joris
CMPX · Compass Therapeutics, Inc.

Tovecimig (CTX-009), a DLL4 x VEGF-A bispecific antibody

Cleared (data check passed)

Second-line unresectable advanced, metastatic or recurrent biliary tract cancer

BPIQ drug id 17398 · tovecimig-biliary-tract-cancer

Analysis as of 2026-10-17 Framework v6.2.0 NCT05506943
Contaminated Another program's result is due close to this one, so a price move cannot be tied to this event alone.
  • CTX-8371 2026-12-31 · BPIQ id 17481 · not yet analysed (BPIQ data row): overlaps, a date the company has given
  • CTX-10726 (PD-1 x VEGF-A bispecific antibody) 2026-12-31 · BPIQ id 18722 · not yet analysed (BPIQ data row): overlaps, a date the company has given

Readout window opens 2026-10-23. Re-checks done: T-5, T-1 (T-5, T-2 and T-1 mean 5, 2 and 1 months before the window opens).

Readout window

When the result of this big event is expected, and what each source says

Our expected window (earliest to latest date) Most likely date Checked against the original source Not checked Estimate from a web search
Q3 2026Q4 2026Q1 2027Q2 2027Q3 2027Q4 2027Q1 2028 Window Our expected window: 2026-10-23 to 2026-10-27. Most likely date: 2026-10-24. Basis: The company named the day and time of the talk (Saturday 2026-10-24, 10:15-11:45 central European time) in its 2026-07-17 release, and the BPIQ note and the 2026-09-22 8-K reaffirm it. Earliest and latest are the congress's opening and closing days from data/congresses.json. The first US trading session that can react to a Saturday talk is Monday 2026-10-26. Correction to the 5.10.3 basis: its claim that ESMO releases abstracts on the opening day was never verified; ESMO's 2026 embargo page returned 404. An OncoDaily preview of 2026-10-10 listed abstract 459O without any duration-of-response figure, which fits an embargo until the talk. An earlier online abstract would bring the readout forward; that risk is UNVERIFIED and is why the run-up exits at T-2. Most likely date: 2026-10-24BPIQBPIQ: 2026-10-24 (“October 24, 2026”). Evidence tag: VERIFIED — BPIQ fetch_company_drugs 2026-10-17. catalyst_date_text names a day, so catalyst_date is not a period-end placeholder. catalyst_source is now the 2026-09-22 regulatory-update release.CT.govCT.gov: 2027-08-31. Evidence tag: VERIFIED — CT.gov get_trial_details NCT05506943, read 2026-10-17. Moved from 2026-12-01 (read 2026-08-19) with no stated reason. Dates the trial's own completion, not the presentation; recorded and not used for the window.CompanyCompany: 2026-10-24 (“oral Proffered Paper presentation at the ESMO Congress 2026, Saturday October 24, 2026, 10:15-11:45 CET; reaffirmed 2026-09-22 as 'the expectation to present the complete dataset at a medical conference this year'”). Evidence tag: VERIFIED — Compass release 2026-07-17; 8-K exhibit 99.1 2026-09-22. Presenter Nilofer Azad; abstract 459O per the ESMO programme as summarised by OncoDaily on 2026-10-10.CongressCongress: 2026-10-23/2026-10-27. Evidence tag: VERIFIED — data/congresses.json, ESMO Congress 2026, as_of 2026-08-06. Admissible because Compass has said it will present there. Supplies the window's outer edges.ModelledModelled: 2027-10/2027-12. Evidence tag: UNVERIFIED — modelled, default lag. Dates the trial's final readout, a later and different event; not used for the window.

5 of 5 sources we checked gave a date. The company named the day and time of the talk (Saturday 2026-10-24, 10:15-11:45 central European time) in its 2026-07-17 release, and the BPIQ note and the 2026-09-22 8-K reaffirm it. Earliest and latest are the congress's opening and closing days from data/congresses.json. The first US trading session that can react to a Saturday talk is Monday 2026-10-26. Correction to the 5.10.3 basis: its claim that ESMO releases abstracts on the opening day was never verified; ESMO's 2026 embargo page returned 404. An OncoDaily preview of 2026-10-10 listed abstract 459O without any duration-of-response figure, which fits an embargo until the talk. An earlier online abstract would bring the readout forward; that risk is UNVERIFIED and is why the run-up exits at T-2. (BPIQ is a biotech data service. CT.gov is ClinicalTrials.gov, the US trial registry.)

The sources agree.

Table view (the same information as the chart, as text)
SourceDate givenEvidence tagNote
BPIQ2026-10-24VERIFIED — BPIQ fetch_company_drugs 2026-10-17catalyst_date_text names a day, so catalyst_date is not a period-end placeholder. catalyst_source is now the 2026-09-22 regulatory-update release.
CT.gov2027-08-31VERIFIED — CT.gov get_trial_details NCT05506943, read 2026-10-17Moved from 2026-12-01 (read 2026-08-19) with no stated reason. Dates the trial's own completion, not the presentation; recorded and not used for the window.
Company2026-10-24VERIFIED — Compass release 2026-07-17; 8-K exhibit 99.1 2026-09-22Presenter Nilofer Azad; abstract 459O per the ESMO programme as summarised by OncoDaily on 2026-10-10.
Congress2026-10-23/2026-10-27VERIFIED — data/congresses.json, ESMO Congress 2026, as_of 2026-08-06Admissible because Compass has said it will present there. Supplies the window's outer edges.
Modelled2027-10/2027-12UNVERIFIED — modelled, default lagDates the trial's final readout, a later and different event; not used for the window.

Verdict

A written summary that is not scored. The scored predictions below are the official prediction.

What I would do (buy, watch, avoid or no view) watch

Why

The shares trade at cash, so a weak duration has limited downside and a strong one real upside, and at today's price the two weigh almost exactly: expected value -0.5% against spot, our band inside a very wide break-even range. The FDA has already said what matters most: patients did not live longer, and the crossover check made it worse. A duration figure does not answer that.

What would change this

A median duration of response of 9 months or more with a supportive discussant would argue for the positive high edge; an FDA statement accepting a filing without a survival trial would change the asset verdict; a duration below the arm's 4.7-month median PFS would argue for the miss low edge.

What to watch

  • 2026-10-19 to 2026-10-22: any online posting of abstract 459O before the talk
  • 2026-10-24, 10:15-11:45 CET: the talk by Nilofer Azad and the discussant's verdict
  • 2026-10-26: first US trading session after the talk
  • About 2026-11-05: Q3 results - runway restatement after the FDA feedback, any survival-trial plan and funding
  • By 2026-11-14: Q3 13Fs for Tang, OrbiMed and Vivo
  • From about 2026-11-19: earliest possible Nasdaq minimum-bid notice
  • Q4 2026: CTX-8371 and CTX-10726 Phase 1 data (the attribution conflicts)
  • Q1 2027: BLA target reported by secondary sources only

Prediction

The outcome and the stock direction are scored separately

Outcome (the trial or regulator decision) modelled
Uncertain
45% [32–58]

Down from 55% (band 42-68) on 2026-08-19. Conditions: ORR >= 15.0% (18.0% disclosed, about 0.97), no new grade 3+ category above 44% (about 0.97), median DoR >= 5.0 months reported (about 0.5 given it is reported, about 0.95 that it is reported). Product about 0.45. Direction 'uncertain' because the band spans 50. No third party has published a probability on this event.

Stock direction (share price) price when locked $0.86 · 2026-10-16
$0.62–$0.84 miss positive $0.87–$1.15
Scenario Low High Anchors (evidence behind the range) How it was worked out
Positive $0.87 $1.15
  • UNVERIFIED Cash floor at the expected event date — 0.86
  • VERIFIED This ticker's own positive data moves over the protocol event window, median (2025-04-01 +1.0%, 2024-11-08 +0.8%, 2024-08-12 +11.0%) — +1.0% median, giving $0.87; maximum +11.0%, giving $0.95, on the high edge only
Low edge: duration just reaches 5.0 months - clears the bar but falls short of the 6.9-month Korean figure and leaves the FDA's survival concern untouched ('missed expectations' bucket), so the shares stay near cash. High edge: duration clearly beats 6.9 months with a supportive discussant ('significantly exceeded' bucket), set above the ticker's own +11.0% maximum for three named facts - 25.6% of the float short, spot at the cash floor, and a long duration being the company's best answer to the FDA - calibrated on oncology +9% x 6.2 for a company under $1 billion [WEB ESTIMATE — IQVIA 2024]. Far below the commercial ceiling allowance of $6.77.
Miss $0.62 $0.84
  • UNVERIFIED Cash floor at the expected event date — 0.86
  • VERIFIED This ticker's own negative tovecimig moves over the protocol event window, median (2026-04-27 -64.6%, 2026-09-22 -35.2%) — -49.9% median, giving $0.43
A duration below 5.0 months, or none reported, removes the company's best argument for filing. The range sits below the floor because spending continues on three programs at about $15M a quarter, a Nasdaq minimum-bid notice can arrive from about 2026-11-19, and an unused $100M ATM makes dilution easy to start; and far above the -49.9% own-history anchor because those moves came from prices well above cash. The Q4 2026 CTX-8371 and CTX-10726 readouts could hold it near the floor. Shallower than the positive row is high, against the class default, because spot sits at the cash floor (07-benchmarks.md names this as legitimate).
$0.86-0.5% modelled estimate

-4.7% to +3.8% compared with the current price, across our 32–58% range of probabilities. We cannot tell yet whether it is above or below the current price.

probability_pct (45%) applied to the midpoint of each scenario range: 0.45 x $1.01 + 0.55 x $0.73

No edge how sure we are of the direction None

callFromBands returns no-edge: our band [32, 58] overlaps the break-even range [6.5, 96.0]. The expected value is -0.5% against spot, consistent with no-edge. Materiality is dominant per company.md C.2; a dominant program presenting a known dataset after the regulator has spoken is a smaller event than its materiality suggests. Attribution is CONTAMINATED: CTX-8371 (17481) and CTX-10726 (18722) carry Q4 2026 windows covering this program's disclosed day, so a move in this window is not attributable to this program alone. Spot is a settled close (2026-10-16), not an intraday snapshot. No reverse split is authorised, so none can fall inside the window; spot_scale should be 1 at settlement unless EDGAR shows otherwise.

2026-10-19 → 2026-10-30 · Opens with the first session of the congress week, covers the Saturday 2026-10-24 talk and the first trading session after it (2026-10-26), and closes before the expected early-November third-quarter results.

break-even chance of success from our own price ranges: 46.4% (range 6.5–96%) · our band overlaps it · highest sensible price from sales potential: 4.8

published base rate for this phase: 47.7% (BIO2021 Figure 2, p. 7 (Oncology, Phase III to NDA/BLA), via lib/success-rates.mjs baseRate (area fallback; no bispecific row); equals asset_evaluation.pos.base_rate)

Rationale

Program refresh at the T-1 cadence gate (06-prompts.md § Run due), with a full company-tier and program-tier re-sweep on 2026-10-17; not a re-lock. Rewritten to the 6.2.0 layout. Since the 2026-08-19 lock: the FDA recommended a survival trial before a BLA (2026-09-22), the shares fell from $2.48 to $0.86 and now sit at the cash floor, the final ORR is 18.0%, and the 5.10.3 upside anchor (+49.73% on 2025-04-01) failed to reproduce from the price cache.

Locked (cannot be changed) Awaiting the result Prediction dated 2026-10-17

Prediction history

2 predictions locked on this program, with the changes between one and the next

Every prediction ever locked on this program, oldest first. A refresh does not edit an old prediction. It adds a new one that replaces ("supersedes") the old one. Each earlier prediction is kept exactly as written, because it shows what we believed before the outcome was known. It is not an out-of-date draft that needs fixing. A column marked "moved" shows a value that changed since the previous prediction.

We re-check each program at set points: 5, 2 and 1 months before the expected result. Each column is labelled with the checkpoint it fell in. The checkpoint is judged against the readout window that prediction itself assumed. “Ad-hoc” means there was no readout window yet, so no checkpoint could be set. This applies to every program in this corpus today, so expect to see it below. It means the refreshes that add a window are still waiting to be done. It does not mean this table has a gap.

Field
2026-08-19 CMPX-17398-2026-08-19 T-5 Superseded
2026-10-17 CMPX-17398-2026-10-17 T-1 Live
Probability of a good outcome 55% 45% moved
Probability range 42–68% 32–58% moved
Stock-call window (dates the stock call covers) 2026-10-19 → 2026-11-07 2026-10-19 → 2026-10-30 moved
Share price when locked $2.48 2026-08-19 $0.86 2026-10-16 moved
Expected share price $2.51 (+1.0%) $0.86 (-0.5%) moved
Share price ranges (positive result and miss) positive result $2.70–$3.30 · miss $1.65–$2.15 positive result $0.87–$1.15 · miss $0.62–$0.84 moved
Readout window 2026-10-23 → 2026-10-27 most likely 2026-10-24 2026-10-23 → 2026-10-27 most likely 2026-10-24
Run-up score 49 · move 5–35% score 56 · move -5–15% moved

Catalyst

The binary event being scored

Name
Oral Proffered Paper presentation (abstract 459O) of the complete COMPANION-002 Phase 2/3 dataset, including duration of response, at the ESMO Congress 2026 in Madrid
Catalyst Date
2026-10-24
Catalyst Date Text
October 24, 2026
Catalyst Date Is Exact
Yes
Nct
NCT05506943
Date Slips
5

Clinical

Trial history and readouts

Mechanism
One antibody with two binding functions: its main body binds VEGF-A, the protein a tumour releases to make blood vessels sprout toward it, and two single-chain variable fragments attached to the ends bind DLL4, a protein on the surface of the vessel cells inside the tumour. Blocking VEGF-A alone drives tumours to switch on DLL4-Notch signalling as an escape route, so blocking both closes that route; when the disordered vessels regress and normalise, paclitaxel reaches more tumour cells.
Pivotal Trial
NCT05506943 (COMPANION-002): Phase 2/3, open-label, randomised 2:1, n=168 at 34 US sites, crossover allowed. Primary endpoint overall response rate by RECIST v1.1. Registry primary completion 2027-08-31 (2026-12-01 on 2026-08-19).
Results To Date
Orr
18.0% (20/111) vs 5.3% (3/57), p=0.0228 (final, 8-K 2026-09-22); earlier 17.1% (19/111), p=0.031
Pfs
median 4.7 vs 2.6 months, hazard ratio 0.44, p<0.0001
Os
median 8.9 vs 9.4 months, hazard ratio 1.05, p=0.78; RPSFT-adjusted 1.13; 54% crossover (31/57)
Dor
not reported; the 2026-10-24 event's new content
Safety
hypertension 69% all grades, 44% grade 3+; neutropenia 36% grade 3+; fatigue 67%
Powering Gap
Designed to detect a 23-point absolute difference in response rate with 90% power (PMID 38861293); the final difference is 12.7 points (18.0 - 5.3). The 5.10.3 figure was 11.8 points on the earlier 17.1%.
Crossover Argument Fails Its Own Check
The company attributes the flat overall survival to 54% crossover. The standard test of that claim is a rank-preserving structural failure time analysis, which Compass reported: hazard ratio 1.13 against the unadjusted 1.05. Removing the crossover made the result worse. The supportive subset comparison (crossover 12.8 months vs non-crossover 6.1 months, HR 0.54, p=0.04) compares patients who lived long enough to cross over against those who did not, which is survivorship bias rather than evidence.
Fda Feedback 2026 09 22
The FDA recommended that Compass conduct a trial demonstrating a survival benefit before proceeding with a BLA; Compass does not believe a new trial is warranted and will further engage the FDA while preparing a BLA (8-K exhibit 99.1). A Q1 2027 BLA target is reported by secondary sources only.
Designations
Fast Track (2024-04-25); Orphan Drug for BTC (announced by 2026-04-08; the 5.10.3 'around 2026-05-05' was the BPIQ note date); no Breakthrough Therapy.

Competitive landscape

Comparators and benchmarks

Mechanism Differentiation
First DLL4 x VEGF-A bispecific in a pivotal trial; no anti-angiogenic drug approved in BTC; no anti-DLL4 drug approved anywhere.
Timeline Advantage
No other randomised all-comer second-line drug is near filing, but a survival trial would push tovecimig back about three years (Leerink via BioSpace).
Proof Of Concept
A randomised trial met response rate and progression-free survival, but survival was flat and the FDA asked for a survival trial.
Load Bearing Fact
Whether the FDA accepts response rate and progression-free survival without a survival benefit. Tested on 2026-09-22: the FDA recommended a survival trial first.

Treatment algorithm

Standard of care and where the asset fits

First Line
Gemcitabine and cisplatin, usually with durvalumab (Imfinzi label) or pembrolizumab.
Second Line Targeted
IDH1 (about 13% of intrahepatic cholangiocarcinoma) to ivosidenib; FGFR2 fusions to pemigatinib or futibatinib (accelerated approvals); HER2 IHC 3+ to zanidatamab (accelerated approval). All three accelerated approvals rest on response rate and duration of response from single-arm trials (DailyMed labels).
Second Line Unselected
About 70-80% of second-line patients have no targetable alteration and receive FOLFOX (ABC-06, survival 6.2 vs 5.3 months) or nal-IRI plus 5-FU (NIFTY, progression-free survival 4.2 vs 1.7 months).
Where This Fits
In the unselected group, added to paclitaxel; COMPANION-002 excludes patients eligible for a targeted drug.

Intellectual property

Exclusivity and royalty burden

Orphan Exclusivity Years
7
Orphan Exclusivity Note
Seven years of US exclusivity for BTC from approval.
Patent
Two licensed patent families (DLL4/VEGF antibody program) with 2 issued US and 27 foreign patents, expected to start to expire in 2033, subject to possible extension (Form 10-K for 2025).
Patent Note
Named correction: the 5.10.3 version found no patent estate. Whether the 2033 patents claim the molecule itself is not stated. A new biologic would also receive 12 years of US reference-product exclusivity from licensure (statutory default, UNVERIFIED for this product).
Licence
Exclusive global licence excluding South Korea from ABL Bio (November 2018); Handok holds South Korea; China out-licensed to Elpiscience (15% sublicense share to ABL Bio). ABL Bio received $5M upfront and a $6M milestone and is eligible for up to $96M development/regulatory and $303M commercial milestones plus tiered single-digit royalties in oncology (Form 10-Q filed 2026-08-06; Form 10-K for 2025). Verified from filings; the 5.10.3 version carried these as web estimates.

Valuation

Peak-sales scenarios and capital needs

Peak Sales Scenarios
Low
Usd Per Year
50,000,000
Assumptions
2,000 patients x $100,000 x 25% penetration
Base
Usd Per Year
146,000,000
Assumptions
2,600 patients x $125,000 x 45% penetration
High
Usd Per Year
312,000,000
Assumptions
3,200 patients x $150,000 x 65% penetration
Addressable Us Patients Per Year
2,080 to 3,190
Funnel
US biliary tract cancer diagnoses 12,130 (SEER-based, 2022) to 18,600 (American Cancer Society, 2024) x ~68% advanced/unresectable x ~70% receiving first-line gem/platinum x ~45% progressing and fit for second line x ~80% without an actionable alteration.
Weakest Input
The price. No tovecimig price exists and no analogous all-comers add-on has ever launched in this disease, so the $20,000-$30,000/month assumption has no verified anchor. Penetration is the second weakest. Patient count is the strongest, having two independent published sources at its first step and a cited literature range at its last.
Launch Assumption
Launch near 2029-2030 (was 2028 in 5.10.3), peak around 2034.
Rejected Inputs
William Blair's $1.46 billion worldwide peak sales on about 15,000 eligible US patients (pre-readout) rejected: 15,000 is close to all US BTC diagnoses before any funnel step, and it predates the April 2026 survival result and the FDA position. LifeSci Capital's ~$1.4 billion US opportunity set aside for the same reasons.
Enpv Range Usd
-60000000 to 250000000
Enpv Assumptions
Value at approval (peak x multiple 2/3.5/5 x economic share 0.9) times approval odds of 5/12/22%, discounted about four years at 12%, less a risk-weighted survival trial of about $70M. No point estimate (rule 10). The 5.10.3 range was $150-700M before the FDA position.
Economic Share
90.0%
Economic Share Basis
US economics retained; tiered single-digit royalties and up to $303M of commercial milestones owed to ABL Bio leave roughly 90% (UNVERIFIED judgement on verified terms).

Market timing

Positioning into this event

Spot
$0.86
Spot As Of
2026-10-16
Spot Kind
settled close
Months To Catalyst
0.2
Months To Catalyst Basis
Six days from the 2026-10-17 lock to readout.window.earliest 2026-10-23; precision DAY.
Expected Move Bracket Pct
10 to 35
Expected Move Basis
Options chain unreadable (no strike at spot; $1.00 straddle 29% to 76% of spot). Bracket calibrated on Phase 3 +11/-22%, oncology +9/-9% and the under-$1B size factor of 6.2x [WEB ESTIMATE — IQVIA 2024], and this ticker's own event-window moves (+0.8% to +11.0% on presentations, -35% and -65% on hypothesis-level news).
Nearest Comparable
company.md C.7 2024-11-08 CTX-471 SITC presentation, +0.8% over the event window: same kind, much smaller stakes.
Materiality
dominant (company.md C.2)
Cash Floor
$0.9M

Chemistry (ChEMBL)

Compound identity and properties

Molecule Chembl Id
CHEMBL6068308
Pref Name
TOVECIMIG
Molecule Type
Antibody
Max Phase
2
Structure Type
SEQ
Usan Stem
-mig (monoclonal antibody, multiple immunoglobulin)

Readout

Window
Earliest
2026-10-23
Likeliest
2026-10-24
Latest
2026-10-27
Precision
DAY
Confidence
HIGH
Basis
The company named the day and time of the talk (Saturday 2026-10-24, 10:15-11:45 central European time) in its 2026-07-17 release, and the BPIQ note and the 2026-09-22 8-K reaffirm it. Earliest and latest are the congress's opening and closing days from data/congresses.json. The first US trading session that can react to a Saturday talk is Monday 2026-10-26. Correction to the 5.10.3 basis: its claim that ESMO releases abstracts on the opening day was never verified; ESMO's 2026 embargo page returned 404. An OncoDaily preview of 2026-10-10 listed abstract 459O without any duration-of-response figure, which fits an embargo until the talk. An earlier online abstract would bring the readout forward; that risk is UNVERIFIED and is why the run-up exits at T-2.
Sources
Source 1
Kind
bpiq
Value
2026-10-24
Text
October 24, 2026
Tag
VERIFIED — BPIQ fetch_company_drugs 2026-10-17
As Of
2026-10-17
Source 2
Kind
ctgov
Value
2027-08-31
Tag
VERIFIED — CT.gov get_trial_details NCT05506943, read 2026-10-17
As Of
2026-10-17
Field
primary_completion_date
Nct
NCT05506943
Source 3
Kind
company
Value
2026-10-24
Text
oral Proffered Paper presentation at the ESMO Congress 2026, Saturday October 24, 2026, 10:15-11:45 CET; reaffirmed 2026-09-22 as 'the expectation to present the complete dataset at a medical conference this year'
Tag
VERIFIED — Compass release 2026-07-17; 8-K exhibit 99.1 2026-09-22
As Of
2026-09-22
Url
https://www.sec.gov/Archives/edgar/data/1738021/000117184326006162/exh_991.htm
Source 4
Kind
congress
Value
2026-10-23/2026-10-27
Tag
VERIFIED — data/congresses.json, ESMO Congress 2026, as_of 2026-08-06
As Of
2026-08-06
Name
ESMO Congress 2026, IFEMA MADRID, Madrid, Spain
Url
https://www.esmo.org/meeting-calendar/esmo-congress-2026
Source 5
Kind
modelled
Value
2027-10/2027-12
Tag
UNVERIFIED — modelled, default lag
As Of
2026-10-17
Method
CT.gov primary_completion_date 2027-08-31 plus the default two to four months to database lock and analysis (01-rules.md rule 34); data/benchmarks/readout-lag.json holds no matching entry.
Disagreement
CONSISTENT
Slips
Count
5
Sequence
Sequence 1
As Of
2025-02-27
Text
Report topline data by the end of Q1 2025
Sequence 2
As Of
2025-08-11
Text
Secondary endpoints including OS data expected in Q1 2026
Sequence 3
As Of
2025-11-05
Text
OS/PFS secondary endpoint data expected late Q1 2026; potential first BLA filing planned H2 2026
Sequence 4
As Of
2026-03-05
Text
PFS/OS secondary endpoint analyses from COMPANION-002 expected in April 2026
Sequence 5
As Of
2026-04-27
Text
COMPANION-002 met key PFS endpoint; complete dataset with DoR to present later 2026; FDA meeting planned
Sequence 6
As Of
2026-07-17
Text
Oral ESMO 2026 presentation of COMPANION-002 Ph2/3 BTC results on October 24, 2026
Sequence 7
As Of
2026-08-06
Text
Final ORR 18%; FDA feedback Q3 2026; ESMO oral data October; BLA planned 2026
Sequence 8
As Of
2026-09-22
Text
FDA advised survival-benefit trial before BLA; company will further engage FDA. ESMO data Oct 24, 2026.

Attribution

Status
CONTAMINATED
Conflicts
Conflict 1
Bpiq Drug Id
17,481
Label
CTX-8371
Date
2026-12-31
Analysed
No
Gap Days
0
Confirmed
Yes
Conflict 2
Bpiq Drug Id
18,722
Label
CTX-10726 (PD-1 x VEGF-A bispecific antibody)
Date
2026-12-31
Analysed
No
Gap Days
0
Confirmed
Yes

Kol

As Of
2026-10-17
Investigators
Investigator 1
Name
Nilofer Azad
Role
First author of the COMPANION-002 design paper; presenter of ESMO abstract 459O on 2026-10-24
Affiliation
Professor of Oncology and Associate Director of Clinical Research, Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins - a COMPANION-002 site
Nct
NCT05506943
Conflicts
Conflict 1
Party
sponsor
Kind
trial investigator and designated presenter of the sponsor's data
Disclosed In
Compass Therapeutics press release, 2026-07-17
As Of
2026-07-17
Conflicts Checked
PubMed get_article_metadata PMID 38861293, 2026-10-17 - author list returned, no financial-disclosure section, CT.gov search_investigators (condition biliary tract cancer, name Azad), 2026-10-17 - zero rows, PubMed get_full_text_article PMC11509068, 2026-08-19 - no disclosure section (5.10.3 sweep)
Tag
VERIFIED — Compass release 2026-07-17 and PubMed 2026-10-17
Investigator 2
Name
Milind Javle
Role
Senior (last) author of the COMPANION-002 design paper
Affiliation
Department of Gastrointestinal Medical Oncology, The University of Texas MD Anderson Cancer Center - a COMPANION-002 site, and separately the sponsor of the first-line trial NCT06548412
Nct
NCT05506943
Conflicts
Conflict 1
Party
sponsor
Kind
trial investigator, disclosed by authorship of the sponsor's design paper
Disclosed In
Future Oncology 2024;20(30):2241-2248, PMID 38861293
As Of
2024-06-04
Conflicts Checked
PubMed get_article_metadata PMID 38861293, 2026-10-17 - no financial-disclosure section
Tag
VERIFIED — PubMed 2026-10-17
Investigator 3
Name
Rachna Shroff
Role
COMPANION-002 design-paper author
Affiliation
Division of Hematology and Oncology, University of Arizona Cancer Center - a COMPANION-002 site
Nct
NCT05506943
Conflicts
Conflict 1
Party
sponsor
Kind
trial investigator, disclosed by authorship
Disclosed In
Future Oncology 2024;20(30):2241-2248, PMID 38861293
As Of
2024-06-04
Conflicts Checked
PubMed get_article_metadata PMID 38861293, 2026-10-17 - no financial-disclosure section
Tag
VERIFIED — PubMed 2026-10-17
Investigator 4
Name
Lipika Goyal
Role
COMPANION-002 design-paper author
Affiliation
Division of Hematology and Oncology, Stanford Cancer Center - a COMPANION-002 site
Nct
NCT05506943
Conflicts
Conflict 1
Party
sponsor
Kind
trial investigator, disclosed by authorship
Disclosed In
Future Oncology 2024;20(30):2241-2248, PMID 38861293
As Of
2024-06-04
Conflict 2
Party
competitor
Kind
first author of Taiho's futibatinib FOENIX-CCA2 trial (FGFR2-rearranged intrahepatic cholangiocarcinoma)
Disclosed In
N Engl J Med 2023 byline, PMID 36652354
As Of
2023-01-19
Conflicts Checked
PubMed get_article_metadata PMIDs 38861293 and 36652354, 2026-10-17 - author lists returned, no disclosure section in either payload
Tag
VERIFIED — PubMed 2026-10-17
Investigator 5
Name
Minori Rosales
Role
COMPANION-002 design-paper author; an employee of the sponsor rather than an independent investigator
Affiliation
Compass Therapeutics, 80 Guest Street, Boston, MA 02135
Nct
NCT05506943
Conflicts
Conflict 1
Party
sponsor
Kind
employment
Disclosed In
the author affiliation printed on Future Oncology 2024;20(30):2241-2248, PMID 38861293
As Of
2024-06-04
Conflicts Checked
PubMed get_article_metadata PMID 38861293, 2026-10-17 - no financial-disclosure section
Tag
VERIFIED — PubMed 2026-10-17
Judgement
Endpoint Supported
UNKNOWN
Basis
No independent clinician has commented on this endpoint in any source reached, and no clinical result is peer-reviewed (four PubMed papers, none independent clinical commentary). The FDA's 2026-09-22 recommendation is a regulator's position, recorded in A.4, and analyst comment (Leerink via BioSpace) is not a KOL view under rule 40. The ESMO discussant on 2026-10-24 will be the first independent named view.
Tag
UNVERIFIED — judgement

Asset Evaluation

Mode
A
Search Date
2026-10-17
Modality
bispecific
Therapeutic Area
oncology
Orphan
Us
Yes
Scores
Score 1
Section
A.2
Criterion
Target validation
Score
5
Tag
VERIFIED — Open Targets 2026-10-17: no DLL4 association with cholangiocarcinoma or BTC; VEGFA 0.152
Score 2
Section
A.2
Criterion
Mechanism and target engagement
Score
5
Tag
VERIFIED — PubMed PMID 39375952 (PK model); hypertension as PD sign
Score 3
Section
A.2
Criterion
Translational biomarkers
Score
2
Tag
VERIFIED — CT.gov NCT05506943, no molecular stratification
Score 4
Section
A.2
Criterion
Clinical effect against standard of care
Score
5
Tag
VERIFIED — 8-K exhibit 99.1 2026-09-22 and Compass 2026-04-27
Score 5
Section
A.2
Criterion
Consistency and replication
Score
4
Tag
VERIFIED — PMID 38861293 (Korean Phase 2 37.5%) against 18.0% randomised
Score 6
Section
A.2
Criterion
Therapeutic window and safety
Score
3
Tag
VERIFIED — Compass 2026-04-27, 44% grade 3+ hypertension
Score 7
Section
A.2
Criterion
Data transparency and expert view
Score
2
Tag
UNVERIFIED — expert view score
Score 8
Section
A.5
Criterion
Recruitment feasibility
Score
8
Tag
VERIFIED — CT.gov NCT05506943, 168 enrolled at 34 sites
Score 9
Section
A.5
Criterion
Endpoint acceptance
Score
3
Tag
VERIFIED — 8-K exhibit 99.1, FDA recommended a survival trial
Score 10
Section
A.5
Criterion
Design robustness
Score
5
Tag
VERIFIED — CT.gov and PMID 38861293: randomised, open-label, crossover
Score 11
Section
A.5
Criterion
Execution track record
Score
4
Tag
VERIFIED — BPIQ note field and company.md C.1a
Score 12
Section
M.6
Criterion
Process robustness and CQAs
Score
5
Tag
UNVERIFIED — no public process data; validation batches per 10-Q
Score 13
Section
M.6
Criterion
Regulatory CMC record
Score
6
Tag
VERIFIED — openFDA CRL API NOT_FOUND; no application inspected yet
Score 14
Section
M.6
Criterion
Capacity and supply
Score
5
Tag
VERIFIED — 10-K: unnamed third-party CDMOs
Score 15
Section
M.6
Criterion
Cost of goods
Score
5
Tag
UNVERIFIED — judgement, 10 mg/kg biweekly dose
Score 16
Section
B.1
Criterion
Mechanism differentiation
Score
7
Tag
VERIFIED — ChEMBL max_phase 2; no DLL4 x VEGF-A agent approved
Score 17
Section
B.1
Criterion
Timing against competitors
Score
5
Tag
WEB ESTIMATE — Leerink via BioSpace 2026-09-22, about three years' delay if a new trial
Score 18
Section
B.1
Criterion
Clinical data against competitors
Score
4
Tag
UNVERIFIED — judgement on indirect B.1a comparison
Score 19
Section
B.1
Criterion
IP and exclusivity
Score
6
Tag
VERIFIED — Form 10-K for 2025, patents expiring from 2033
Score 20
Section
B.3
Criterion
Comparator alignment
Score
4
Tag
VERIFIED — CT.gov, comparator paclitaxel not FOLFOX
Score 21
Section
B.3
Criterion
Endpoints and certainty
Score
3
Tag
VERIFIED — 8-K exhibit 99.1; survival flat
Score 22
Section
B.3
Criterion
Supplementary evidence plan
Score
2
Tag
VERIFIED — QLQ-C30 planned in PMID 38861293, never reported
Score 23
Section
B.3
Criterion
Value for money at the target price
Score
3
Tag
UNVERIFIED — judgement
Score 24
Section
B.3
Criterion
Payer precedent
Score
4
Tag
WEB ESTIMATE — NICE TA948 summary, 2026-10-17
Stage Gate
Stage Gate 1
Id
T1
Gates
phase2, phase3
Rating
amber
Confidence
M
Stage Gate 2
Id
T2
Gates
phase2, phase3
Rating
amber
Confidence
M
Stage Gate 3
Id
T3
Gates
phase2, phase3, filing
Rating
green
Confidence
M
Stage Gate 4
Id
D1
Gates
phase2, phase3, filing
Rating
red
Confidence
M
Stage Gate 5
Id
D2
Gates
phase2, phase3, filing
Rating
amber
Confidence
M
Stage Gate 6
Id
D3
Gates
phase2, phase3, filing
Rating
amber
Confidence
L
Stage Gate 7
Id
D4
Gates
phase3, filing
Rating
amber
Confidence
M
Stage Gate 8
Id
P1
Gates
phase2, phase3, filing
Rating
red
Confidence
H
Stage Gate 9
Id
P2
Gates
phase2, phase3, filing
Rating
amber
Confidence
M
Stage Gate 10
Id
P3
Gates
phase2, phase3
Rating
red
Confidence
L
Stage Gate 11
Id
P4
Gates
phase3, filing
Rating
red
Confidence
M
Stage Gate 12
Id
R1
Gates
phase3, filing
Rating
amber
Confidence
M
Stage Gate 13
Id
R2
Gates
phase3, filing
Rating
red
Confidence
M
Domains
Scientific Rationale
Score
3.7
Weight
15
Method
mean
Basis
Mean of the seven A.2 criteria (5, 5, 2, 5, 4, 3, 2), lib/scorecard.mjs.
Clinical Data Design
Score
4
Weight
20
Method
direct
Basis
A.3: P1, P3, P4, R2 and D1 red (FDA recommended a survival trial; survival hazard ratio 1.05; 44% grade 3+ hypertension; no exposure-response model); T3 green; the rest amber. The randomised design and met primary endpoint keep it above the red band.
Regulatory Pathway
Score
3
Weight
10
Method
direct
Basis
A.4: the FDA recommended a trial demonstrating a survival benefit before any BLA (2026-09-22); Fast Track and Orphan Drug held; same-disease accelerated approvals rest on single-arm, marker-selected data; FDA's August 2025 draft guidance treats survival as a harm check.
Operations Cmc
Score
5.1
Weight
10
Method
mean
Basis
Mean of the A.5 parameter mean (5.0) and the M.6 CMC score (5.25), lib/scorecard.mjs.
Competitive Position
Score
5.5
Weight
15
Method
mean
Basis
Mean of the four B.1 metrics (7, 5, 4, 6), lib/scorecard.mjs.
Market Access
Score
3.2
Weight
15
Method
mean
Basis
B.3 Market Access Score, mean of five criteria (4, 3, 2, 3, 4), lib/scorecard.mjs.
Risk Profile
Score
3
Weight
15
Method
direct
Basis
C: science-and-clinical and regulatory rows are high likelihood and high impact; execution and financing high impact; CMC medium.
Weighted Total
3.9
Cmc Score
5.3
Market Access Score
3.2
Pos
Base Rate
Value Pct
47.7%
Phase
phase3
Area
oncology
Modality
bispecific
Transition
next_phase
Adjustments
Adjustment 1
Factor
Biomarker-selected population
Direction
down
Evidence
None; all comers (CT.gov NCT05506943).
Tag
VERIFIED — CT.gov
Adjustment 2
Factor
Human genetic support for the target
Direction
down
Evidence
No Open Targets association between DLL4 and cholangiocarcinoma or BTC; VEGFA 0.152, mostly literature.
Tag
VERIFIED — Open Targets 2026-10-17
Adjustment 3
Factor
Rare disease or orphan designation
Direction
up
Evidence
Orphan Drug and Fast Track designations; clear unmet need in the all-comer second line.
Tag
VERIFIED — designation coverage
Adjustment 4
Factor
Randomised Phase 2 with a meaningful effect
Direction
up
Evidence
Randomised gains in response rate (18.0% vs 5.3%) and progression-free survival (hazard ratio 0.44).
Tag
VERIFIED — 8-K exhibit 99.1 2026-09-22
Adjustment 5
Factor
First-in-class, novel endpoint, or prior failures in the class
Direction
down
Evidence
No anti-DLL4 drug approved; survival hazard ratio 1.05, 1.13 after crossover adjustment.
Tag
VERIFIED — Compass 2026-04-27
Adjustment 6
Factor
Other: FDA recommended a survival trial before a BLA
Direction
down
Evidence
The regulator's own position on these data, 2026-09-22.
Tag
VERIFIED — 8-K exhibit 99.1
Next Phase
Low
30
Base
45
High
60
Approval
Low
5
Base
12
High
22
Asset Verdict
Verdict
pass
Reason
The regulator has said the current data are not enough to file, and the survival result gives no sign a new trial would succeed. Monitor only if the 2026-10-24 duration figure is strong and the FDA later accepts a filing, or a partner funds the survival trial. Confidence Medium. Does not set any scored call (rule 45).

Risk flags

  • The FDA recommended a trial demonstrating a survival benefit before any BLA (2026-09-22).
  • Overall survival is flat (hazard ratio 1.05) and the crossover adjustment makes it worse (1.13).
  • Grade 3 or higher hypertension in 44% of patients, against no survival gain.
  • The final response-rate difference (12.7 points) is about half the 23 points the trial was powered for.
  • No peer-reviewed clinical result exists.
  • No predictive biomarker.
  • Attribution is CONTAMINATED: CTX-8371 and CTX-10726 Q4 2026 windows cover the disclosed day.
  • Shares at the cash floor and below $1; a Nasdaq minimum-bid notice is possible from about 2026-11-19.
  • Manufacturing at unnamed contract sites; validation spending ahead of a delayed filing.

Full analysis

The full written analysis. Each claim is tagged with the kind of evidence behind it.

CMPX / tovecimig-biliary-tract-cancer — Tovecimig (CTX-009) for previously treated biliary tract cancer

Program analysis · bpiq_drug_id 17398 · Mode A (prospective) · prepared 2026-10 · search date 2026-10-17 · USD · framework v6.2.0 · Coverage: CLEARED

Asset. Tovecimig · CTX-009, ABL001, NOV1501, TR009, ES104 · no brand name · sponsor Compass Therapeutics, under an exclusive global licence (excluding South Korea) from ABL Bio; Handok holds South Korea; Compass out-licensed China to Elpiscience [1][2] Biology and modality. DLL4 and VEGF-A · one antibody that blocks two signals tumours use to grow blood vessels · modality bispecific Indication. Biliary tract cancer (bile-duct and gallbladder cancer) · second line, after gemcitabine and platinum chemotherapy · no biomarker selection · therapeutic area oncology · orphan status US yes, EU unknown Status. Phase 2/3 · COMPANION-002 (NCT05506943) [3] · Fast Track (2024-04-25) and Orphan Drug (announced by 2026-04-08) · no approval anywhere Launch markets. United States only. The filings describe no EU, UK or Japan launch plan [2]

Company context, financials, ownership, options and the full pipeline: ../company.md.

How to read this. Written for a curious reader with little or no background in science, medicine, the drug industry or finance. Terms specific to this program are explained in the Glossary below. General terms are explained once in the glossary. All money figures are in US dollars (USD). Every fact carries one of three tags. [VERIFIED — source] means a primary source confirmed it, and the source is named. [UNVERIFIED] means it could not be confirmed. [WEB ESTIMATE — source, date] means it is a third-party figure or a labelled model.

Confidence legend. Every conclusion carries a confidence beside it (01-rules.md rule 1). A conclusion is a judgement drawn from several facts: a score, a rating, an answer to a Go / No-Go question, or a verdict. High means the evidence under it is primary, consistent and recent. Medium means the evidence is partial, indirect or mixed. Low means the judgement rests mostly on inference or on sources that could not be verified.

Sources. Every source is cited as a number in square brackets, [n]. The numbers run through the document in the order of first use. The Sources list at the end gives one link per number (01-rules.md rule 20). “No public record as of 2026-10-17” means a search on that date found nothing.

This is a refresh. It replaces the 2026-08-19 document, which was written to framework 5.10.3. It re-ran the company tier and the program tier of the sweep in full. It keeps the earlier version’s corrections, its audit trails for rejected inputs and its analyses of past events, and it marks where this sweep corrected them. The largest change since August is the FDA feedback of 2026-09-22: the agency recommended a trial showing that patients live longer before Compass files for approval [4].

Glossary

TermPlain-language meaning
Tovecimig (also CTX-009, ABL001, NOV1501, TR009, ES104)The drug. One antibody built with two kinds of binding arm, one for VEGF-A and one for DLL4. The code ABL001 is also used by Novartis for an unrelated leukaemia drug (asciminib), which is why searches on that code return the wrong papers.
Biliary tract cancer (BTC)Cancer of the tubes that carry bile from the liver to the gut, and of the gallbladder. Four subtypes: intrahepatic cholangiocarcinoma (inside the liver), extrahepatic cholangiocarcinoma (outside it), gallbladder cancer, and ampullary cancer (where the bile duct meets the small intestine).
CholangiocarcinomaCancer of the bile duct itself, the first two subtypes above. Often used loosely for all biliary tract cancer; it is not the same thing.
DLL4 (delta-like ligand 4)A protein on blood-vessel cells inside a tumour. It switches on the Notch signal, which controls how new vessels branch and mature.
VEGF-A (vascular endothelial growth factor A)The main protein a tumour releases to make new blood vessels grow toward it.
Bispecific antibodyAn antibody built to grip two different targets at once. Tovecimig grips VEGF-A with its main body and DLL4 with two small fragments fixed to its ends.
COMPANION-002 (NCT05506943)The pivotal trial: tovecimig plus paclitaxel against paclitaxel alone, in patients whose cancer grew after one earlier chemotherapy.
COMPANION-003 (NCT05513742)A separate, finished Phase 2 of tovecimig in colorectal cancer. It did not advance. Not this program.
PaclitaxelAn old, cheap chemotherapy that stops cells dividing. Both arms of COMPANION-002 receive it.
FOLFOXA chemotherapy combination (folinic acid, fluorouracil and oxaliplatin). The ABC-06 trial made it the usual second-line treatment.
ABC-06The randomised UK trial in which FOLFOX beat symptom care alone in second-line biliary tract cancer, by 0.9 months of median survival [5].
CrossoverA trial rule that lets a patient whose cancer grows on the control arm switch to the new drug. It is humane, and it blurs the survival comparison, because the control group stops being a pure control. 54% of COMPANION-002 control patients crossed over.
RPSFT (rank-preserving structural failure time)A standard statistical method that estimates what the control group’s survival would have been if nobody had crossed over.
ESMO CongressThe European Society for Medical Oncology’s yearly meeting. The 2026 meeting runs 23–27 October in Madrid.
Proffered PaperESMO’s highest-visibility abstract category: a podium talk followed by a named expert discussant who critiques the data in public.
Abstract 459OESMO’s number for the COMPANION-002 talk [6].
ABL BioThe South Korean company that invented the molecule and licensed it to Compass in 2018. It is owed royalties and milestone payments.

Executive scorecard

Asset snapshot

Asset snapshot
Asset and sponsorTovecimig (CTX-009), Compass Therapeutics; licensed from ABL Bio (global excluding South Korea), China sub-licensed to Elpiscience, South Korea with Handok [1][2]
Modality and mechanismbispecific; an IgG antibody against VEGF-A with two single-chain fragments against DLL4 fused to its ends; blocks two blood-vessel growth signals at once [7]
Targeted indications (Mode A)Unresectable, advanced or metastatic biliary tract cancer after one gemcitabine-platinum regimen, all comers, no biomarker [3]
Stage and modePhase 2/3 (COMPANION-002 completed its primary and secondary analyses) · Mode A
Next catalysts (Mode A)ESMO oral presentation of the full COMPANION-002 dataset, including duration of response, 2026-10-24 (window 2026-10-23 to 2026-10-27, see Readout) [8]. Then further FDA engagement; a BLA targeted for the first quarter of 2027 per secondary reports only [9]
Asset verdictPass — see below

Domain scores

The weights are v2.2’s defaults [UNVERIFIED — default weight]. Every figure below is computed by lib/scorecard.mjs from the criterion scores in A.2, A.5, M.6, B.1 and B.3 and from the three direct scores.

DomainSectionsWeightScore (1–10)RatingKey driver (one line)
Scientific rationaleA.1–A.215%3.7redDLL4 has no genetic link to this disease and no anti-DLL4 drug has been approved; 44% severe high blood pressure
Clinical data and designA.320%4.0amberA randomised trial met response rate and progression-free survival, but survival was flat (hazard ratio 1.05)
Regulatory pathwayA.410%3.0redThe FDA recommended a survival trial before any marketing application (2026-09-22)
Operations and CMCA.5, M10%5.1amberEnrolment done; manufacturing at unnamed contract sites; guidance slipped repeatedly
Competitive positionB.115%5.5amberFirst of its kind for the 70–80% without a targetable mutation, but no survival edge over FOLFOX
Market accessB.2–B.315%3.2redCompared with paclitaxel, not FOLFOX; no survival benefit; no quality-of-life data
Risk profile (10 = low risk)C15%3.0redRegulatory and efficacy risks are both likely and severe
Weighted total100%3.9red

Key metrics

Key metricLowBaseHighSource or assumption [n]
Probability of success to the next phase (A.6)30%45%60%The next phase is a filing; base rate 47.7% (BIO 2021, oncology, Phase III to filing), cut for the FDA recommendation, held up by the company’s stated intent to file [4]
Probability of success to approval (A.6)5%12%22%Base rate 43.9% to approval, cut hard for the FDA’s request for a survival trial [4][10]
Peak sales, USD per year (B.4a)$50M$146M$312MBottom-up US model, conditional on approval [UNVERIFIED — modelled]
eNPV, USD (B.4b)−$60M+$5M+$250MRisk-adjusted, after the cost of a survival trial [UNVERIFIED — modelled]
Market Access Score (B.3)3.2Mean of the five B.3 criteria, lib/scorecard.mjs

Top-3 box

  • Top-3 value drivers. 1. A randomised trial met its primary endpoint: response rate 18.0% against 5.3% [4]. 2. Progression-free survival nearly doubled, 4.7 against 2.6 months (hazard ratio 0.44) [4]. 3. No approved drug serves the 70–80% of second-line patients without a targetable mutation, and the drug holds Orphan Drug and Fast Track designations [5][11].
  • Top-3 risks, each with its mitigation. 1. The FDA wants a survival trial before a filing [4]; mitigation: further FDA engagement, which the company says it is pursuing [4]. 2. Survival was flat and the crossover-adjusted figure was worse (hazard ratio 1.13) [12]; mitigation: none in the current data. 3. Severe high blood pressure in 44% of patients [12]; mitigation: standard blood pressure drugs, untested as a fix at scale.
  • Evidence gaps, and whether each is still fixable at this stage. 1. Duration of response is unpublished. Fixable on 2026-10-24. 2. No peer-reviewed clinical result exists. Fixable by publication within months. 3. No survival benefit and no quality-of-life data. Fixable only by a new trial, which takes years and money the company has not said it will spend.

Asset verdict

Asset verdict: Pass. The regulator has said the current data are not enough to file, and the survival result gives no sign that a new trial would succeed. Monitor only if the 2026-10-24 duration figure is strong and the FDA later accepts a filing, or a partner funds the survival trial. · confidence Medium

This verdict judges the drug’s long-term value. It did not set any of the three scored calls in Locked prediction (01-rules.md rule 45).

Visual 1: the seven domain scores

Seven domain scores

Source(s): the domain table above, computed by lib/scorecard.mjs from the scores in A.2, A.3, A.4, A.5, M.6, B.1, B.3 and C [4][12][3][1]

Visual 2: milestones

Milestones

Source(s): CT.gov [3]; Compass releases [4][8][12]; designation coverage [11][13]; BLA target from secondary reports only [9]

Visual 2b: clinical study timeline

Clinical study timeline

Source(s): dates and n from CT.gov [3] and the design paper [7]; NCT06548412, NCT05513742, NCT07392957 and NCT07662031 from CT.gov search, read 2026-10-17. Date basis: registry dates as of the search date. COMPANION-002’s bar runs to its registry primary completion date, 2027-08-31, which moved from 2026-12-01 between the two sweeps (see Readout).

Studies supporting key target-profile claims.

  1. COMPANION-002 [4][12][3]: the response-rate, progression-free survival, survival and safety claims in A.1.
  2. The Korean Phase 1b/2 with paclitaxel (NCT04492033) [7]: the 6.9-month duration of response and the 37.5% response rate used as reference points.
  3. The first-in-human study (NCT03292783) and its population pharmacokinetic model [7][14]: the 10 mg/kg dose and the dosing claims.

Rationale for the clinical development plan. Compass ran one randomised Phase 2/3 trial with response rate as its primary endpoint. That design aimed at accelerated approval, as three other second-line biliary drugs have won it [15][16][17]. Allowing crossover protected patients and gave away the survival comparison, which is now the FDA’s main concern [4].

Key Go / No-Go to the next decision point. The next decision is a filing (A.3, filing column). Go: a median duration of response of 5 months or more on 2026-10-24, and an FDA position that accepts a filing on response rate and progression-free survival. No-Go: a duration below the 4.7-month median progression-free survival, or an FDA position that keeps a survival trial as a condition. The FDA position is undated; the company said on 2026-09-22 it would “further engage” [4].

Clinical pharmacology studies

Study (ID or NCT)TypeDesign and nKey findingLabel or dosing impactSource
NCT03292783Dose escalation (RP2D)Phase 1 monotherapy, nine dose levels 0.3–17.5 mg/kg, n=45No dose-limiting toxicity; responses at ≥10 mg/kg; 10 mg/kg carried forward [VERIFIED — design paper]10 mg/kg on days 1 and 15 of 28[7]
—Drug–drug interactionNot conductedNo public record as of 2026-10-17None known[3][7]
—QTcNo dedicated studyCOMPANION-002 excluded patients with QTcF above 450 ms [VERIFIED — CT.gov]None known[3]
—Hepatic or renal impairmentNot conductedTrial required bilirubin ≤1.5× the upper normal limit and creatinine clearance ≥30 mL/min [VERIFIED — CT.gov]Untested in severe organ impairment[3]
NCT03292783 dataPopulation PK and exposure–responseModel on 712 samples from 30 patientsTwo-compartment model; body weight the only covariate; fixed and weight-based doses give equal exposure; 15 mg/kg every 3 weeks matches 10 mg/kg every 2 weeks. No exposure–response model [VERIFIED — PubMed]Supports a less frequent schedule, not yet tested[14]
—Dose optimisationNot conductedNo randomised comparison of two doses, which the FDA’s Project Optimus asks for in oncology. No public record as of 2026-10-17A possible review question[3]
—Immunogenicity, food effect, ADME, ethnic bridgingNot reportedNo public record as of 2026-10-17. Food effect does not apply to an infused antibodyNone known[7][14]

0. Program-tier coverage — CLEARED

Every mandatory row was called. Company-tier coverage is in ../company.md C.0. The manufacturing tier was run, because the program sits at its filing decision after a Phase 2/3 pivotal trial. The payer tier was not, because the launch plan names no EU, UK or Japan market [2]. The regulatory tier was not, because the catalyst is a congress presentation, not a regulatory decision.

ToolStateNote / verbatim error
CT.gov search_trials + get_trial_details on the pivotal NCTCALLEDsearch_trials on “CTX-009 OR tovecimig OR ABL001” returned 63 trials, most of them Novartis’s asciminib, which shares the code ABL001; the 5.10.3 query without ABL001 had found the 6 tovecimig trials. get_trial_details on NCT05506943 returned 168 patients, 34 US sites, six outcome measures, primary_completion_date 2027-08-31 (2026-12-01 on 2026-08-19), has_results false, and no investigator names.
PubMed search_articles + get_article_metadata on every hitCALLEDA first query including “ABL001” returned 389 hits, the asciminib collision 02-connectors.md warns about for short codes. A restricted query returned the same 4 papers as in August. Metadata was retrieved for all 4, author lists included.
Open Targets search_entitiesCALLEDReturned DLL4 (ENSG00000128917), VEGFA (ENSG00000112715), cholangiocarcinoma (MONDO_0019087), biliary tract cancer (MONDO_0003060) and tovecimig (CHEMBL6068308). No throttle message this time. A follow-up association query found no DLL4 association with either disease, and VEGFA scores of 0.152 (cholangiocarcinoma) and 0.118 (biliary tract cancer) [18].
ChEMBL compound_searchCALLEDCHEMBL6068308 TOVECIMIG, “Antibody”, max_phase 2, USAN stem -mig [19]. ChEMBL’s small-molecule activity tables do not answer an antibody’s selectivity question, so the absence of binding data costs nothing here.
web_search ×4: peak sales · competitive · exclusivity + royalty · analystCALLEDPeak sales: one analyst figure found, William Blair’s $1.46 billion worldwide, rejected in B.4a with its audit trail. Competitive: second-line options and pipeline. Exclusivity and royalty: the 10-K and 10-Q answered both (patents from 2033; tiered single-digit royalties). Analyst: post-2026-09-22 rating changes, recorded in ../company.md C.8.
B.1a competitor efficacy and side-effect search (v2.2 prompt P4)CALLEDPrimary papers retrieved through PubMed for ABC-06, NIFTY, HERIZON-BTC-01, FIGHT-202, FOENIX-CCA2 and ClarIDHy [5][20][21][22][23][24].
Orange Book and Purple BookCALLEDWeb search. Tovecimig is an unlicensed biologic, so it has no Purple Book entry yet. Orange Book entries for the small-molecule competitors were reached only through third-party mirrors and are not used [25].
DailyMedCALLEDAPI. Current labels read for zanidatamab (Ziihera), pemigatinib (Pemazyre), futibatinib (Lytgobi), ivosidenib (Tibsovo) and durvalumab (Imfinzi) [15][16][17][26][27].
ICERCALLEDWeb search. No ICER assessment of a biliary tract cancer drug was found as of 2026-10-17 [28].
CMS Medicare drug price negotiation pageCALLEDWeb search. The 2025 law widened the orphan exclusion [29][30].
optional EDGAR full-text search on the drug’s namesNOT CALLEDOptional row. The 10-K, 10-Q and 8-K were read directly from EDGAR instead.
optional Europe PMC searchNOT CALLEDOptional row. PubMed answered with full metadata.
optional CTIS searchNOT CALLEDOptional row. All 34 COMPANION-002 sites are in the US [3].
Manufacturing tier: CMC search (v2.2 prompt P5)CALLEDTwo web searches on tovecimig with CRL, Form 483, warning letter, shortage, recall and contract-manufacturer terms. No contract manufacturer is named anywhere. No public record of a CMC problem as of 2026-10-17.
Manufacturing tier: openFDA CRL dataCALLEDAPI. “Compass”, “ABL” and “tovecimig” return NOT_FOUND. The endpoint itself answers (458 letters in total), so the empty result is real [31].
Manufacturing tier: openFDA CRL web pageCALLEDRead through a web search on the sponsor’s name. No letter found.
Manufacturing tier: FDA warning lettersCALLEDWeb search on the sponsor’s name. None found. The contract site is unnamed, so no site-level search could be run.
Manufacturing tier: FDA inspections dashboardCALLEDWeb search, as the method column directs. No record found. Same limit as the row above.
Manufacturing tier: openFDA drug shortagesCALLEDAPI. Tovecimig and paclitaxel return NOT_FOUND; the endpoint holds 1,581 records, so the result is real [32].

A. Scientific and clinical value

A.1 Target product profile

Target label. Tovecimig plus paclitaxel for adults with unresectable advanced, metastatic or recurrent biliary tract cancer whose disease progressed after one gemcitabine- and platinum-containing regimen, and who are not eligible for a targeted drug [3].

AttributeMinimum viableCompetitive (base)High caseStandard of care and key competitors, data [n]Current evidence for tovecimig [n]Rating · confidence
Indication and populationSecond line, all comersSecond line, all comersSecond line plus first lineFOLFOX for all comers [5]; zanidatamab (HER2), pemigatinib and futibatinib (FGFR2), ivosidenib (IDH1) for marker-selected patients [15][16][17][26]Second line, all comers; excludes patients eligible for a targeted drug [3]green · High
Mechanism and differentiationA new mechanism with a plausible rationaleFirst in class with clinical proofBest in class across angiogenesis drugsNo anti-angiogenic drug is approved in this diseaseFirst DLL4 × VEGF-A bispecific in a pivotal trial; no anti-DLL4 drug has ever been approved [19]amber · Medium
Efficacy (primary and key secondary endpoints)Response rate above FOLFOX’s with no survival harmResponse rate and progression-free survival gains plus a survival trendA survival benefitFOLFOX: survival 6.2 against 5.3 months on symptom care (hazard ratio 0.69) [5]; nal-IRI + 5-FU: progression-free survival 4.2 against 1.7 months (hazard ratio 0.61) [20]Response rate 18.0% against 5.3%; progression-free survival 4.7 against 2.6 months (hazard ratio 0.44); survival 8.9 against 9.4 months (hazard ratio 1.05) [4][12]red · High
Safety and tolerabilityManageable, no new class signalGrade 3+ rate near FOLFOX’sBetter tolerated than FOLFOXFOLFOX: grade 3–5 adverse events 69% against 52% on symptom care [5]High blood pressure 69% all grades, 44% grade 3+; neutropenia 36% grade 3+; fatigue 67% [12]red · Medium
Dose, regimen, route, formulationIntravenous, standard infusionSame visits as the comparatorEvery three weeksFOLFOX every two weeks, intravenous [5]Intravenous 10 mg/kg every two weeks with weekly paclitaxel; a three-weekly schedule is modelled, not tested [14]amber · Medium
Biomarker or companion diagnosticNone neededNone neededA marker that picks respondersCompetitors need HER2, FGFR2 or IDH1 tests [15][16][17][26]None tested or planned [3]amber · Medium
Payer value (B.3)Reimbursed at an orphan priceReimbursed without restrictionPremium price on a survival gainFOLFOX is genericNo survival gain and no quality-of-life data; Market Access Score 3.2 (B.3)red · Medium

Stakeholder check

StakeholderWhat they requireDoes the competitive profile meet it? (tagged)Basis [n]
Patient and caregiverBenefit the patient notices; conveniencePartly. Two more months before the cancer grows, at the cost of severe high blood pressure in 44% and no longer life [VERIFIED — Compass 2026-04-27] · Medium[12]
RegulatorA favourable balance of benefit against riskNo, as of 2026-09-22. The FDA recommended a trial showing a survival benefit before a filing [VERIFIED — 8-K exhibit 99.1] · High[4]
Payer and HTA bodyAdded benefit against the comparator, at a price they acceptNo. The trial compared against paclitaxel, not FOLFOX, and shows no survival gain [UNVERIFIED — judgement] · Medium[3][5]
PrescriberA fit in treatment guidelines; ease of usePartly. It would fill the all-comer gap, but blood-pressure management adds work [UNVERIFIED — judgement] · Medium[12]

Value link. No sourced figure exists for what a response-rate gain is worth in this disease. The illustrative oncology link in 07-benchmarks.md § Rules of thumb values each extra month of overall survival at roughly $150–300 million of peak sales [UNVERIFIED — illustrative]. Applied here it gives nothing, because the survival gain is zero. Kept from the 5.10.3 version, which stated the same unflattering result.

A.2 Rationale and evidence

Tovecimig is one antibody with two jobs. Its main body binds VEGF-A, the protein a tumour releases to make blood vessels sprout toward it. Two small fragments on its ends bind DLL4 on the vessel cells inside the tumour [7]. When VEGF-A is blocked alone, tumours can switch on the DLL4–Notch signal as an escape route. Blocking both shrank tumours more than blocking either one in mouse studies, and the regressing vessels let chemotherapy reach more of the tumour [VERIFIED — Yeom et al. 2020, PubMed] [33][34].

Tovecimig blocks VEGF-A and DLL4 together, closing the escape route that anti-VEGF drugs alone leave open

Source(s): design paper [7]; preclinical papers [33][34]

How well is the target validated? Unevenly. VEGF-A blockade is proven in other cancers, but Open Targets finds no association at all between DLL4 and cholangiocarcinoma or biliary tract cancer. Its VEGFA association score for cholangiocarcinoma is 0.152, most of it from literature [VERIFIED — Open Targets association query, 2026-10-17] [18]. No anti-DLL4 drug has ever been approved. The 44% rate of severe high blood pressure is the expected on-target effect of blocking these vessel signals [12].

The exact step the next readout must prove. That the tumour shrinkage lasted. Duration of response is the one pre-specified secondary endpoint never reported [3]. It matters more now. Every accelerated approval in second-line biliary cancer rested on “overall response rate and duration of response” [15][16][17]. A long duration is the company’s best argument against the FDA’s request.

The honest scientific risk. The drug shrank tumours and delayed progression, and patients did not live longer. The company blames crossover. Its own crossover-adjusted analysis made the result worse (A.2 evidence below). That contradiction is what the FDA’s 2026-09-22 recommendation addresses [4].

Evidence quality

The seven scores are recorded in asset_evaluation.scores[]. Their mean, 3.7, is the Scientific rationale domain score from lib/scorecard.mjs.

CriterionHigh (7–10)Medium (4–6)Low (1–3)Where to checkScoreBasis (tagged) [n] · confidence
Target validationHuman genetic evidence, or a clinical precedent with the same targetStrong preclinical data, indirect human dataPreclinical only, or conflicting dataOpen Targets, GWAS Catalog, literature5VEGF-A blockade has clinical precedent in other cancers; DLL4 has no Open Targets link to this disease and no approved drug [VERIFIED — Open Targets 2026-10-17] [18] · Medium
Mechanism and target engagementProof of mechanism in people (PK and PD, receptor occupancy)A PD signal with an unclear dose relationNo human PD dataPhase 1 data, publications5A pharmacokinetic model exists [14]; high blood pressure is a pharmacodynamic sign that VEGF is blocked; no receptor-occupancy or vessel-imaging data published [VERIFIED — PubMed] · Medium
Translational biomarkersA validated PD or selection biomarker with an available assayAn exploratory biomarkerNonePublications, FDA biomarker qualification2None tested or planned [3] · High
Clinical effect against standard of careRandomised, clinically meaningful effectSingle-arm or small randomised signalNo signal, or a weak oneRegistries, publications, congress data5Randomised gains in response rate and progression-free survival, against paclitaxel rather than FOLFOX; survival hazard ratio 1.05 [4][12] · High
Consistency and replicationReplicated across trials or subgroupsOne trialInconsistentTrial results4The Korean Phase 2’s 37.5% response rate fell to 18.0% under randomisation; the colorectal trial failed [7] · Medium
Therapeutic window and safetyWide margin, no class liabilityManageable side effectsDose-limiting toxicity or a class liabilityToxicology, clinical safety, FAERS, class labels344% grade 3+ high blood pressure, an on-target class effect, against no survival gain [12] · Medium
Data transparency and expert viewPeer-reviewed; supportive named experts (A.2b)Abstracts only; mixed viewsPress releases only; sceptical expertsPubMed, congresses, A.2b2No clinical result is peer-reviewed; every efficacy figure is from a press release; A.2b finds no independent voice (judgement UNKNOWN) [UNVERIFIED — expert view score] · Medium

The single most important evidence finding, updated. The design paper says the trial “was designed to detect a 23% absolute difference” in response rate with 90% power [7]. The final difference is 12.7 percentage points (18.0% against 5.3%), a little over half the design target [4]. It cleared significance (p=0.0228) because the control arm also did worse than the design assumed. “Significant” and “as large as planned” are different facts. Kept from 5.10.3, which computed 11.8 points from the earlier 17.1% response rate.

The crossover argument does not survive its own check. Kept from 5.10.3 unchanged. 54% of control patients (31 of 57) crossed over [12]. The standard check, the RPSFT analysis, gave a survival hazard ratio of 1.13, worse than the unadjusted 1.05 [12]. Removing the effect of crossover made the drug look worse. The supportive subset the company cites (crossover patients 12.8 months against 6.1 months for non-crossover patients, hazard ratio 0.54, p=0.04) compares patients who lived long enough to cross over against those who did not. That is survivorship bias, not evidence. The FDA’s August 2025 draft guidance asks sponsors to treat survival as a safety endpoint, to rule out harm, and to limit crossover for this very reason [VERIFIED — FDA draft guidance page] [10].

Endpoints

Endpoint (full name)What it measuresScale / rangeBetter directionMCID / note
Overall response rate (ORR), the primary endpointThe share of patients whose tumours shrink enough to count as a complete response (all measurable disease gone) or a partial response (the sum of target-lesion diameters down at least 30%), by RECIST version 1.10–100%HigherNo established minimal clinically important difference. Final 18.0% (20/111) against 5.3% (3/57), p=0.0228 [4].
Duration of response (DoR), the figure 2026-10-24 turns onAmong responders, the time from the scan that first confirmed the response to the scan that first confirmed growthMonthsLongerNo established difference. Reference points: 6.9 months in the Korean Phase 2 [7]; the arm’s own 4.7-month median progression-free survival. Not yet reported.
Progression-free survival (PFS)Time from randomisation until the cancer grows or the patient diesMonthsLonger4.7 against 2.6 months, hazard ratio 0.44, p<0.0001 [4]. In an open-label trial it is the endpoint most exposed to bias.
Overall survival (OS)Time from randomisation until death from any causeMonthsLonger8.9 against 9.4 months, hazard ratio 1.05, p=0.78; RPSFT-adjusted 1.13 [12]. The endpoint the FDA has asked to be shown.
Disease control rate (DCR)The share whose tumours shrank or stayed stable0–100%HigherRegistry secondary endpoint [3]. Not reported.
Treatment-emergent adverse events (TEAEs)Any medical problem arising during treatment, graded 1 (mild) to 5 (fatal) on the CTCAE version 5.0 scaleCounts and percentages by gradeFewer, milderGrade 3 means severe. Grade 3+ high blood pressure 44% [12].
EORTC QLQ-C30 (quality of life)A 30-item patient questionnaire on functioning and symptoms0–100 per scaleHigher for functioning, lower for symptomsIn the design paper [7], not in the registry’s outcome list [3], never reported.

A.2b Named experts

Panel as of. 2026-10-17.

Investigators

CT.gov search_investigators again returned zero rows (condition “biliary tract cancer”, name “Azad”), and get_trial_details on NCT05506943 carries no investigator block [3]. As in 5.10.3, the investigators below come from the design paper’s author list, each tied to a site on the trial’s own CT.gov list [3][7].

NameRole / affiliationTrial (NCT)Conflicts (party, kind, disclosed in, as of)Conflicts checked (source, date)SourceTag
Nilofer AzadFirst author of the design paper; presenter of abstract 459O on 2026-10-24; Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins, a trial siteNCT05506943Sponsor: trial investigator and designated presenter, disclosed in Compass’s release of 2026-07-17, as of 2026-07-17. No competitor relationship found.PubMed get_article_metadata PMID 38861293, 2026-10-17: author list returned, no financial-disclosure section. CT.gov search_investigators, 2026-10-17: zero rows.[8][7]VERIFIED — Compass release 2026-07-17 and PubMed 2026-10-17
Milind JavleSenior author of the design paper; MD Anderson Cancer Center, a trial site and sponsor of the first-line study NCT06548412NCT05506943Sponsor: trial investigator, disclosed by authorship of the design paper, as of 2024-06-04. No competitor relationship found.PubMed metadata PMID 38861293, 2026-10-17: no disclosure section.[7]VERIFIED — PubMed 2026-10-17
Rachna ShroffDesign-paper author; University of Arizona Cancer Center, a trial siteNCT05506943Sponsor: trial investigator, disclosed by authorship, as of 2024-06-04.PubMed metadata PMID 38861293, 2026-10-17: no disclosure section.[7]VERIFIED — PubMed 2026-10-17
Lipika GoyalDesign-paper author; Stanford Cancer Center, a trial site. Also first author of the futibatinib FOENIX-CCA2 paper, a competitor in FGFR2-altered disease.NCT05506943Sponsor: trial investigator, disclosed by authorship, as of 2024-06-04. Competitor: lead author of Taiho’s futibatinib trial, disclosed in the NEJM paper’s byline, as of 2023-01-19.PubMed metadata PMIDs 38861293 and 36652354, 2026-10-17: no disclosure section in either payload.[7][23]VERIFIED — PubMed 2026-10-17
Minori RosalesDesign-paper author at “Compass Therapeutics, 80 Guest Street, Boston”; a sponsor employeeNCT05506943Sponsor: employment, disclosed by the affiliation on the paper, as of 2024-06-04.PubMed metadata PMID 38861293, 2026-10-17: affiliation read directly.[7]VERIFIED — PubMed 2026-10-17

A limit, kept from 5.10.3. Neither PubMed payload carries a disclosure section, so “no competitor relationship found” records that a search was made, nothing more (02-connectors.md § KOL sources). This sweep did find one competitor relationship by reading the author lists: Lipika Goyal led the futibatinib trial [23]. That row is new.

Independent voices

None found. The literature is still four papers, none of them independent clinical commentary [7][33][34][14]. No named clinician has commented publicly on the 2026-09-22 FDA position in any source this sweep reached. The critical coverage is from analysts, not clinicians: Leerink, quoted by BioSpace, said the FDA “simply recommended a study but did not say it was mandatory” [35]. An analyst is not a key opinion leader under 01-rules.md rule 40, so that view is not recorded in this table.

NameAffiliationView (close enough to quote)As ofConflicts checked (source, date)SourceTag

The ESMO discussant of abstract 459O on 2026-10-24 will be the first independent named view. It is on the What to watch list.

Judgement

Endpoint supportedBasis (one or two sentences)Tag
UNKNOWNNo independent clinician has commented on this endpoint in any source reached, and no clinical result is peer-reviewed. The FDA’s recommendation is a regulator’s position, recorded in A.4, not an expert view.UNVERIFIED — judgement

Dissent

NameView (close enough to quote)Source

Empty, correctly: the judgement is UNKNOWN, so there is no claimed support to contest.

A.3 Go / No-Go questions

The next decision is a filing. The filing column marks the questions answered for it. The others are answered too, for the record.

#QuestionGo to Phase 2Go to Phase 3FilingAnswer (tagged, at most three points) [n]RatingConfidence
Target
T1Is proof of mechanism or principle established in people (target engagement, PD) with this candidate?●●High blood pressure shows VEGF blockade; tumours shrank; no direct DLL4 engagement data [VERIFIED — Compass 2026-04-27] [12][14]amberM
T2If the early population differs from the intended population, how is the evidence translated?●●The Korean Phase 2 (37.5% response) did not translate to the US randomised trial (18.0%) [4][7]amberM
T3Is there a rationale for combinations, and is it pursued?●●●Paclitaxel synergy in mice [33]; a first-line combination with gemcitabine, cisplatin and durvalumab is under way (NCT06548412) [3]greenM
Dose and drug
D1Is the exposure–response relationship and the therapeutic window defined for the intended regimen and route?●●●No exposure–response model [14]; 44% grade 3+ high blood pressure at 10 mg/kg [12]redM
D2Is the dose and regimen compatible with the clinical and non-clinical safety data?●●●No dose-limiting toxicity up to 17.5 mg/kg in Phase 1 [7]; the combination’s blood-pressure rate is high [12]amberM
D3Does the formulation deliver therapeutic exposure, and is it commercially viable (the final formulation for Phase 3)?●●●Intravenous antibody; validation batches under way at unnamed contract sites [1]amberL
D4Are intrinsic and extrinsic factors (organ impairment, drug interactions, ethnicity, body weight) characterised?●●Body weight only [14]; no organ-impairment or interaction studiesamberM
Patient and design
P1Are the design and endpoints accepted by the FDA and EMA, clinically meaningful, competitive, and relevant for HTA bodies (B.3)?●●●No. The FDA recommended a survival trial before a filing [VERIFIED — 8-K exhibit 99.1] [4]; the comparator is not FOLFOX [3][5]redH
P2What is the rationale for the population; is a selection biomarker validated, with a companion diagnostic plan?●●●All comers is a strong commercial rationale; no selection marker [3]amberM
P3What is the probability of success of the next trial (effect-size assumptions, statistical power, trial simulation or assurance)?●●A survival trial is not designed. The randomised data (hazard ratio 1.05) give no sign it would succeed [12]redL
P4Has proof of concept with a positive benefit–risk balance been shown in the intended population?●●Concept yes (response rate, progression-free survival); positive benefit–risk no, on the FDA’s reading [4][12]redM
Benefit–risk
R1Is the safety database adequate for the indication (exposure, duration); are class effects or a REMS expected?●●111 treated in the tovecimig arm plus 31 crossover patients; the blood-pressure class effect is known [12]amberM
R2Is the overall benefit–risk balance favourable against the current standard of care?●●Not shown: no survival gain over paclitaxel and no comparison with FOLFOX [12][5]redM

The Clinical data and design domain is a direct score of 4 (amber). Basis: P1, P3, P4, R2 and D1 are red; T3 is green; the rest amber. The randomised design and a met primary endpoint keep it above the red band. · confidence Medium

A.4 Regulatory pathway

TopicQuestionsAnswer (tagged) [n]Where to look
PathwayWhich route: NDA or BLA, 505(b)(2), EU centralised MAA, conditional authorisation or accelerated approval? Is a confirmatory trial required, and is it feasible?A BLA for accelerated approval on response rate and progression-free survival was the plan [36]. On 2026-09-22 the FDA “recommended that Compass conduct a trial demonstrating a survival benefit before proceeding with a Biologics License Application” [VERIFIED — 8-K exhibit 99.1] [4]. Compass “does not believe that a new trial is warranted” and is preparing a BLA anyway [4]. A filing can be submitted against the FDA’s advice; the FDA can refuse to file it or reject it. Leerink estimated a new trial would mean about a three-year delay [WEB ESTIMATE — BioSpace, 2026-09-22] [35].FDA guidance [10]
DesignationsBreakthrough Therapy, Fast Track, RMAT, Orphan Drug, priority review voucher (FDA); PRIME, orphan (EMA); Sakigake, orphan (PMDA): granted, or realistic?Fast Track (2024-04-25) and Orphan Drug (announced by 2026-04-08) are held [11][13]. No Breakthrough Therapy designation. No EU or Japanese designation found as of 2026-10-17.FDA registers
Agency alignmentEnd-of-Phase-2 meeting, scientific advice, SPA, parallel EMA and HTA advice: was the pivotal design agreed?No Special Protocol Assessment is disclosed. The design allowed crossover [7]. The FDA’s 2026-09-22 position followed a meeting the company called “discussions last month” [WEB ESTIMATE — BioSpace quoting the release, 2026-09-22] [35].Company disclosures
PrecedentHave drugs been approved on the same endpoint and design in this indication? Any recent rejections in the class?Three accelerated approvals in second-line biliary cancer rest on “overall response rate and duration of response”: zanidatamab (HER2), pemigatinib (FGFR2) and futibatinib (FGFR2) [VERIFIED — DailyMed labels] [15][16][17]. All three came from single-arm trials in marker-selected patients. Tovecimig is different: it has a randomised survival result, and it points the wrong way. The FDA’s draft guidance treats that as a harm signal to rule out [10].Drugs@FDA, DailyMed
Review riskLikelihood of an advisory committee; safety signals; manufacturing inspection; co-approval of a diagnostic; post-marketing requirementsIf a BLA is filed without a survival trial, an advisory committee is likely [UNVERIFIED — judgement]. No diagnostic is needed. The contract manufacturing sites would face their first pre-approval inspection (M.3).AdComm materials

The Regulatory pathway domain is a direct score of 3 (red). Basis: the agency has recommended a survival trial before any filing; the designations held do not lower the evidence bar. · confidence High

Designations.

  • Fast Track, granted 2024-04-25. It grants more frequent FDA meetings and a rolling review; it does not lower the evidence bar [VERIFIED — BPIQ note 4/25/24; OncLive 2024-04-25] [13].
  • Orphan Drug designation for biliary tract cancer. It grants seven years of US exclusivity for this disease from approval, tax credits and a fee waiver [VERIFIED — CancerNetwork, 2026-04-08] [11]. Named correction: the 5.10.3 version dated it “around 2026-05-05”, the date of the BPIQ note; press coverage shows it was announced by 2026-04-08.
  • No Breakthrough Therapy designation. None appears in the note history or the press feed [UNVERIFIED — absence of evidence].

A.5 Clinical operations

Development timeline

Source(s): CT.gov [3]; design paper [7]; Compass releases [4][12]; competitor trial dates [5][20][21][22]

Milestone dates. Enrolment completed 2024-08-06 [VERIFIED — BPIQ note]. The primary-endpoint readout came on 2025-04-01 and the survival readout on 2026-04-27 [12]. The registry’s primary_completion_date moved from 2026-12-01 (read 2026-08-19) to 2027-08-31 (read 2026-10-17) [3]. The registry does not say why. It is consistent with longer survival follow-up, but that reading is [UNVERIFIED].

Trials

TrialSponsor / whose drugDesign (arms, blinding, n)PopulationStatus / key resultLink
COMPANION-002 (NCT05506943), pivotalCompass; its own drugPhase 2/3, multicentre, open-label, randomised 2:1, tovecimig + paclitaxel against paclitaxel; n=168 at 34 US sites; crossover allowed at progressionAdults with advanced biliary tract cancer after one gemcitabine-platinum regimen, performance status 0–1, not eligible for a targeted drugActive, not recruiting. Response rate 18.0% against 5.3% (p=0.0228); progression-free survival 4.7 against 2.6 months (hazard ratio 0.44); survival 8.9 against 9.4 months (hazard ratio 1.05). Duration of response unreported.NCT05506943 [4][12]
NCT04492033, the Korean study COMPANION-002 was built onHandok; ABL Bio’s drugPhase 1b/2a, open-label, single-arm; n=41Advanced biliary tract cancer after one or two therapiesTerminated. Phase 2 cohort (n=24): response rate 37.5%, median duration 6.9 months, progression-free survival 9.4 monthsNCT04492033 [7]
NCT03292783, first in humanABL Bio; its own drugPhase 1 dose escalation, nine levels, 0.3–17.5 mg/kg; n=45Heavily pretreated solid tumoursCompleted. No dose-limiting toxicity; high blood pressure 37.8%[7]
COMPANION-003 (NCT05513742)Compass; its own drugPhase 2, single-arm; n=49Colorectal cancer after two or three regimensCompleted; did not advance. Response rate 5% (2/41)NCT05513742
NCT06548412, first lineMD Anderson; Compass’s drugPhase 1/2, open-label; n=50; with gemcitabine, cisplatin and durvalumabUntreated advanced biliary tract cancerRecruiting; primary completion 2027-05-01NCT06548412
NCT07662031 and NCT07392957Washington University; Compass’s drugsPhase 2 (n=25, colorectal) and Phase 1b/2 (n=54, glioblastoma)Other cancersInvestigator-sponsored; not this programNCT07392957
ABC-06, competitor precedentUK academic group; generic chemotherapyPhase 3, open-label, randomised 1:1, FOLFOX + symptom care against symptom care; n=162Second-line biliary tract cancer after cisplatin and gemcitabineSurvival 6.2 against 5.3 months (hazard ratio 0.69, p=0.031); grade 3–5 events 69% against 52%[5]

Pivotal programme

TrialSettingDesign / comparatorKey result [n]Use
COMPANION-002Second line, all comersRandomised against paclitaxel, open-label, crossover allowedResponse rate and progression-free survival met; survival not met [4][12]Intended for US accelerated approval; the FDA recommends a survival trial first [4]

Effect by trial

Not applicable as a cross-trial comparison — one randomised trial so far. The table records that trial’s own effects for reference.

TrialEndpointEffect (hazard ratio or other, named)95% confidence intervalnSource
COMPANION-002Progression-free survivalHazard ratio 0.44Not disclosed in the sources read168[4]
COMPANION-002Overall survivalHazard ratio 1.05 (unadjusted); 1.13 (RPSFT-adjusted)Not disclosed in the sources read168[12]

Operational parameters

ParameterHigh (7–10)Medium (4–6)Low (1–3)Where to checkScoreBasis (tagged) [n] · confidence
Recruitment feasibilitySites and countries adequate for the disease’s prevalence; enrolment on or ahead of benchmarkAdequate, with competing trials for the same patientsRare population, few sites, enrolment delaysClinicalTrials.gov, EU CTIS8168 patients at 34 US academic sites, enrolment complete 2024-08-06 [3] · High
Endpoint acceptanceRegulatory precedent in FDA or EMA guidanceNovel endpoint with documented agency agreementUnvalidated surrogateFDA and EMA guidance, EPARs3Response rate has precedent in this disease [15][16][17], but the FDA has asked for survival here [4] · High
Design robustnessRandomised, controlled, agreed with the agencies (SPA)Adaptive or seamless, with a pre-specified interimSingle-arm without a robust external controlProtocol, SPA, scientific advice5Randomised and controlled, but open-label, 2:1, crossover allowed, no SPA [3][7] · High
Execution track recordSponsor met its prior milestonesMinor delaysRepeated delays, CRO or site problemsCompany history, 8-K filings, ../company.md C.1a4Survival data a month late; BLA guidance missed; CTX-471 start missed five dates (../company.md C.1a) · High

Resourcing sufficiency. For the 2026-10-24 presentation, yes: the data are locked and the slot is booked. For what comes next, not clearly. The company held $179.9M at 2026-06-30 and guides to a runway “into 2028” (../company.md C.3). That guidance predates the FDA’s request for a survival trial. A new randomised survival trial in this disease would take years (B.4c). The company has not said whether it will run one or how it would pay for it. Its 39-person staff and missed guidance (C.1a) argue against a fast start.

A.6 Probability of success

Base rate. baseRate in lib/success-rates.mjs was run with phase phase3, area oncology and modality bispecific. The phase is phase3 because COMPANION-002 is the registrational Phase 2/3 trial and the next step is a filing. BIO 2021 has no bispecific row, so the lookup falls back to the oncology row.

Base rate: 47.7% to the next phase (Phase III to NDA/BLA) · 43.9% to approval · source BIO 2021, Figure 2, p. 7 · fallback level: area (oncology) · Wong 2019: not broken out for Phase 3 to the next phase; 35.5% for Phase 3 to approval [WEB ESTIMATE — BIO/Informa/QLS 2021]

Adjustments

FactorDirection (up / down)Evidence [n]Tag
Biomarker-selected populationdownNone; all comers. Programs with a selection biomarker approve more often; this one has none [3]VERIFIED — CT.gov
Human genetic support for the targetdownNo Open Targets association between DLL4 and the disease; VEGFA 0.152 [18]VERIFIED — Open Targets 2026-10-17
Rare disease or orphan designationupOrphan Drug and Fast Track designations; a clear unmet need [11][13]VERIFIED — designation coverage
Randomised Phase 2 with a meaningful effectupRandomised gains in response rate and progression-free survival (hazard ratio 0.44) [4]VERIFIED — 8-K exhibit 99.1
First-in-class, novel endpoint, or prior failures in the classdownNo anti-DLL4 drug approved; survival hazard ratio 1.05, 1.13 after crossover adjustment [12]VERIFIED — Compass 2026-04-27
Other, named: the FDA recommended a survival trial before a BLAdownThe regulator’s own position on these data, 2026-09-22 [4]VERIFIED — 8-K exhibit 99.1

Result

LowBaseHighBasis
To the next phase30%45%60%The next phase is a filing. The company says it will file; the FDA’s advice, the cost and the refuse-to-file risk pull the figure down from 47.7% [UNVERIFIED — modelled]
To approval5%12%22%Approval on the current data needs the FDA to reverse its advice; approval after a new survival trial needs that trial to succeed where the current one showed no trend [UNVERIFIED — modelled]

A.7 Modality

  • Modality: bispecific. A full IgG antibody against VEGF-A with two single-chain fragments against DLL4 fused to its ends (“IgG with 2× C-terminal scFv”) [7][19].
  • Manufacturing is assessed in M below. Its CMC score feeds the Operations and CMC domain.

M. Manufacturing (CMC)

The full form applies: the program sits at its filing decision after a Phase 2/3 pivotal trial. The manufacturing tier of the sweep was run (section 0). CMC means chemistry, manufacturing and controls: how the drug is made and quality-checked.

M.1 Product and process

ElementDescribeThis asset [n]
Molecule or constructFor an antibody: isotype and expression system. For a conjugate: linker chemistry, payload class, conjugation site, DAR and its distribution. For a gene therapy: vector or capsid. For an RNA drug: the lipid nanoparticle. For a radioligand: the isotopeA humanised IgG against VEGF-A with two anti-DLL4 single-chain fragments fused to the heavy-chain ends [7]. Isotype and cell line: no public record as of 2026-10-17.
Drug substance stepsUpstream (growing the cells), downstream (purifying the product), payload and linker synthesis, conjugation, purification; in-house or CDMO for each stepAll at unnamed contract manufacturers. Compass owns no plant [VERIFIED — Form 10-K for 2025] [2]
Drug productFormulation (liquid or freeze-dried), presentation, device, stability and shelf life, refrigerated transport (cold chain)Intravenous infusion; formulation and shelf life not disclosed
AnalyticsThe critical quality attributes (such as DAR, free payload, aggregates, potency assay, vector genomes) and the release and characterisation methodsNot disclosed. A dual-binding potency assay would be expected for a bispecific [UNVERIFIED — judgement]
Companion diagnosticThe assay, its partner, and the timing of its co-approval (A.4)Not applicable — no biomarker

M.2 Challenges by process step

Process stepTypical challengesStatus and readiness (tagged) [n]Rating
Drug substance (biologic)Titre, cell-line stability, glycosylation, scale-upClinical supply delivered through a 168-patient trial; commercial-scale status undisclosed [UNVERIFIED]amber
Payload and linkerHPAPI synthesis and containment, supply chain, impuritiesNot applicable — not a conjugatenot applicable
Conjugation or assemblyDAR control and distribution, unconjugated antibody, aggregation, yieldNot applicable as conjugation. Fused single-chain fragments can raise aggregation risk [UNVERIFIED — class judgement]amber
Analytics and comparabilityPotency assay, CQA definition, comparability after a process change or a site transferNot disclosed [UNVERIFIED]amber
Drug product and stabilityFormulation, freeze-drying, shelf life, cold chainNot disclosed [UNVERIFIED]amber
Scale-up and tech transferCommercial-scale process, transfer to several sites, validation (PPQ)The 10-Q lists “registration and validation batches” among manufacturing costs and $7.5M of accrued manufacturing expense, mostly tovecimig [VERIFIED — Form 10-Q filed 2026-08-06] [1]amber
Capacity and supplyCapacity against the demand forecast, CDMO dependency, dual sourcing, shortages or over-capacityFully dependent on unnamed CDMOs; dual sourcing not disclosed [2]amber
Cost of goodsCOGS against the price corridor (B.3); the effect of dose (mg per kg) and yieldA 10 mg/kg dose every two weeks is a lot of antibody per patient; no cost disclosed [UNVERIFIED]amber
Regulatory CMCPre-approval inspections, Form 483 observations, CRLs, post-approval changesNo CRL, warning letter or inspection record found for the sponsor [31]; the sites are unnamed, so site records could not be searchedamber

M.3 Regulatory CMC track record

ItemWhat to recordWhere to find itFinding (tagged) [n]
CMC-related CRLsDate, product, reason, time to resolutionopenFDA’s complete response letter data, company press releasesNo public record as of 2026-10-17 (openFDA CRL API, NOT_FOUND for Compass, ABL and tovecimig) [31]
InspectionsForm 483 observations and warning letters at the drug substance, drug product and CDMO sitesFDA warning letters, the FDA inspection dashboardNo public record as of 2026-10-17 for the sponsor. CDMO sites unnamed, so not searchable
Review documentsThe product quality review and post-marketing CMC commitments (FDA); the quality section of the EPAR (EMA)Drugs@FDA, EMA EPARsNot applicable — no application filed
Supply eventsShortages, recalls, allocationFDA drug shortages, company filingsNo public record as of 2026-10-17 (openFDA shortages API) [32]

M.4 Network and capacity

The company-level facts are in ../company.md C.1a: no owned plants, unnamed contract manufacturers.

TopicFinding (tagged) [n]
Sites for each step (own or CDMO), regions, planned investments and their timing against demandAll contract sites, unnamed; validation batches were being made in 2026 for a planned BLA [1]. A survival trial would delay the commercial need by years [UNVERIFIED — judgement]
Geopolitical and tariff exposure; single points of failureUnknown, because the sites are unnamed. The originator is Korean; whether any drug substance is made in Asia is not disclosed [UNVERIFIED]
Capacity strategy: risk-adjusted or maximum-demand planning; take-or-pay contractsNot disclosed. Validation-batch spending ahead of an FDA position that now points to delay creates a write-down risk [UNVERIFIED — judgement]

M.5 Modality questions

ModalityKey CMC and technology questionsAnswer for this asset (tagged) [n]
Small moleculeSolubility and bioavailability, polymorphs, route of synthesis at scale, nitrosamine risk—
Antibody, protein or bispecificTitre and yield, stability, high-concentration subcutaneous formulation, immunogenicityTitre, stability and immunogenicity: no public data as of 2026-10-17. Intravenous only; no subcutaneous form planned. A 10 mg/kg dose makes yield matter for cost [UNVERIFIED]
Antibody-drug conjugateLinker and payload stability, DAR control, payload toxicity, payload supply—
Cell therapyVein-to-vein time, manufacturing success rate, capacity, cost of goods, treatment-centre network—
Gene therapyVector yield and potency assay, dose-related toxicity, durability, redosing, cost of goods—
RNA or oligonucleotideDelivery (liver against other tissues), dosing interval, class safety (platelets, kidney)—
Radioligand or otherIsotope supply, logistics and half-life, site infrastructure—

M.6 CMC score and top risks

CriterionHigh (7–10)Medium (4–6)Low (1–3)ScoreBasis (tagged) [n] · confidence
Process robustness and CQAsValidated commercial process; well-defined CQAsProcess changes pending; comparability neededQuality problems recognised late5Validation batches in progress; nothing published on quality attributes [1] · Low
Regulatory CMC recordNo CMC-related CRL or Form 483 issueMinor observations, resolvedA CMC CRL, a warning letter or a refusal6No record found, but no application has been inspected yet [31] · Low
Capacity and supplyCapacity matched to demand; dual sourcingConstrained, or a single siteShortages, or a major over-capacity write-down5Unnamed CDMOs; no dual sourcing disclosed [2] · Low
Cost of goodsCompatible with the price corridorMargin pressureCOGS threatens viability5High dose per patient; orphan pricing likely covers it [UNVERIFIED] · Low

CMC score: 5.3 (mean of the four, from lib/scorecard.mjs)

Top-3 CMC risks. 1. Contract sites unnamed, so no inspection history can be checked [2]. 2. Validation spending ahead of a delayed filing could be written off [1]. 3. Bispecific fusion formats can aggregate; no stability data are public [UNVERIFIED].


B. Commercial value

B.1 Treatment today and competitors

Treatment algorithm. First line is gemcitabine and cisplatin, now usually with durvalumab (TOPAZ-1) [VERIFIED — Imfinzi label] [27] or pembrolizumab [UNVERIFIED — not read this sweep]. At progression the path forks by tumour test:

  • About 13% of intrahepatic cholangiocarcinomas carry an IDH1 mutation and go to ivosidenib [VERIFIED — ClarIDHy abstract] [24][26].
  • FGFR2 fusions go to pemigatinib or futibatinib, both accelerated approvals [16][17].
  • HER2-positive (IHC 3+) disease goes to zanidatamab, an accelerated approval [15].
  • Everyone else gets chemotherapy: FOLFOX on the strength of ABC-06 [5], or liposomal irinotecan with 5-FU on the strength of NIFTY [20]. The share without a targetable alteration is about 70–80% [WEB ESTIMATE — Frontiers in Oncology 2026 review and DelveInsight, via the 2026-08-19 sweep; URL not recorded].

Where tovecimig would fit. In that last group, added to paclitaxel. COMPANION-002 excludes patients eligible for a targeted drug [3].

Competitors

Competitor (drug, company)StageKey data [n]Note (indirect comparison)
FOLFOX (generic)Standard of care, off-labelABC-06: survival 6.2 against 5.3 months, hazard ratio 0.69; grade 3–5 events 69% [5]Indirect. Unlike COMPANION-002, ABC-06 shows a survival benefit.
Liposomal irinotecan + 5-FU/LV (nal-IRI)Off-label; Phase 2b evidenceNIFTY (Korea): progression-free survival 4.2 against 1.7 months, hazard ratio 0.61 [20]Indirect; Korean population; a German trial (NALIRICC) was not read this sweep.
Zanidatamab (Ziihera), JazzApproved (accelerated), HER2-positive onlyResponse rate 41.3%; grade 3 treatment-related events 18% [21][15]Indirect; single-arm; marker-selected.
Pemigatinib (Pemazyre), IncyteApproved (accelerated), FGFR2 onlyResponse rate 35.5%; grade 3+ events 64% [22][16]Indirect; single-arm; marker-selected.
Futibatinib (Lytgobi), TaihoApproved (accelerated), FGFR2 onlyResponse rate 42%; median duration of response 9.7 months [23][17]Indirect; single-arm; marker-selected.
Ivosidenib (Tibsovo), ServierApproved, IDH1 onlyProgression-free survival 2.7 against 1.4 months on placebo, hazard ratio 0.37 [24][26]Indirect; marker-selected; randomised against placebo.

Competitive position

MetricHigh value (7–10)Medium value (4–6)Low value (1–3)Where to checkScoreBasis (tagged) [n] · confidence
Mechanism differentiationFirst-in-class, or clearly best-in-classNext-generation improvementMe-tooPublications, the competitor list above7First DLL4 × VEGF-A bispecific in a pivotal trial; no anti-angiogenic drug approved in this disease [19] · Medium
Timing against competitorsMore than 12 months ahead of the nearest competitorWithin 6 to 12 monthsLagging by more than 12 monthsClinicalTrials.gov5No other randomised all-comer drug is near filing, but a survival trial would push tovecimig back about three years [35] · Medium
Clinical data against competitorsSuperior efficacy, safety or convenience (B.1a)ComparableInferiorPublications, congress data4Better response rate than FOLFOX, no survival gain, more severe high blood pressure (B.1a) · Medium
IP and exclusivity10 years or more after launch (composition of matter)5 to 10 years, or method of use onlyUnder 5 years, or pending onlyOrange Book, Purple Book, patent databases6Two licensed patent families, 2 US patents, expiring “starting in 2033, subject to possible extension” [VERIFIED — Form 10-K for 2025] [2]; a new biologic gets 12 years of US exclusivity from licensure and the orphan grant 7 years [UNVERIFIED — statutory default] · Medium

Named correction. The 5.10.3 version said no patent estate could be found. The 10-K describes it. Whether the 2033 patents cover the molecule itself is not stated.

Strategic insight. In oncology, first movers with novel mechanisms have shown roughly 3.2× higher peak sales and 1.8× higher development risk than fast followers [UNVERIFIED — illustrative] (07-benchmarks.md § Rules of thumb). Tovecimig now carries the second half of that line. It wins only if the field accepts tumour control without a survival gain. The single fact the case rests on is unchanged from 5.10.3 but has now been tested: whether the FDA accepts response rate and progression-free survival without survival. On 2026-09-22 the answer was no [4].

B.1a Efficacy against side effects

Every comparison here is indirect (01-rules.md rule 17). Patients, lines of therapy, follow-up and side-effect definitions differ across these trials. Three of the four points are single-arm or marker-selected. Tovecimig’s point is early data and uses a lower bound for harm.

Efficacy against side effects, second-line biliary tract cancer

Source(s): COMPANION-002 [4][12] · ABC-06 [5] · HERIZON-BTC-01 [21] · FIGHT-202 [22]. Indirect comparison across trials (rule 17). Scaling: x = response rate ÷ 50%; y = 1 − (grade 3+ rate ÷ 100%).

IndicationRegimen (trial)Efficacy (endpoint, value)Grade 3 or higher adverse events (%)Adverse event definition (any cause, treatment-emergent or treatment-related)Source
2L BTC, all comersTovecimig + paclitaxel (COMPANION-002)Response rate 18.0%At least 44% (grade 3+ high blood pressure alone; the overall rate is not published)Treatment-emergent, by event[4][12]
2L BTC, all comersPaclitaxel (COMPANION-002 control)Response rate 5.3%Not published—[4]
2L BTC, all comersFOLFOX + symptom care (ABC-06)Survival 6.2 months; response rate about 5% [WEB ESTIMATE — via the 2026-08-19 sweep]69% (grade 3–5)Any cause[5]
2L BTC, all comersnal-IRI + 5-FU/LV (NIFTY)Progression-free survival 4.2 monthsNot in the abstract read—[20]
2L BTC, HER2-positiveZanidatamab (HERIZON-BTC-01)Response rate 41.3%18% (grade 3)Treatment-related[21]
2L CCA, FGFR2 fusionPemigatinib (FIGHT-202)Response rate 35.5%64%Any cause[22]
2L iCCA, FGFR2 fusionFutibatinib (FOENIX-CCA2)Response rate 42%Hyperphosphataemia 30% (grade 3, treatment-related); overall rate not in the abstractTreatment-related[23]
2L CCA, IDH1Ivosidenib (ClarIDHy)Progression-free survival 2.7 monthsAscites 7% the most common; overall rate not in the abstractAny cause[24]

B.2 Addressable market

Launch market: the United States only. Compass holds rights outside South Korea and China but has described no European or Japanese launch [2]. Orphan market, so the guide’s 100,000-patient threshold does not apply (01-rules.md rule 9).

ComponentHow it is estimatedAssessment guideThis asset (tagged)Source [n]
Drug-treated patientsPrevalence or incidence × diagnosis rate × treatment rate × eligible shareMore than 100,000 (US, EU5, Japan) is a premium threshold for a common disease; judge an orphan market relative to its areaAbout 2,080–3,190 US patients a year (funnel in B.4a). Two incidence figures disagree and are used as the two ends, not averaged: 18,600 (American Cancer Society, quoted in the design paper) and 12,130 (a SEER-based estimate). Kept from 5.10.3.ACS [VERIFIED — via the design paper] [7]; SEER [WEB ESTIMATE — epidemiology papers, via the 2026-08-19 sweep]
ExclusivityPatent term + extensions (PTE, SPC) + regulatory, orphan or paediatric exclusivityYears of exclusivity after launch; the generic or biosimilar entry datePatents from 2033 plus possible extension [2]; 12 years of biologic exclusivity from approval [UNVERIFIED — statutory default]; 7 years orphan [11]. Likely 10 years or more from a launch near 2029–2030[2][11]
Reimbursement precedentApprovals and payer decisions in two or more major marketsPositive NICE or CADTH guidance for a drug like this oneNone for tovecimig. Nearest analogue: NICE recommended ivosidenib only with a confidential discount [37][37]
Price corridorComparator and analogue prices; a value-based priceWorked out in B.3Modelled at $20,000–30,000 a month, orphan-oncology range [UNVERIFIED — assumption]B.3
Patient and system impactHospitalisations, caregiver burden, HRQoLThe size of the effect and the evidence for itNot shown; blood-pressure monitoring adds burden; no quality-of-life data [12][7][12][7]

B.3 Payers and price

The United States is assessed. The EU, Germany, the UK and Japan are not in the launch plan, so their rows say Not applicable and the payer tier was not run.

Per market

MarketIn scope?Expected outcomePrice implicationKey riskSource [n]
United States: ICER benchmark; Medicare negotiation exposure (IRA timing, orphan exclusion)AlwaysCoverage likely if approved, as for other orphan cancer drugs; no ICER review of any biliary drug foundOrphan-level pricing; a biologic is not negotiable until 11 years after approval, and the widened 2025 orphan exclusion would shelter an orphan-only label [WEB ESTIMATE — Milliman, 2026]No survival gain to justify a price above generic chemotherapy[29][30][28]
EU joint clinical assessment (JCA)NoNot applicable — no EU launch plan——[2]
Germany: G-BA and IQWiG (AMNOG)NoNot applicable — no EU launch plan——[2]
UK: NICENoNot applicable — no UK launch plan——[2]
Japan: ChuikyoNoNot applicable — no Japan launch plan——[2]

Questions

QuestionAnswer (tagged) [n]
Comparator: does the pivotal trial compare against what payers require (the G-BA’s appropriate comparator, the NICE scope, the JCA’s PICO, the US standard of care)? If not, is a credible bridge planned, such as an indirect comparison or an external control?No. The comparator is paclitaxel; the US standard of care is FOLFOX [3][5]. No bridge disclosed.
Endpoints and certainty: patient-relevant endpoints (mortality, morbidity, quality of life, safety) or surrogates? Will payers discount the design (single-arm, crossover, immature data)?Surrogates (response rate, progression-free survival); survival flat; crossover confounds it [4][12].
Supplementary evidence: quality-of-life data (EQ-5D), early payer advice (EU joint scientific consultation, G-BA, NICE), real-world or natural-history data, an early cost-effectiveness model?None reported. QLQ-C30 was planned but never reported [7].
Value against price: a realistic G-BA added-benefit category; cost per QALY against the NICE threshold; the ICER price benchmark. Do they support the target price against the comparator?With no survival or quality-of-life gain, the QALY gain is close to zero, so no cost-per-QALY threshold supports a premium [UNVERIFIED — judgement].
Access conditions: likely restriction to sub-populations, managed access or confidential discounts, post-launch data obligations?A confirmatory survival trial would be a condition of any accelerated approval [10].
US comparator annual cost (list price), the ICER benchmark, and the Medicare negotiation timing for this drug (small molecule 7 years, biologic 11 years after approval; orphan exclusion)FOLFOX is generic. No ICER benchmark exists. As a biologic, 11 years; the widened orphan exclusion may apply [29].

Market access criteria

CriterionHigh (7–10)Medium (4–6)Low (1–3)ScoreBasis (tagged) [n] · confidence
Comparator alignmentRandomised trial against the payers’ comparatorRobust indirect comparisonPlacebo or single-arm only4Randomised, but against paclitaxel, not FOLFOX [3] · High
Endpoints and certaintyHard endpoint, mature, randomisedSurrogate, immatureUnvalidated surrogate, single-arm3Surrogates only; survival flat [4][12] · High
Supplementary evidence planQuality of life (EQ-5D), payer advice, real-world evidence, a cost-effectiveness modelPartialNot addressed2Nothing reported [7] · Medium
Value for money at the target priceWithin the NICE threshold, or at or below the ICER benchmarkNeeds a discountFar above the thresholds3No QALY gain against a generic [UNVERIFIED — judgement] · Medium
Payer precedentPositive, with a premium priceRestricted or mixedRejected, or priced at the comparator4Marker-selected biliary drugs are reimbursed; ivosidenib only with a discount in England [37] · Medium

Market Access Score: 3.2 (mean of the five, from lib/scorecard.mjs)

Payer decisions

MarketIndication / evidenceOutcome (body’s own words, date)Main reasonSource [n]
England (NICE TA948)Ivosidenib, IDH1-mutant cholangiocarcinoma after one or more treatmentsRecommended “only if the company provides it according to the commercial arrangement” [WEB ESTIMATE — search summary of the NICE page, 2026-10-17]Cost-effectiveness needed a discount[37]

B.4 Valuation and financing

B.4a Peak sales

Built bottom-up, US only, conditional on approval, gross of ABL Bio’s royalty. Kept from 5.10.3. Launch now assumed near 2029–2030 rather than 2028, because a survival trial or a contested filing both push it back. The uptake shape is in Visual 4.

Patient funnel: 12,130–18,600 US diagnoses a year × ~68% advanced × ~70% treated in first line × ~45% fit for second line × ~80% without a targetable alteration = 2,080–3,190 patients [UNVERIFIED — assumptions, kept from 5.10.3]. Price: $20,000–30,000 a month × about 5 months of treatment = $100,000–150,000 a patient [UNVERIFIED — assumption].

ScenarioAssumptions (patients × price × share × duration)Estimate (USD/year at peak)Basis / tag
Low2,000 × $100,000 × 25%$50MNarrow approval; most oncologists stay with FOLFOX [UNVERIFIED — modelled]
Base2,600 × $125,000 × 45%$146MApproval; adoption in about half of eligible patients [UNVERIFIED — modelled]
High3,200 × $150,000 × 65%$312MApproval plus a durable response that reframes survival, plus first-line pull-through [UNVERIFIED — modelled]

The weakest input is the price, then penetration. Kept from 5.10.3. No tovecimig price exists and no all-comer add-on has launched in this disease. Patient count is the strongest input.

Rejected input, with its audit trail. The 5.10.3 sweep found no published peak-sales figure. This sweep found one: William Blair’s $1.46 billion worldwide in previously treated biliary cancer, resting on “about 15,000 patients annually in the United States likely eligible” [WEB ESTIMATE — William Blair initiation, as summarised by search coverage; pre-dates the 2025 readout] [38]. It is not used. 15,000 eligible second-line patients is close to the total number of US biliary cancer diagnoses (12,130–18,600), before any loss to early-stage disease, death before second line or targeted therapy. It also predates the April 2026 survival result and the FDA’s position. LifeSci Capital’s “around $1.4 billion” US opportunity is set aside for the same reasons.

B.4b Value

ScenarioeNPV (USD)Assumptions (discount rate, costs, royalties and milestones, loss of exclusivity, probability of success)Tag
Low−$60MValue at approval $90M (low peak × 2 × 0.9) × 5% approval odds, less a survival trial costing about $70M that the company runs and fails [UNVERIFIED — modelled]UNVERIFIED — modelled
Base+$5MValue at approval $460M (base peak × 3.5 × 0.9) × 12% × about 0.64 for four years of discounting at 12%, less the risk-weighted trial costUNVERIFIED — modelled
High+$250MValue at approval $1,404M (high peak × 5 × 0.9) × 22% × about 0.8, the FDA accepting a filing without a new trialUNVERIFIED — modelled

No single figure, per 01-rules.md rule 10. The 5.10.3 range was $150–700M, before the FDA asked for a survival trial. The market’s current valuation, about −$10M to −$25M of enterprise value (../company.md C.4), sits at the bottom of this range.

B.4c Funding

QuestionAnswer (tagged) [n]
Capital needed to the next decision pointEffectively nil to 2026-10-24; the data are locked [8].
Capital needed to approval, and the funding planNot disclosed. A randomised survival trial in this disease would cost tens of millions and take years [UNVERIFIED — judgement]; Leerink estimated about three years of delay [35]. The $400M shelf and $100M ATM are unused (../company.md C.3).
Cash runway against those needs (../company.md C.3)$179.9M at 2026-06-30, “into 2028” by the company’s own guidance, which predates the FDA feedback (../company.md C.3).
Financing need and dilution risk before the next decision (../company.md C.3, dilution history)No raise is likely before 2026-10-24. After it, a raise at about cash value would be heavily dilutive. A partner is the other path.

B.4d Launch, partnering, rights and economic share

QuestionAnswer (tagged) [n]
Launch capability: can the company launch alone, or must it partner?It has no product and no sales force, a Chief Commercial Officer since January 2026, and 39 staff (../company.md C.1a). A small US specialty launch is within reach; nothing beyond it.
Partnering history and comparable deal termsABL Bio in-licence and the Elpiscience China sub-licence (../company.md C.1a). No US or European partner.
Commercialisation rights: retained, split, or out-licensed?US and rest of world retained, except South Korea (Handok) and China (out-licensed to Elpiscience) [1][2].
Economic share0.9 [UNVERIFIED — judgement]. Basis: Compass keeps the US economics. It owes ABL Bio tiered single-digit royalties and up to $303M of commercial milestones in oncology [VERIFIED — Form 10-Q filed 2026-08-06] [1]. A mid single-digit royalty plus milestones spread over peak years leaves roughly 90%.

Visual 4: scenario chart

Annual sales by scenario

Source(s): B.4a assumptions, launch assumed in 2029 with peak around 2034 [UNVERIFIED — modelled] [7][1]. Lines: low, base and high, in that order.


C. Risk

Framing questions

  1. Does the development plan support the target product profile claims (A.1)? For response rate and progression-free survival, yes. For survival, no: the design allowed crossover, and the FDA now asks for a trial that shows it.
  2. Would the identified risks change the target product profile? Yes. Without a survival result the regulator row of A.1 fails, and the payer row fails with it.
  3. If a risk cannot be mitigated, is the asset still differentiated from competitors? On response rate, yes. But a response-rate edge without a survival gain, at 44% severe high blood pressure, is a weaker product than FOLFOX in the eyes of the regulator.
Risk groupCoversTime / quality / cost riskLikelihoodImpactMitigationSource [n]
Science and clinicalResearch, biomarker, clinical pharmacology, clinical efficacy, drug safetyQuality: flat survival, crossover-adjusted hazard ratio 1.13, no biomarker, 44% grade 3+ high blood pressurehighhighNone in hand. A safety signal of this kind is a near-veto for a drug with no survival gain.[4][12]
Non-clinicalDMPK, safety pharmacology, toxicologyNo toxicology finding found; the pharmacokinetics are modelledlowlow—[14]
RegulatoryPathway, endpoints, review, post-marketing requirementsTime and quality: the FDA recommends a survival trial before filinghighhighFurther FDA engagement; a filing against advice[4][10]
CMC and supplyProcess, scale-up, comparability, capacity, diagnostics (from M)Unnamed contract sites; validation spending that a delay could strand; aggregation risk in the format (M.6)mediummediumStandard CDMO qualification[1][2]
IP and legalPatents, freedom to operate, litigation, licencesPatents from 2033; ABL Bio terms; investigation notices without a filed suitlowmediumBiologic and orphan exclusivity[1][2]
Access and commercialEvidence and value, payers, pricing, competitionNo survival or quality-of-life gain against a generichighmediumOrphan pricing; a niche US launch[5]
Execution and financingProject management, clinical operations, funding, partnersRepeated slips; a survival trial needs money the shares cannot raise cheaplymediumhigh$180M cash; unused shelf and ATM[1]

The Risk profile domain is a direct score of 3 (red). Basis: the regulatory and the science rows are both high likelihood and high impact; execution and financing is high impact. · confidence Medium

Visual 5: risk heat map

Risk heat map

Source(s): the risk table above [4][12][1][2]

Top-3 risks. 1. The FDA’s request for a survival trial before filing [4]. 2. Flat survival and the crossover check that made it worse [12]. 3. Severe high blood pressure in 44% of patients [12].


D. Look-back

Not applicable until settled. The look-back is written at settlement in lookback.md beside this file (06-prompts.md § Settlement).


Readout

Sources

KindWhere it comes fromWhat it is worthValue (or null)TagNote
bpiqfetch_company_drugs, fields catalyst_date and catalyst_date_textThe one source every program has. On most rows it is built from period text, so on its own it is the latest plausible date (a ceiling), not an estimate.2026-10-24VERIFIED — BPIQ fetch_company_drugs 2026-10-17Text “October 24, 2026” names a day, so it is not a placeholder. catalyst_source is now the 2026-09-22 release.
ctgovCT.gov get_trial_details, field primary_completion_dateIndependent of the company’s own messaging. Known to the month. It moves when the trial moves.2027-08-31VERIFIED — CT.gov NCT05506943, read 2026-10-17Moved from 2026-12-01 (read 2026-08-19). It dates the trial’s own completion, not the presentation. Recorded, not used for the window.
companyfetch_company_press_releasesThe company’s most recent dated statement. Also the source of the Date slippage table below.2026-10-24VERIFIED — Compass release 2026-07-17; 8-K exhibit 99.1 2026-09-22The July release gives Saturday 2026-10-24, 10:15–11:45 central European time, presenter Nilofer Azad [8]. The 2026-09-22 8-K reaffirms “the expectation to present the complete dataset at a medical conference this year” [4].
congressdata/congresses.json, only when the company has said it intends to present there (02-connectors.md § Data limits)Answers where the result will be presented.2026-10-23/2026-10-27VERIFIED — data/congresses.json, ESMO Congress 2026Admissible because the company has said it will present there [8][39]. Supplies the window’s outer edges.
modelledTrial arithmetic, see belowThe only estimate independent of anyone’s guidance.2027-10/2027-12UNVERIFIED — modelled, default lag2027-08-31 plus two to four months. Dates the trial’s final readout, a later and different event. Not used for the window.

The modelled estimate. Registry primary completion 2027-08-31, plus the default two to four months to database lock and analysis (01-rules.md rule 34), gives 2027-10 to 2027-12. data/benchmarks/readout-lag.json holds no matching entry.

Window

EarliestLikeliestLatestPrecisionConfidence
2026-10-232026-10-242026-10-27DAYHIGH

Basis. The company named the day and time of the talk, and the BPIQ note and the 2026-09-22 8-K reaffirm it. The congress’s opening and closing days bound the window. The talk is on a Saturday morning in Madrid, so the first US trading session that can react is Monday 2026-10-26.

A correction to the 5.10.3 basis. That basis said the earliest date “additionally covers ESMO’s own pre-congress abstract release making the headline numbers public on the opening day”. Neither sweep verified ESMO’s 2026 embargo policy: the policy page returned 404 and the abstracts page loads without text. One fact bears on it. On 2026-10-10 an OncoDaily preview listed abstract 459O and quoted the already-public response and progression-free survival figures, but no duration of response [6]. That fits an embargo until the talk. If ESMO instead publishes the abstract online before 2026-10-23, the readout would come earlier than this window says. That risk is [UNVERIFIED] and is carried in the Run-up exit choice below.

Disagreement. CONSISTENT. Three sources name the same day and the congress source brackets it. The ctgov and modelled rows date a different event, the trial’s final completion. Kept from 5.10.3.

Date slippage. Five slips, unchanged since 2026-07-17. Every slip moved a reporting date; none was an enrolment delay. The two newest statements reaffirm October 24.

As ofGuidance text
2025-02-27”Report topline data by the end of Q1 2025”
2025-08-11”Secondary endpoints including OS data expected in Q1 2026”
2025-11-05”OS/PFS secondary endpoint data expected late Q1 2026; potential first BLA filing planned H2 2026”
2026-03-05”PFS/OS secondary endpoint analyses from COMPANION-002 expected in April 2026”
2026-04-27”COMPANION-002 met key PFS endpoint; complete dataset with DoR to present later 2026; FDA meeting planned”
2026-07-17”Oral ESMO 2026 presentation of COMPANION-002 Ph2/3 BTC results on October 24, 2026”
2026-08-06”Final ORR 18%; FDA feedback Q3 2026; ESMO oral data October; BLA planned 2026”
2026-09-22”FDA advised survival-benefit trial before BLA; company will further engage FDA. ESMO data Oct 24, 2026.”

Attribution

Status.

CONTAMINATED

Computed by attributionFor in lib/clustering.mjs over the six pipeline rows in ../company.md C.2, with CATALYST_CLUSTER_MIN_MONTHS = 6, for bpiq_drug_id 17398. Copied, not estimated.

Conflicts

bpiq_drug_idLabelDateAnalysed?Gap (days)Confirmed?
17481CTX-83712026-12-31No0Yes
18722CTX-10726 (PD-1 x VEGF-A bispecific antibody)2026-12-31No0Yes

Note. Both conflicts are confirmed because this program’s own window is disclosed to the day. CTX-8371 and CTX-10726 carry only “Q4 2026”, which the module widens to the whole half-year from 2026-07-01 to 2026-12-31. So a known day sits inside two half-year windows. Either Phase 1 readout could land in the days around 2026-10-24. Both programs are rated meaningful in C.2, and the company’s value is now almost entirely cash plus these three programs, so a move in the stock-call window cannot be credited to tovecimig alone. A third event is not in the table because it is not a separate pipeline row: the company’s answer to the FDA, whose timing is unknown. Kept in substance from 5.10.3, which reached the same status from the same two rows.


Market and timing for this event

  • Plain takeaway. The shares trade at the company’s cash. The market prices almost nothing for tovecimig after the FDA’s request. The 2026-10-24 talk will add one new number, the duration of response, and the first independent expert reaction. A good duration helps the case for a filing; it does not answer the FDA’s survival concern.
  • Months to this catalyst. 0.2 months (6 days) from the 2026-10-17 lock to readout.window.earliest of 2026-10-23. Precision is DAY, so this is a real countdown.
  • Expected move around this event. ../company.md C.6 records the chain as unreadable: no strike at spot, and a $1.00 straddle quoted anywhere from 29% to 76% of spot. A bracket of ±10% to ±35% is used instead. Calibration: Phase 3 readouts average +11% and −22%, oncology +9% and −9%, and companies under $1 billion react 6.2 times more strongly at Phase 3 [WEB ESTIMATE — IQVIA 2024]. This ticker’s own presentation-type events moved +0.8% to +11.0% over the event window, and its hypothesis-level events −35% and −65% (C.7).
  • Nearest comparable past reaction. C.7’s 2024-11-08 row, the CTX-471 congress presentation: +0.8% over the event window. It is the same kind of event, a talk on data already described. It is a poor match in size: a minor asset then, the only late-stage asset now, and a stock at cash with 25.6% of its float sold short now.
  • Materiality. ../company.md C.2 rates this program dominant. The stock call below is consistent with that. A dominant program presenting a known dataset, after the regulator has already spoken, is a smaller event than its materiality alone suggests.
  • Date slippage. Five slips; see Readout. None since 2026-07-17.

Spot. $0.86, the settled close of Friday 2026-10-16, read from data/prices/CMPX.json through lib/prices.mjs after the 2026-10-17 fetch (../company.md C.4). It is a settled close, not an intraday snapshot: it matches BPIQ’s last price and Yahoo’s daily close.

Cash floor.

Cash floor (USD per share)Cash, debt, quarterly burn (as of)Event dateEconomic share countResidual pipeline valueTag
$0.86$179.879M cash and securities, no debt, $14.738M quarterly burn (2026-06-30)2026-10-24186,797,915$0UNVERIFIED — computed by lib/edge.mjs cashFloor

cashFloor returns $0.86, net cash at the event of $161.2M over 1.27 quarters. At the first-half average burn of $16.0M a quarter it would be $0.85. The residual pipeline value is $0, with a reason. C.2 rates CTX-8371 and CTX-10726 meaningful, and rule 30 allows their value to be added. No defensible figure exists for two Phase 1 assets with no partner and no deal comparable read this sweep, so none is added (rules 9 and 10). The miss row’s basis says how they could hold the price above the floor. Spot equals the cash floor.

Scenario prices

ScenarioLowHighAnchors (each named, tagged, with its kind)Basis
Positive$0.87$1.15(1) Cash floor $0.86 [UNVERIFIED — computed by lib/edge.mjs cashFloor] · cash_floor. (2) This ticker’s own positive data moves over the event window, median +1.0% (2025-04-01 +1.0%, 2024-11-08 +0.8%, 2024-08-12 +11.0%), which gives $0.87; maximum +11.0%, which gives $0.95 [VERIFIED — data/prices/CMPX.json event-window computation, ../company.md C.7] · own_past_moves.The low edge is the price if the duration just reaches 5.0 months. It lands in the “missed expectations” bucket of 07-benchmarks.md: it clears the bar, but it falls short of the 6.9-month Korean figure and does not touch the FDA’s survival concern, so the shares stay near cash. The high edge is the price if the duration clearly beats 6.9 months and the discussant is supportive: the “significantly exceeded” bucket. It sits above the ticker’s own +11.0% maximum, for three named facts: 25.6% of the float is short (../company.md C.5); spot is the cash floor, so the stock prices no value for the asset; and a long duration is the company’s best answer to the FDA. Calibration: oncology +9% × 6.2 for a company under $1 billion [WEB ESTIMATE — IQVIA 2024]. Far below the commercial ceiling allowance of $6.77.
Miss$0.62$0.84(1) Cash floor $0.86 [UNVERIFIED — computed by lib/edge.mjs cashFloor] · cash_floor. (2) This ticker’s own negative tovecimig moves over the event window, median −49.9% (2026-04-27 −64.6%, 2026-09-22 −35.2%), which gives $0.43 [VERIFIED — data/prices/CMPX.json event-window computation, ../company.md C.7] · own_past_moves.A duration below 5.0 months, or none reported, removes the company’s best argument for filing. The range sits below the floor for named reasons: spending continues on three programs at about $15M a quarter; a Nasdaq bid-price notice can arrive from about 2026-11-19 (../company.md C.4); and an at-the-money ATM makes dilution cheap to start. It sits far above the −49.9% own-history anchor, because those two moves came from prices well above cash. Cash cannot be sold twice. The two Phase 1 readouts in Q4 2026 are what could hold it near the floor. The miss is shallower than the positive row is high, against the class default, because spot sits at the cash floor, the reason 07-benchmarks.md names as legitimate.

Expectation benchmark. The Korean Phase 2’s 6.9-month median duration of response is the only published reference point [7]. With the shares at cash, the market appears to price no route to a filing at all [UNVERIFIED — judgement].

Expected value. 45% × the positive midpoint ($1.01) + 55% × the miss midpoint ($0.73) = $0.856, or −0.5% against the $0.86 spot. Method: probability_pct applied to the midpoint of each range. This is arithmetic, not advice, and not a price target.

Break-even probability.

Break-even pointBreak-even range (low–high)Our probability (band)Clearance (pp)Call (callFromBands)Evidence for the gap (each named and tagged)
46.4%6.5%–96.0%45% (32%–58%)noneno-edge — band [32, 58] overlaps range [6.5, 96]None

The range is 89.5 points wide because spot sits almost on the miss row’s high edge. The figure comes from our own scenarios, not from the market.

What a positive result can be worth.

Value at approval (low / base / high)Ceiling priceAllowance (1.5x, or 3x with a named exception)Ratio to market cap
$90M / $459.9M / $1,404M$4.80$6.77 (1.5×); $12.68 (3×, no exception claimed)1.556

Computed by commercialCeiling in lib/edge.mjs. Inputs: B.4a peak sales; economic share 0.9 (B.4d); the sales multiple of 2, 3.5 and 5 and the 1.5× allowance, both [UNVERIFIED — default]; the probability of approval after a positive Phase 2 readout, 52.4% [WEB ESTIMATE — BIO/Informa/QLS 2021]; 186,797,915 shares; a $154.9M market cap. The Phase 2 figure is used instead of the Phase 3 one (90.6%), because the FDA’s request puts a confirmatory survival trial ahead of approval. The ratio is high because the market cap equals cash. It is an upper bound that assumes none of the program is priced in, which is literally true at this price.

Run-up

readout.precision is DAY and readout.confidence is HIGH, so date_confidence scores 100. Rule 39 withholds a run-up call only when the precision is UNKNOWN. It does not apply here, so a call is made.

Entry and exit

Entry dateEntry priceEntry basisExit rule
2026-10-17$0.86The lock date, a Saturday, at the settled close of Friday 2026-10-16. The first session in which the position could exist is Monday 2026-10-19.T-2 trading days before readout.window.earliest, which resolves to the close of 2026-10-21. exit is null at lock: resolveExit returns null because the cache ends before the window.

Why T-2 and not T-5. T-5 before 2026-10-23 is 2026-10-16, the entry’s own price bar. A T-5 exit would close the position before it opened. T-2 is chosen over T-1 to leave one more session before a possible early online abstract (Readout, correction).

Predicted move and peak

LowHighEstimated dateBasis
Predicted move, entry to exit−5%+15%—Three sessions. For a rise: the price sits at cash with 25.6% of the float short (C.5), and a known Saturday event invites pre-positioning. Against: the shares fell from $1.02 to $0.86 over the past seven sessions, and tax-loss selling below $1 is a seasonal headwind. 07-benchmarks.md does not calibrate run-up drift. The band includes a loss because a band that excluded one would be a hope, not a forecast.
Predicted peak, from entry+3%+30%2026-10The peak may come after the exit, on Monday 2026-10-26, if the duration is good. The high edge matches the positive row’s high edge (+33.7%), less a little fade.

Priority score drivers

DriverReadsScore (0–100)Basis
Unmet-need relevanceThis document’s A.1 and B.1. A judgement, with no formula.85No approved drug for the 70–80% of second-line patients without a targetable alteration; FOLFOX adds 0.9 months of survival (B.1, A.1). Held below 100 because marker-selected subsets have drugs.
Value-uplift potentialstock_direction.commercial_ceiling.ratio_base, computed by valueUpliftScore in lib/edge.mjs.88commercial ceiling ratio R=1.556 (approval-adjusted value at a positive readout / market capitalisation), scored 80 + 40*log10(R), clamped to [0, 100]. High because the market cap equals cash.
Probability of a positive outcomeThis record’s own outcome_prediction.probability_pct. Computed.45outcome_prediction.probability_pct = 45
Date confidencereadout.precision and readout.confidence. Computed. It is the gate the other six drivers depend on (01-rules.md rule 39).100readout.precision=DAY, readout.confidence=HIGH
Squeeze mechanicsFloat, short interest as a percentage of float, and average daily dollar volume. Computed.64float 157636355 shares, short_float_pct 25.6, average dollar volume 6226804. Float is the filed count less the 12.467% insider holding; short interest is FINRA’s 2026-09-30 count; dollar volume is the cache’s 20-session average volume at spot.
Priced-in-ness52-week position, price drift since the last catalyst, ownership crowding, and the spread between analyst targets. Computed. A HIGH score means room LEFT to run.99price 0.86 sits at 1% of its 52-week range (low 0.822, high 6.88, as of 2026-10-16) — closer to the 52-week low — room left to run. Reads in the opposite sense to its name: 99 means nothing is priced in. Only the 52-week leg is computed by the module.
Financing and clustering riskCash runway against the catalyst date, plus attribution.status and attribution.conflicts. Computed. The one NEGATIVE driver.100attribution.status=CONTAMINATED is the larger of the two independent risks (clustering). Reads in the opposite direction from the other five: 100 is the worst score and lowers the total. Runway alone (OK) would score 10.

Priority score. From priorityScore in lib/runup.mjs.

Priority score 56 · formula_version 2.0.0

The 5.10.3 record scored 49 under formula 1.0.0. The two are not comparable: 2.0.0 computes value uplift from the commercial ceiling, which this low market cap inflates.

Settlement. Null at lock. Settled only on the user’s explicit request, from a confirmed primary source, against the committed price cache (05-prediction-protocol.md § Run-up settlement).

Verdict

What I would do. Watch. Hold no position through the talk on these numbers.

Why. The shares trade at cash, so the downside from a disappointing duration is limited, and the upside from a strong one is real. The scenarios weigh those two almost exactly at today’s price: the expected value is −0.5% against spot, and our probability band sits inside a very wide break-even range. The FDA has already said what matters most for the asset. Patients did not live longer, and the crossover check made it worse. A duration figure, however good, does not answer that.

What would change this. A median duration of response of 9 months or more, with a discussant who backs a filing, would argue for the positive row’s high edge. An FDA statement that it will accept a filing without a survival trial would change the asset verdict. A duration below 4.7 months, the arm’s own median progression-free survival, would argue for the miss row’s low edge.

What to watch.

  • 2026-10-19 to 2026-10-22: any online posting of abstract 459O before the talk.
  • 2026-10-24, 10:15–11:45 central European time: the talk by Nilofer Azad, and the discussant’s verdict. The first independent named view of this dataset (A.2b).
  • 2026-10-26: the first US trading session after the talk.
  • About 2026-11-05: third-quarter results. Watch whether the runway “into 2028” is restated after the FDA feedback, and whether a survival trial is announced and funded.
  • By 2026-11-14: third-quarter 13Fs, showing whether Tang, OrbiMed and Vivo held through 2026-09-22.
  • From about 2026-11-19: the earliest date for a Nasdaq minimum-bid notice, if the price stays below $1.
  • Q4 2026: CTX-8371 and CTX-10726 Phase 1 data, the two conflicts in Attribution.
  • Q1 2027: the BLA target reported by secondary sources, not yet confirmed by the company.

Locked prediction

  • Outcome-direction (will the readout succeed on its primary endpoint?): uncertain — probability 45%, band 32–58% [UNVERIFIED — modelled]. The definition below has three conditions. Two are close to settled: the final response rate is 18.0% [4], and the safety dataset was disclosed on 2026-04-27 [12]. The open one is a median duration of at least 5.0 months. The arm’s median progression-free survival is 4.7 months, and responders usually last longer than the arm as a whole, so 5.0 months is roughly an even bet. Two facts lean against it. The FDA saw the full data, duration included, and still asked for a survival trial. And the company’s 2026-09-22 rebuttal cited response rate, progression-free survival and survival subsets, but not duration [4]. That is why the figure moved down from 55% on 2026-08-19. No third party has published a probability on this event.
  • Base rate 47.7% (to the next phase, BIO 2021 Figure 2, p. 7, oncology) — distance −2.7 pp. The base rate counts oncology Phase 3 programs that reached a filing. This call is narrower: a duration threshold in a presentation of locked data. The figures sit close by coincidence, not by derivation. outcome_prediction.base_rate equals asset_evaluation.pos.base_rate.
  • Stock-direction (which way do the shares move?): no-edge — confidence none — window 2026-10-19 to 2026-10-30. Basis: it opens with the first session of the congress week, covers the talk and the first trading session after it (2026-10-26), and closes before the expected third-quarter results. Materiality is dominant per ../company.md C.2. Attribution is CONTAMINATED: CTX-8371 and CTX-10726 carry “Q4 2026” windows that cover this day, so a move in this window is not attributable to this program alone.
  • Break-even probability 46.4%, range 6.5–96.0% — clearance none — gap evidence: 0 items.
  • Cash floor: $0.86 per share.
  • Scenario prices: positive $0.87–$1.15 · miss $0.62–$0.84.
  • Commercial ceiling: $4.80 — positive high allowed up to $6.77.
  • Expected value: $0.856, −0.5% against spot $0.86 (arithmetic, not advice).
  • Run-up: entry $0.86 on 2026-10-17, exit rule T-2 trading days before readout.window.earliest — predicted move −5% to +15%, predicted peak +3% to +30% around 2026-10 — priority score 56, formula_version 2.0.0.
  • Settles on the ESMO Congress 2026 presentation of abstract 459O (scheduled 2026-10-24), its abstract, and any same-day Compass release or 8-K. Positive definition: all three hold. (1) A confirmed overall response rate for tovecimig plus paclitaxel of at least 15.0% in the primary-endpoint analysis, by independent central review where the presentation separates it from investigator assessment. (2) A median duration of response for that arm of at least 5.0 months. (3) No new grade 3 or higher treatment-emergent adverse-event category at a rate above the 44% grade 3+ high blood pressure reported on 2026-04-27. If no median duration for that arm is reported by 2026-10-27, the outcome is a miss. A duration “not reached” counts as at least 5.0 months only if the responders’ minimum follow-up is stated as at least 5.0 months.
  • Locked: yes · Settled: no

Program data-quality flags

  1. The 5.10.3 positive anchor did not reproduce. The “+49.73% intraday” move on 2025-04-01 was the 08-19 version’s upper bound for tovecimig news. The price cache gives +1.0% over the protocol’s event window and +17.9% close to close (../company.md C.7 and C.8 row 3). This version’s anchors use the cache.
  2. The 5.10.3 positive anchors are inadmissible under 6.2.0. That record anchored its positive row on the 52-week high and on analyst targets, and its miss row on the 52-week low. Rule 30 now bars all three. The old record is history and is not edited.
  3. The registry’s primary completion date moved by nine months between sweeps, from 2026-12-01 to 2027-08-31, with no stated reason [3]. Recorded, not interpreted.
  4. CT.gov carries no investigator names for NCT05506943. search_investigators again returned zero rows. A.2b is built from the design paper’s authors, matched to trial sites.
  5. PubMed’s index mixes tovecimig’s code ABL001 with Novartis’s asciminib. A query including “ABL001” returned 389 hits; the real footprint is 4 papers. Kept as a warning for the next sweep.
  6. No conflict-of-interest statement is available for any named investigator. “No competitor relationship found” means the search had no disclosure section to read. One competitor tie was found from bylines instead (Goyal, futibatinib).
  7. No peer-reviewed clinical result exists. Every efficacy and safety figure traces to a company release or an 8-K. Kept from 5.10.3.
  8. The ESMO 2026 embargo policy could not be read. The policy page returned 404 and the abstracts page carries no text. The window’s earliest edge rests on the congress opening, and the 5.10.3 claim about an opening-day abstract release is withdrawn.
  9. Named correction to the PubMed reading. The 5.10.3 version said two of the four papers had ABL Bio employees as a majority of authors. That holds for PMID 33383646 (10 of 12). PMID 32580836 is Yonsei-led, with 6 of 13 authors from ABL Bio [34].
  10. The ABC-06 FOLFOX response rate (about 5%) and the 70–80% unselected share come from the 2026-08-19 sweep’s secondary sources, whose URLs it did not record. They are tagged as web estimates and drive no scored figure.
  11. The final response rate changed between sources: 17.1% (19/111, p=0.031) on 2026-04-27 [12], 18.0% (20/111, p=0.0228) on 2026-09-22 [4]. This version uses the later figure.

Sources

[1] Compass Therapeutics, Form 10-Q for the quarter ended 2026-06-30, filed 2026-08-06 — https://www.sec.gov/Archives/edgar/data/1738021/000117184326005295/cmpx20260630_10q.htm [2] Compass Therapeutics, Form 10-K for 2025, filed 2026-03-05 — https://www.sec.gov/Archives/edgar/data/1738021/000117184326001337/cmpx20251231_10k.htm [3] ClinicalTrials.gov, NCT05506943 (COMPANION-002), read 2026-10-17 — https://clinicaltrials.gov/study/NCT05506943 [4] Compass Therapeutics, Form 8-K exhibit 99.1, “Compass Therapeutics Provides Regulatory Update on Tovecimig in Biliary Tract Cancer Following FDA Feedback”, 2026-09-22 — https://www.sec.gov/Archives/edgar/data/1738021/000117184326006162/exh_991.htm [5] Lamarca A et al., “Second-line FOLFOX chemotherapy versus active symptom control for advanced biliary tract cancer (ABC-06)”, Lancet Oncol 2021 — https://doi.org/10.1016/S1470-2045(21)00027-9 [6] OncoDaily, “9 Hepatobiliary Cancer Trials to Watch at ESMO 2026”, 2026-10-10 (secondary) — https://oncodaily.com/gastrointestinal-oncology/esmo-2026-2 [7] Azad N et al., “COMPANION-002 A clinical trial of investigational drug CTX-009 plus paclitaxel vs paclitaxel in second line advanced BTC”, Future Oncol 2024;20(30):2241-2248 — https://doi.org/10.1080/14796694.2024.2351351 [8] Compass Therapeutics, press release, “Tovecimig Data Accepted for an Oral Presentation at the ESMO Congress 2026”, 2026-07-17 — https://www.globenewswire.com/news-release/2026/07/17/3329171/0/en/compass-therapeutics-announces-tovecimig-data-accepted-for-an-oral-presentation-at-the-esmo-congress-2026.html [9] Stocktwits, “Compass Therapeutics FDA setback raises concerns on Wall Street; H.C. Wainwright calls tovecimig filing strategy risky”, 2026-09-23 (secondary; the page returned 403 to a direct fetch, read through search summaries) — https://stocktwits.com/news-articles/markets/equity/compass-therapeutics-fda-setback-raises-concerns-on-wall-street-h-c-wainwright-calls-tovecimig-filing-strategy-risky/cZMEwFSRBJl [10] US FDA, draft guidance “Approaches to Assessment of Overall Survival in Oncology Clinical Trials”, August 2025 — https://www.fda.gov/regulatory-information/search-fda-guidance-documents/approaches-assessment-overall-survival-oncology-clinical-trials [11] CancerNetwork, “FDA Grants Orphan Drug Designation to Tovecimig for Biliary Tract Cancer”, 2026-04-08 (secondary) — https://www.cancernetwork.com/view/fda-grants-orphan-drug-designation-to-tovecimig-for-biliary-tract-cancer [12] Compass Therapeutics, press release, COMPANION-002 secondary endpoints, 2026-04-27 — https://www.globenewswire.com/news-release/2026/04/27/3281512/0/en/tovecimig-demonstrates-statistically-significant-benefit-in-companion-002-randomized-phase-2-3-study-in-patients-with-biliary-tract-cancer.html [13] OncLive, “FDA Grants Fast Track Designation to CTX-009 Plus Paclitaxel for Pretreated Biliary Tract Cancer”, 2024-04-25 (secondary) — https://www.onclive.com/view/fda-grants-fast-track-designation-to-ctx-009-plus-paclitaxel-for-pretreated-biliary-tract-cancer [14] Na JY et al., “Population pharmacokinetic model of ABL001/CTX-009 (anti-VEGF/DLL4) in adult cancer patients with solid tumor”, Cancer Sci 2024;115(12):3943-3951 — https://doi.org/10.1111/cas.16363 [15] DailyMed, ZIIHERA (zanidatamab-hrii) label, published 2026-08-27 — https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=ae5d9425-fae5-4541-a158-150998343348 [16] DailyMed, PEMAZYRE (pemigatinib) label, published 2026-06-15 — https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=9e1f2222-1d89-4e63-989c-ccebe2ab1eb4 [17] DailyMed, LYTGOBI (futibatinib) label, published 2026-06-02 — https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=0b1332a1-0581-4707-9bf6-1eccfa39bef4 [18] Open Targets Platform, VEGFA and cholangiocarcinoma evidence (DLL4 has no association), queried 2026-10-17 — https://platform.opentargets.org/evidence/ENSG00000112715/MONDO_0019087 [19] ChEMBL, CHEMBL6068308 (tovecimig) — https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL6068308/ [20] Hyung J et al., “Treatment With Liposomal Irinotecan Plus Fluorouracil and Leucovorin for Patients With Previously Treated Metastatic Biliary Tract Cancer: The Phase 2b NIFTY Randomized Clinical Trial”, JAMA Oncol 2023 — https://doi.org/10.1001/jamaoncol.2023.0016 [21] Harding JJ et al., “Zanidatamab for HER2-amplified, unresectable, locally advanced or metastatic biliary tract cancer (HERIZON-BTC-01)”, Lancet Oncol 2023 — https://doi.org/10.1016/S1470-2045(23)00242-5 [22] Abou-Alfa GK et al., “Pemigatinib for previously treated, locally advanced or metastatic cholangiocarcinoma (FIGHT-202)”, Lancet Oncol 2020 — https://doi.org/10.1016/S1470-2045(20)30109-1 [23] Goyal L et al., “Futibatinib for FGFR2-Rearranged Intrahepatic Cholangiocarcinoma (FOENIX-CCA2)”, N Engl J Med 2023 — https://doi.org/10.1056/NEJMoa2206834 [24] Abou-Alfa GK et al., “Ivosidenib in IDH1-mutant, chemotherapy-refractory cholangiocarcinoma (ClarIDHy)”, Lancet Oncol 2020 — https://doi.org/10.1016/S1470-2045(20)30157-1 [25] FDA, Purple Book (searched 2026-10-17; tovecimig not listed) — https://purplebooksearch.fda.gov [26] DailyMed, TIBSOVO (ivosidenib) label, published 2025-12-22 — https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=65d254c0-67ad-42c4-b972-ad463b755b2d [27] DailyMed, IMFINZI (durvalumab) label, published 2026-09-07 — https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=8baba4ea-2855-42fa-9bd9-5a7548d4cec3 [28] ICER, assessments (searched 2026-10-17; no biliary tract cancer assessment found) — https://icer.org/assessments/ [29] Milliman, “Key takeaways: third Medicare drug price negotiation” (secondary) — https://www.milliman.com/en/insight/key-takeaways-third-medicare-drug-price-negotiation [30] CMS, Medicare Drug Price Negotiation — https://www.cms.gov/inflation-reduction-act-and-medicare/medicare-drug-price-negotiation [31] openFDA, complete response letter data API (queried 2026-10-17) — https://api.fda.gov/transparency/crl.json [32] openFDA, drug shortages API (queried 2026-10-17) — https://api.fda.gov/drug/shortages.json [33] Yeom DH et al., “ABL001, a Bispecific Antibody Targeting VEGF and DLL4, with Chemotherapy, Synergistically Inhibits Tumor Progression in Xenograft Models”, Int J Mol Sci 2020;22(1):241 — https://doi.org/10.3390/ijms22010241 [34] Kim DH et al., “Synergistic antitumor activity of a DLL4/VEGF bispecific therapeutic antibody in combination with irinotecan in gastric cancer”, BMB Rep 2020;53(10):533-538 — https://doi.org/10.5483/BMBRep.2020.53.10.103 [35] BioSpace, “Compass vows to press on with FDA application despite new trial request”, 2026-09-22 (secondary) — https://www.biospace.com/drug-development/compass-vows-to-press-on-with-fda-application-despite-new-trial-request [36] Compass Therapeutics, second-quarter 2026 results, 8-K exhibit 99.1, 2026-08-06 — https://www.sec.gov/Archives/edgar/data/0001738021/000117184326005292/exh_991.htm [37] NICE, TA948, ivosidenib for previously treated advanced cholangiocarcinoma with an IDH1 R132 mutation — https://www.nice.org.uk/guidance/ta948 [38] William Blair, Compass Therapeutics initiation (secondary; figure read from search summaries) — https://www.williamblair.com/News/Compass-Therapeutics-Inc-Initiation [39] ESMO, ESMO Congress 2026 (via data/congresses.json) — https://www.esmo.org/meeting-calendar/esmo-congress-2026

Sources

    BPIQ fetch_company_drugs, CMPX, 2026-10-17 CT.gov search_trials and get_trial_details NCT05506943, search_investigators, 2026-10-17 PubMed search_articles (4 real hits) and get_article_metadata on PMIDs 39375952, 38861293, 33383646, 32580836; competitor papers PMIDs 33798493, 36951834, 37276871, 32203698, 36652354, 32416072 - 2026-10-17 Open Targets search_entities and association query (DLL4, VEGFA x MONDO_0019087, MONDO_0003060), 2026-10-17 ChEMBL compound_search tovecimig (CHEMBL6068308), 2026-10-17 DailyMed API labels: Ziihera, Pemazyre, Lytgobi, Tibsovo, Imfinzi, 2026-10-17 openFDA CRL and drug shortages APIs, 2026-10-17 EDGAR: 8-K exhibit 99.1 2026-09-22; Form 10-Q filed 2026-08-06; Form 10-K for 2025 FDA draft guidance, Approaches to Assessment of Overall Survival in Oncology Clinical Trials, August 2025 Web searches 2026-10-17: peak sales, competitive, exclusivity and royalty, analyst, ICER, CMS, Purple Book, Orange Book, CMC (P5), warning letters and inspections data/congresses.json, ESMO Congress 2026 row