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

Program analysis · bpiq_drug_id 17398 · prepared 2026-08 · USD · framework v5.10.3 · Coverage: CLEARED

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

How to read this. Written for a reader who has not studied pharmacology or finance. Terms specific to this program are in the Glossary below; general terms are defined once in framework/04-glossary.md. Tags: [VERIFIED — source] / [UNVERIFIED] / [WEB ESTIMATE — source, date].

Glossary

Executive summary

• What it is (one sentence): Tovecimig is a two-armed antibody that blocks both of the signals a bile-duct tumour uses to grow its own blood supply, given on top of ordinary paclitaxel chemotherapy to patients whose cancer has already returned after first-line treatment.
• The event and when (as disclosed): An oral presentation of the complete COMPANION-002 dataset at the ESMO Congress in Madrid on October 24, 2026, presented by Dr. Nilofer Azad of Johns Hopkins [VERIFIED — BPIQ fetch_company_drugs 2026-08-19, catalyst_date_text "October 24, 2026"; Compass press release 2026-07-17]. This is a date disclosed to a day, which is rare in this corpus and is what makes the timing calls below unusually well-anchored.
• The main reason it could work: Second-line biliary tract cancer has no approved drug for the four-fifths of patients without a targetable mutation, and the trial already met its primary endpoint: 17.1% of patients on tovecimig plus paclitaxel had their tumours shrink, against 5.3% on paclitaxel alone (p=0.031), with progression delayed from 2.6 to 4.7 months (hazard ratio 0.44, p<0.0001) [VERIFIED — Compass press release 2026-04-27].
• The main risk: Patients did not live longer. Median overall survival was 8.9 months on tovecimig against 9.4 months on paclitaxel alone (hazard ratio 1.05, p=0.78), and the crossover-adjusted analysis that was supposed to rescue that number made it worse, not better (hazard ratio 1.13) [VERIFIED — Compass press release 2026-04-27]. Alongside that, 44% of patients had severe (Grade 3 or higher) high blood pressure. The shares fell about 65% in one session on that readout.
• What it means for the stock: The October event presents a dataset whose headline numbers are already public. The one genuinely unknown figure is how long responses lasted. That is a real but bounded catalyst, and it lands in the same quarter as two other company readouts, so a move around it is not cleanly attributable to this program. The expected value below is essentially flat against spot, and the stock-direction call is no-edge accordingly. The run-up call is separate and is the one with an edge.

0. Program-tier coverage — CLEARED

Every mandatory program-tier row was called. Company-tier coverage is in ../company.md C.0. The regulatory tier was not called, correctly: this catalyst is a congress data presentation, not a regulatory decision, and no marketing application has been submitted yet.

A. Scientific & clinical assessment

A.1 Mechanism of action and scientific rationale

Tovecimig is a single antibody molecule with two jobs. Its main body binds VEGF-A, the protein a tumour releases to make blood vessels sprout toward it. Attached to the ends of that body are two small fragments (single-chain variable fragments) that bind DLL4, a protein sitting on the surface of the vessel cells inside the tumour [VERIFIED — COMPANION-002 design paper, Azad et al., Future Oncol 2024, PMID 38861293, DOI 10.1080/14796694.2024.2351351, retrieved from PubMed 2026-08-19].

The reason to block both, rather than VEGF-A alone, is a resistance story. Blocking VEGF-A starves the tumour of its sprouting signal, and tumours respond by switching on DLL4-Notch signalling instead, which is a different route to the same result. Preclinical work reports that DLL4-Notch signalling is part of how tumours become resistant to anti-VEGF drugs, and that blocking both pathways together shrinks tumours more than blocking either alone in mouse models [VERIFIED — Yeom et al., Int J Mol Sci 2020, PMID 33383646, DOI 10.3390/ijms22010241, retrieved from PubMed 2026-08-19]. Combining that blockade with chemotherapy adds a second effect: when the disordered tumour vessels regress and the survivors normalise, chemotherapy reaches more of the tumour [VERIFIED — same paper].

How well is the target validated? Unevenly, and this is the honest weak point. DLL4 is independently confirmed as a real human gene and drug target [VERIFIED — Open Targets search_entities 2026-08-19, ENSG00000128917], as is VEGFA [VERIFIED — ENSG00000112715], and cholangiocarcinoma is an established disease entity in the same database [VERIFIED — MONDO_0019087]. But target existence is not target validation for this disease. The claim that DLL4 is highly expressed in biliary tract cancer and that its overexpression predicts poor survival comes from the sponsor-authored design paper citing primary literature this sweep did not read [UNVERIFIED — cited but not independently checked]. More telling: no anti-DLL4 agent has ever been approved for anything, and the class has a history of dose-limiting cardiovascular and vascular toxicity. The 44% Grade 3-or-higher hypertension rate in COMPANION-002 is that class effect showing up on schedule.

flowchart LR
  A["Biliary tract tumour<br/>needs new blood vessels<br/>to keep growing"] --> B["Tumour secretes VEGF-A,<br/>driving vessel sprouting"]
  B --> C["Vessels grow, tumour is fed<br/>chemotherapy reaches it poorly"]
  D["Anti-VEGF drugs alone<br/>block sprouting"] --> E["Tumour switches on<br/>DLL4-Notch signalling<br/>= escape route, resistance"]
  F["Tovecimig CTX-009<br/>one antibody, two arms"] --> G["Arm 1 binds VEGF-A<br/>blocks the sprouting signal"]
  F --> H["Arm 2 binds DLL4<br/>closes the escape route"]
  G --> I["Tumour vessels regress<br/>and normalise"]
  H --> I
  I --> J["Paclitaxel reaches<br/>more tumour cells"]
  J --> K["Intended benefit:<br/>more tumours shrink<br/>= higher response rate"]
  C -.-> I
  E -.-> H

The exact scientific step the next readout must prove. Not that the drug shrinks tumours — that is already established and statistically significant. The October presentation must show that the shrinkage lasted. Duration of response is the one pre-specified secondary endpoint from COMPANION-002 that has never been reported publicly, and the company said on 2026-04-27 that the complete dataset with duration of response would be presented later in 2026 [VERIFIED — BPIQ fetch_company_drugs 2026-08-19, note field entry dated 04/27/26].

The honest scientific risk. The mechanism produced tumour shrinkage and delayed progression, and produced no survival benefit at all. That combination — a real anti-tumour effect that does not translate into patients living longer — is the classic signature of an agent whose toxicity costs back what its efficacy earns. The company attributes the flat survival to 54% crossover. Its own crossover-adjusted analysis does not support that attribution (A.5, Evidence quality), and that unresolved contradiction is the single most important scientific fact in this document.

A.2 Clinical development plan, timeline, feasibility, resourcing

gantt
  title Tovecimig (CTX-009) clinical development, with competitor precedent
  dateFormat YYYY-MM-DD
  axisFormat %Y-%m
  section Tovecimig - Compass / ABL Bio
    NCT03292783 Ph1 monotherapy dose escalation n=45      :done, 2017-09-01, 2021-06-30
    NCT04492033 Ph1b/2a + paclitaxel or irinotecan n=41   :done, 2020-06-22, 2024-01-08
    NCT05506943 COMPANION-002 Ph2/3 randomised n=168      :active, 2023-01-09, 2026-12-01
    ORR primary endpoint met, reported                    :milestone, 2025-04-01, 0d
    PFS met, OS missed, reported                          :milestone, 2026-04-27, 0d
    ESMO oral, full dataset with duration of response     :milestone, 2026-10-24, 0d
    Planned BLA submission, guided 2026                   :crit, 2026-11-01, 2026-12-31
    NCT05513742 COMPANION-003 Ph2 colorectal n=49         :done, 2022-12-08, 2025-05-15
    NCT06548412 Ph1/2 first-line BTC, MD Anderson n=50    :active, 2025-01-22, 2027-05-01
  section Competitor precedent
    ABC-06 Ph3 FOLFOX vs symptom control, 2L BTC          :done, 2014-03-01, 2019-06-01
    TOPAZ-1 Ph3 durvalumab + gem/cis, 1L BTC              :done, 2019-04-01, 2021-08-01
    HERIZON-BTC-01 zanidatamab, HER2+ BTC, accelerated approval :done, 2019-09-01, 2024-11-20

Milestone dates. Last patient in is not separately disclosed; enrollment completion was announced on 2024-08-06 [VERIFIED — BPIQ fetch_company_drugs 2026-08-19, note entry dated 8/6/24]. The primary-endpoint database lock produced a readout on 2025-04-01, roughly eight months after enrollment completed [VERIFIED — Compass press release 2025-04-01]. The secondary-endpoint (progression-free and overall survival) lock produced a readout on 2026-04-27 [VERIFIED — Compass press release 2026-04-27]. The registry’s primary_completion_date is 2026-12-01 [VERIFIED — CT.gov get_trial_details NCT05506943, retrieved 2026-08-19], which now postdates the October presentation — see the Readout section for what that disagreement means and does not mean.

Trials

Feasibility matrix

Resourcing sufficiency. Yes for this readout, and it is not really a question: the trial is enrolled, both analyses are locked, and the presentation is written. The company holds $179.9M of cash and marketable securities against a $4.9M monthly burn (../company.md C.3), so there is no scenario where the October presentation fails to happen for lack of money. The real resourcing question is one step later — whether a company with 18 to 36 months of clinical-rate cash can fund a marketing application, a launch and an ongoing first-line programme without returning to the $400M shelf that has sat effective and unused since January (../company.md C.3). That question does not bear on this catalyst; it bears on what happens after it.

A.3 Target label, target product profile, strategic questions, designations

A.3a Target label. Tovecimig in combination with paclitaxel for adults with unresectable advanced, metastatic or recurrent biliary tract cancer who have progressed after one prior gemcitabine- and platinum-containing regimen.

A.3b Target product profile

A.3c Strategic Go/No-Go questions. The set below is Pre-Phase-III / registration, because COMPANION-002 is complete and the next decision is whether to file.

A.3d Regulatory designations.

• Fast Track designation, granted 2024-04-25. Grants more frequent FDA meetings, eligibility for accelerated approval and priority review if criteria are met, and rolling review of a marketing application. It is a process benefit, not a lower evidentiary bar. [VERIFIED — BPIQ fetch_company_drugs 2026-08-19, note entry dated 4/25/24]
• Orphan Drug designation for biliary tract cancer, granted around 2026-05-05. Grants seven years of US market exclusivity for the designated indication from approval, tax credits on qualifying trial costs, and a waiver of the marketing-application user fee. The seven years run independently of any patent. [VERIFIED — BPIQ fetch_company_drugs 2026-08-19, note entry dated 05/05/26; corroborated by CancerNetwork and Targeted Oncology coverage retrieved 2026-08-19]
• No Breakthrough Therapy designation has been announced, and none appears in the BPIQ note history or the press feed. In a disease this poorly served, a drug with a real survival benefit would normally have one. Its absence is consistent with the flat overall survival. [UNVERIFIED — absence of evidence; not evidence of refusal]

A.4 Target product profile valuation matrix

Calibration examples, not asset claims: in oncology each incremental month of overall survival is worth roughly $150–300M in peak-sales potential; oral formulations command roughly 15–25% price premiums over injectables. Applied here, the calibration is unflattering and worth stating plainly: the incremental overall survival is zero months, so that particular lever contributes nothing to this asset’s peak-sales potential.

A.5 Evidence quality and endpoints

The single most important evidence finding in this document. The COMPANION-002 design paper states the trial “was designed to detect a 23% absolute difference in response rate between the two arms, with 90% power and a two-sided alpha of 0.05” [VERIFIED — PMID 38861293, § 3.5 Statistical methods]. The delivered difference was 11.8 percentage points (17.1% minus 5.3%) — roughly half of what the trial was powered to find. It still cleared significance at p=0.031, because the control arm underperformed the design’s assumption too, but “significant” and “as large as expected” are different facts and only the first one was in the press release.

framework/07-benchmarks.md’s Surprise, early stops and hype section is explicit that share-price moves scale with the expectation gap rather than with the result alone, and that the dramatic moves live in the outer buckets. This asset has now landed in the missed expectations bucket twice — once on effect size in April 2025, and once on overall survival in April 2026 — and the second one produced a 65% single-session decline. That history, not the class average, is what the scenario ranges below are calibrated against.

And the crossover argument does not survive its own check. The company attributes the flat survival to 54% of control patients crossing over to tovecimig. The standard way to test that claim is a rank-preserving structural failure time analysis, which estimates what the control arm’s survival would have been without crossover. Compass reported it: hazard ratio 1.13, against the unadjusted 1.05 [VERIFIED — Compass press release 2026-04-27]. Removing the crossover made the result worse. A crossover explanation that its own adjustment contradicts is not an explanation. A supportive subset (crossover patients 12.8 months vs non-crossover 6.1 months, hazard ratio 0.54, p=0.04) was also reported, but comparing patients who crossed over against patients who did not is comparing patients who lived long enough to cross over against those who did not — a well-known form of survivorship bias, not evidence.

Endpoints

A.5b Key opinion leaders.

Panel as of. 2026-08-19.

Investigators

CT.gov search_investigators returned zero rows for this trial on two different query shapes (condition: "biliary tract cancer" with investigator_name: "Javle", and with institution: "Compass Therapeutics"), and get_trial_details on NCT05506943 returned no overall-official or contact block — the registry record simply does not carry investigator names. The investigators below are therefore taken from the COMPANION-002 design paper’s author list, every one of whom is affiliated to an institution that appears in the trial’s own CT.gov site list, which is what ties them to this trial rather than to the disease generally. Five of the twenty-five are listed; they are the ones with a named role in this catalyst or a leading position in the disease.

A limit worth stating rather than burying. 02-connectors.md § KOL sources is explicit that a conflict statement is only as good as the journal’s disclosure practice, and that an absent disclosure is not evidence of no conflict. Here there is no disclosure section at all in either PubMed render, so the “no competitor relationship found” entries above rest on a search that had nothing to search. They record that the attempt was made, and nothing more.

Independent voices

None were found, and the reason is structural. An independent voice on this endpoint would be a named commentator on tovecimig’s data who is not an investigator on its trial and holds no disclosed relationship with Compass. The PubMed footprint for this molecule is four papers, of which three are ABL Bio’s own and one is the sponsor’s design paper — there is no independent literature to draw a voice from. COMPANION-002’s own 34 sites enrol most of the US academic centres that specialise in this disease, so the disease’s leading clinicians are largely investigators on it. No independent named critique of the April 2026 dataset was located in the press feed or by web search; the critical coverage that exists (Fierce Biotech, Benzinga) is unattributed reporting rather than a named expert view.

This empty table is itself a finding, and it is the reason the judgement below reads UNKNOWN rather than anything warmer. It will not stay empty: an ESMO Proffered Paper carries a formal expert discussant who critiques the data from the podium immediately after the presentation. That discussant is the first genuinely independent voice this programme will have had, and their verdict on 2026-10-24 is a specific, dated, checkable item — it is on the “What to watch” list below.

Judgement

Dissent

Empty, and correctly so: endpoint_supported is UNKNOWN rather than SUPPORTIVE_CONTESTED, so there is no claimed support for anyone to contest. Inventing a dissenter would be as wrong as inventing a supporter.

B. Commercial assessment

B.0 Current treatment algorithm

First line. Gemcitabine plus cisplatin chemotherapy, now routinely with an immunotherapy antibody added (durvalumab or pembrolizumab) following the TOPAZ-1 and KEYNOTE-966 trials. This is what essentially every patient with advanced biliary tract cancer receives [WEB ESTIMATE — Frontiers in Oncology 2026 review, retrieved 2026-08-19].

On progression, the path forks by molecular test result.

• Roughly 10–15% of intrahepatic cholangiocarcinomas carry an FGFR2 fusion → pemigatinib (Pemazyre) or futibatinib (Lytgobi), both accelerated approvals.
• Roughly 10–15% of intrahepatic cholangiocarcinomas carry an IDH1 mutation → ivosidenib (Tibsovo), full approval.
• A minority are HER2 immunohistochemistry 3+ → zanidatamab (Ziihera), accelerated approval November 2024, or tucatinib plus trastuzumab.
• Small fractions carry BRAF V600E (4–5%), NTRK fusions (<1%), RET fusions (<1%) or microsatellite instability (1–2%) → the corresponding tumour-agnostic approvals.
• Everyone else — 70–80% of the second-line population — gets FOLFOX, off-label, on the strength of ABC-06’s 0.9-month median survival gain over symptom control, at a response rate of about 5%. [WEB ESTIMATE — Frontiers in Oncology 2026 review and DelveInsight, retrieved 2026-08-19; ABC-06 figures VERIFIED — PMID 38861293]

Where tovecimig would fit. Exactly into that last bullet, and nowhere else. It would be a second-line regimen for the unselected majority, replacing or competing with FOLFOX. COMPANION-002’s own eligibility criteria enforce this position: patients eligible for an approved targeted therapy on a labelled regimen are excluded [VERIFIED — CT.gov get_trial_details NCT05506943, exclusion criterion 1]. It adds to paclitaxel rather than displacing chemotherapy — the regimen is tovecimig plus paclitaxel, so it is an add-on, not a replacement.

B.1 Competitive landscape and positioning

Where this asset wins, and the single fact the thesis rests on. It wins on being the only randomised, statistically positive, non-biomarker-gated asset aimed at the largest unserved slice of a disease whose incumbent second-line therapy produces a 5% response rate. That is a real and defensible position.

The single fact the thesis rests on is that the US Food and Drug Administration will accept response rate and progression-free survival as sufficient, in a disease this poorly served, without an overall-survival benefit. Everything else — the market size, the price, the penetration, the share price — follows from that one judgement. It is not a scientific question and this document cannot resolve it. The company expected FDA feedback on exactly this in the third quarter of 2026 [VERIFIED — Compass second-quarter release, 2026-08-06], which means the answer is likely to arrive before the October catalyst this analysis is written about.

Calibration example, not an asset claim: oncology first-movers in novel mechanisms have shown roughly 3.2× higher peak-sales potential but about 1.8× higher development risk. Tovecimig is a first-mover in a mechanism that has never produced an approval, so the second half of that calibration applies at least as strongly as the first.

B.2 Addressable market

Launch markets. The United States first and, on the base case, only. Compass holds worldwide rights except Korea and China but has no commercial infrastructure outside the US, and its Chief Commercial Officer was appointed in January 2026. Any ex-US revenue requires a partner that does not exist yet, so ex-US is treated as upside in B.3a rather than as base-case revenue.

B.3 Value and feasibility

B.3a Expected peak sales. Built bottom-up. No published analyst peak-sales estimate for tovecimig in second-line biliary tract cancer was found in this sweep’s peak-sales web search, so nothing here is a pass-through of a third-party figure, and nothing here should be read as one. Every figure is conditional on approval. All three scenarios are US-only and gross of the tiered single-digit royalty owed to ABL Bio.

The patient funnel, applied to both ends of the disputed incidence figure:

1. US biliary tract cancer diagnoses per year: 12,130 to 18,600 [VERIFIED — 18,600 quoted from the American Cancer Society in PMID 38861293] / [WEB ESTIMATE — 12,130, SEER-based, retrieved 2026-08-19]
2. × ~68% presenting with or progressing to advanced/unresectable disease [UNVERIFIED — assumption]
3. × ~70% who receive first-line gemcitabine/platinum therapy [UNVERIFIED — assumption]
4. × ~45% who progress and remain fit enough (performance status 0–1) for second-line treatment [UNVERIFIED — assumption, consistent with real-world second-line initiation rates in this disease]
5. × ~80% without an actionable alteration that routes them to an approved targeted agent — the complement of the 20–30% covered by FGFR2, IDH1, HER2 and MSI-high agents [WEB ESTIMATE — Frontiers in Oncology 2026 review, retrieved 2026-08-19]

= 2,080 to 3,190 addressable US patients a year.

Price: no US price exists. Annual cost per treated patient is modelled as $20,000–$30,000 a month of net revenue (the range branded orphan oncology biologics commonly occupy) × 5 months of average treatment duration, anchored on the tovecimig arm’s own 4.7-month median progression-free survival = $100,000–$150,000 per treated patient [UNVERIFIED — assumption].

The weakest input is the price, and the second weakest is penetration. Neither has any verified anchor: no tovecimig price exists, and no analogous all-comers add-on has ever launched in this disease. Patient count is the strongest input, because the funnel’s first step has two independent published sources and its last step has a cited literature range. A reader who disagrees with the price should scale the whole table linearly — it moves the answer proportionally.

These figures sit awkwardly against the published analyst targets, and the gap is worth naming. The base case, ~$146M of peak annual revenue, is below the company’s current ~$267M enterprise value (../company.md C.4). The three post-collapse analyst targets — Stifel $6 (2026-08-14), Piper Sandler $6 (2026-08-07), Leerink $7 (2026-08-07) — imply an enterprise value near $900M at $6 a share, which is roughly six times this base case’s peak revenue and cannot be reached from second-line biliary tract cancer alone. Either those targets carry substantial value for first-line expansion, ex-US rights and the three Phase 1 assets, or this bottom-up build is too conservative. Both are possible and this document cannot settle it; the disagreement is stated rather than smoothed over.

B.3b Feasibility

B.4 Product-development risk

Framing questions.

1. Does the development plan support the target product profile claims? Partly. The plan supports the efficacy claims it makes — response rate and progression-free survival are both demonstrated in a randomised trial with independent central review of the primary endpoint. It does not support the survival claim, because the design permitted crossover and thereby gave that claim away before the first patient was enrolled. A trial cannot demonstrate what its own design makes undemonstrable.
2. Will the identified risks affect the target product profile? Yes, two of them directly. The flat overall survival degrades the payer row of A.3b from premium to minimum. The 44% Grade 3+ hypertension rate degrades the safety row, and it is not mitigable by a dose reduction without also reducing the efficacy that the 10 mg/kg dose was selected for.
3. If a risk cannot be mitigated, is the asset still differentiated from competitors? Yes. Even with flat survival and severe hypertension in nearly half of patients, a 17–18% response rate against FOLFOX’s 5%, in an all-comers population with no approved option, is a differentiated product. The question is not whether it is differentiated but whether the differentiation is worth what it costs the patient.

Readout

Sources

The modelled estimate. Registry primary_completion_date 2026-12-01, plus two to four months to database lock and analysis, gives 2027-02 to 2027-04. data/benchmarks/readout-lag.json holds no observations, so the stated default of 01-rules.md rule 34 applies and the tag is [UNVERIFIED — modelled, default lag] rather than a benchmarked one. As stated in the table, this arithmetic answers a different question from the one the window answers.

Window

Basis. The company named a specific day, October 24, 2026, in a dated press release announcing an accepted ESMO oral presentation, and reiterated it three weeks later in its quarterly release; likeliest is that day. earliest and latest are the congress’s own opening and closing dates from data/congresses.json, which bound where the session could move to if ESMO reschedules it, and earliest additionally covers the possibility that ESMO’s own abstract release makes the headline numbers public on the opening day rather than at the podium. Precision is DAY because a source names a day; confidence is HIGH because two independent sources — the company and the society’s own calendar — agree, and because the underlying data are already generated and locked, so there is nothing left that could slip.

Disagreement. CONSISTENT. Four of the five sources point at the same week and three name the same day. The ctgov row’s 2026-12-01 is a different fact (trial completion) rather than a competing estimate of the same one, and the modelled row is arithmetic over that different fact; neither is treated as disagreeing with the other three, and the reason is stated in each row’s own note rather than resolved by picking a winner.

Date slippage. Five slips, parsed from the BPIQ note field and the company’s own releases, oldest first. Every one of them is a reporting date moving; none is an enrolment delay.

Six dated statements, five transitions, and every transition moved the date later or made it vaguer until the last one, which named a day for the first time. The 2025-08-11 statement pushed a Q1-2025-guided readout to Q1 2026 — a full year — and the Q1 2026 guidance then slipped to April 2026. Against that record, a date that is now pinned to a third party’s conference calendar is considerably more trustworthy than any of the company-only dates that preceded it, which is the substantive reason confidence is HIGH despite five slips.

Attribution

Status.

CONTAMINATED

Computed by lib/clustering.mjs’s attributionFor over this ticker’s full pipeline table (../company.md C.2) with CATALYST_CLUSTER_MIN_MONTHS = 6, for bpiq_drug_id 17398, and transcribed below rather than estimated by eye.

Conflicts

Note. Both conflicts are confirmed, and the mechanism is worth spelling out because it is asymmetric. Neither CTX-8371 nor CTX-10726 has a disclosed date — both carry the BPIQ period-end placeholder 2026-12-31 behind the text “Q4 2026”, which lib/clustering.mjs widens to the whole half-year 2026-07-01 to 2026-12-31 precisely because a placeholder should over-collide rather than under-collide. What makes these conflicts confirmed rather than INDETERMINATE is this program’s own side: a DAY-precision readout window is a real disclosure, and one disclosed side is all a pair needs. So the finding is not “three vague Q4 dates happen to touch” — it is “a known day sits inside two half-year windows that could open on top of it.”

The practical consequence is specific. Both colliding programs are early-stage assets that BPIQ flags is_high_mgmt_interest, and CTX-10726 is a PD-1 × VEGF-A bispecific — the most crowded fashionable class in oncology at the moment, where even a thin Phase 1 signal can move a small-cap share price more than a confirmatory presentation of already-public Phase 2/3 data. If either lands in the days around October 24, a price move in that window is not attributable to tovecimig.

A third event compounds this and is not in the table, because it is not a separate pipeline row: Compass expected FDA feedback on the COMPANION-002 dataset during the third quarter of 2026, ahead of a marketing application [VERIFIED — Compass second-quarter release, 2026-08-06]. That is the same program, so the clustering pass correctly does not count it as a conflict — but it is an undated, potentially larger catalyst on the same asset landing before the one this document scores. The scenario ranges and the stock-direction call below are therefore not presented as attributable to the October presentation alone, and the stock call is no-edge in part for this reason.

Market and timing for this event

• Plain takeaway. The market already knows this trial’s headline numbers, and it has already repriced them — down 65% in a single session in April. What it does not know is how long the responses lasted, and it does not know what the FDA said. The October presentation delivers the first of those two in front of an expert discussant who will say publicly what the field thinks. That is a real event with a genuinely two-sided outcome, sitting on a stock trading at 16.5% of its 52-week range with roughly a fifth of its shares sold short.
• Months to this catalyst. 2.1 months from the 2026-08-19 lock date to readout.window.earliest of 2026-10-23. readout.precision is DAY, so this figure is a real countdown rather than a guess — which is not true of most programs in this corpus.
• Expected move around this event. ../company.md C.6 records the options chain as unusable for this purpose: there is no listed strike at or near the $2.48 spot, so an at-the-money straddle cannot be constructed at all, and the two flanking strikes bracket the implied move anywhere from 43% to 65% of spot at mid prices and from 48% to 81% bid-to-ask. That is not a number. The bracket used instead is ±15% to ±35%, calibrated as follows: the class averages are +11%/−22% at Phase 3 and +9%/−9% for oncology, and the placebo-controlled trial-design amplitude is 29% [WEB ESTIMATE — IQVIA 2024]; those are announcement moves for hypothesis tests, and this event is a presentation of a dataset already reported, which argues for less. Set against that, this ticker’s own C.7 history is far more volatile than any class average (+49.7% and −64.8% on its two tovecimig hypothesis tests), and the specific fact outranks the class average, so the bracket is widened above the class figures rather than narrowed below them.
• Nearest comparable past reaction. ../company.md C.7’s 2024-11-08 row — CTX-471 Phase 1 biomarker data presented at the Society for Immunotherapy of Cancer meeting, +4.28% intraday. It is the closest analogue in kind: a congress presentation of data the company had already described, with no new hypothesis test. It is not a good analogue in magnitude, for three reasons: it was a Phase 1 asset rather than the company’s whole value, the shares were far less crowded then, and the 2024 float carried nothing like today’s 22% short interest. The two dramatic rows (2025-04-01 at +49.7%, 2026-04-27 at −64.8%) are the wrong comparables entirely — both were hypothesis tests — but they set the ceiling and floor of what this ticker is capable of.
• Materiality. ../company.md C.2 records this program as dominant: the only randomised dataset the company owns, the only planned marketing application, and the reason essentially every published analyst target exists. The stock-direction call below must be consistent with that, and it is — but consistency with dominance cuts both ways here. A dominant program presenting known data is not the same event as a dominant program running a test, and the no-edge call reflects the second half of that sentence, not a disagreement with the first.
• Date slippage. Five slips — see Readout above. Every one moved a reporting date, none an enrolment date, and the current date is pinned to a third party’s conference calendar rather than to company guidance.

Spot. $2.48, read 2026-08-19, cited from ../company.md C.4 (last_price, BPIQ fetch_company_info).

Scenario prices

Expected value. 55% × the positive midpoint ($3.00) + 45% × the miss midpoint ($1.90) = $2.51, which is +1.0% against the $2.48 spot. Method: probability_pct applied to the midpoint of each scenario range. This is arithmetic, not advice, and it is not a price target. Its flatness is the finding: the two sides of this event, weighted by a modest lean toward a satisfactory dataset, very nearly cancel, and that is what makes no-edge the honest stock-direction call rather than a hedge.

Run-up

readout.precision is DAY and readout.confidence is HIGH, so the date_confidence driver scores 100 and rule 39 does not withhold this call. That is unusual in this corpus and it is the main reason this program’s priority score is what it is.

Entry and exit

Predicted move and peak

Priority score drivers

Five of the seven are lib/runup.mjs’s scoreDriver output, transcribed. Two — unmet-need and value-uplift — are this document’s own judgement, read from the sections named.

Priority score. Computed by lib/runup.mjs’s priorityScore over the seven rows above, never hand-computed.

Priority score 49 · formula_version 1.0.0

That figure deserves one sentence of context, because it will look anomalous next to the rest of this repository. Every other program in the corpus carries a PERIOD- or UNKNOWN-precision readout, so date_confidence gates their scores down by roughly an order of magnitude. This one has a date disclosed to a day, so the gate is 1.0 and the score is essentially the weighted average of the other six drivers. It ranks high because the date is known, not because the thesis is stronger — and it would rank higher still if the CONTAMINATED clustering finding were not zeroing a quarter of the weighted base.

Settlement. Null at lock time. Settled only on an explicit user request, from a confirmed primary source, against the committed price cache — never on this document’s own initiative. See 05-prediction-protocol.md § Run-up settlement. Note that settling the run-up call on this program will require data/prices/CMPX.json to exist, which it does not today (../company.md C.8 row 4).

Verdict

What I would do. Watch, and do not hold this through October 24 for the presentation itself.

Why. The stock-direction call and the run-up call point different ways here, and that is the whole verdict. Into the event there is a genuine setup — 22% of the shares sold short and rising across three consecutive settlements, a date disclosed to the day, and FDA feedback on the marketing application guided to arrive inside the holding period. Across the event there is no edge at all: the expected value is +1.0% against spot, because the presentation delivers one unknown number (duration of response) on a dataset whose headline results are already public and already repriced. The underlying asset has a real problem that October will not fix — patients did not live longer, and the crossover explanation the company offers is contradicted by its own crossover-adjusted analysis, which moved the hazard ratio from 1.05 to 1.13. Against that, 44% of patients had severe hypertension.

What would change this. Two specific observables, in order of weight. First, FDA feedback: if Compass discloses that the agency has agreed a marketing application may proceed on response rate and progression-free survival without an overall-survival benefit, the single load-bearing judgement in B.1 resolves in the asset’s favour and the positive scenario range in Market and timing is too low. If the agency asks for a confirmatory survival trial, the programme’s timeline extends by years and the miss range is too high. Second, the duration-of-response figure itself: a median above 6.9 months would match the Phase 1b/2 cohort and reframe the durability argument; a median below the trial’s own 4.7-month progression-free survival would mean responders relapse no later than the average patient progresses, and would make the accelerated-approval case very hard to sustain.

What to watch.

• By 2026-09-30 — FDA feedback on the COMPANION-002 dataset, guided for the third quarter of 2026 in the 2026-08-06 release. Undated, larger than the catalyst this document scores, and expected to land first.
• Around 2026-10-20 — ESMO’s own abstract release ahead of the congress. The headline numbers, including duration of response, typically become public before the podium session.
• 2026-10-24 — the oral Proffered Paper presentation itself, by Dr. Nilofer Azad. Watch the formal expert discussant as closely as the presentation: this will be the first genuinely independent named assessment of this dataset, and A.5b records that no independent voice exists today.
• 2026-10-23 to 2026-12-31 — CTX-8371 and CTX-10726 Phase 1 data, both guided “Q4 2026”, both flagged is_high_mgmt_interest. Either could land on top of the tovecimig window; this is what the Attribution section’s CONTAMINATED status is about.
• Any date from 2026-10-24 — a drawdown on the $400M shelf registration, effective and undrawn since 2026-01-07. A raise into post-congress strength is the textbook move for a company funding a first launch.
• No fixed date — the first Form 4 filed since 2026-04-30. Four months of no insider buying at a price 64% below the February high is a standing negative; a genuine open-market purchase would be the first insider signal since the collapse.

Locked prediction

• Outcome-direction (will the readout succeed on its primary endpoint?): positive — probability 55%, band 42–68% [UNVERIFIED — modelled]. The band spans 50 and the lean is deliberately weak. Reasoning: the definition below has three conditions. The response-rate condition is close to settled — the company stated a final overall response rate of 18% on 2026-08-06, above the 15.0% bar. The safety condition is close to settled too — the safety dataset is essentially complete and already disclosed. The duration-of-response condition is genuinely open: the Phase 1b/2 precedent was 6.9 months in an enriched cohort, but the response rate halved between that study and this one, so a proportional shrinkage would put duration below the bar. What argues upward is that Compass is planning a marketing application, and a company about to file on response rate does not usually do so with a duration figure that undermines it. No third party has published a probability on this event.
• Stock-direction (which way do the shares move?): no-edge — confidence low — window 2026-10-19 to 2026-11-07, basis: opens with ESMO’s pre-congress abstract release, covers the 2026-10-24 presentation and the congress close, and runs two weeks past it to capture the digestion of the discussant’s assessment. Materiality is dominant per ../company.md C.2, and the call is consistent with that: a dominant program presenting data it has already reported is not the same event as a dominant program running a test. Attribution is CONTAMINATED — two other has_catalyst rows on this ticker, CTX-8371 and CTX-10726, carry “Q4 2026” windows that both cover this program’s disclosed day, so a move in this window is not attributable to this program alone. A third, undated event on the same program (FDA feedback, guided for the third quarter of 2026) is expected to land before the window opens.
• Scenario prices: positive $2.70–$3.30 · miss $1.65–$2.15
• Expected value: $2.51, +1.0% against spot $2.48 (arithmetic, not advice)
• Run-up: entry $2.48 on 2026-08-19, exit rule T-5 trading days before readout.window.earliest — predicted move +5% to +35%, predicted peak +10% to +45% around 2026-10 — priority score 49, formula_version 1.0.0
• Settles on: the ESMO Congress 2026 oral Proffered Paper presentation of COMPANION-002 (2026-10-24, presenter Nilofer Azad), scored against the presented abstract and slides and any same-day Compass press release. Positive definition: the presentation reports a confirmed overall response rate for tovecimig plus paclitaxel of at least 15.0% by independent central review, and a median duration of response of at least 5.0 months, and discloses no new Grade 3-or-higher treatment-emergent adverse-event category at a rate above the 44% Grade 3+ hypertension already reported on 2026-04-27. All three conditions must hold. If the presentation does not report a median duration of response at all, that counts as a miss, because the figure is the event’s entire content.
• Locked: yes · Settled: no

Program data-quality flags

1. data/prices/CMPX.json does not exist, so three figures that lib/prices.mjs normally owns were substituted: the 52-week range feeding the priced_in run-up driver (BPIQ’s high and low instead of the cache’s), the drift-since-last-catalyst leg of that same driver (not computed at all), and runup.exit (null, which is the correct value at lock time regardless but would otherwise become resolvable once the window is reached). Settling this program’s run-up call will require the cache. See ../company.md C.8 row 4.
2. CT.gov carries no investigator names for NCT05506943. search_investigators returned zero rows on two query shapes and get_trial_details returned no overall-official block, so A.5b’s investigator table is built from the design paper’s author list cross-checked against the trial’s own CT.gov site list. This is a weaker chain than a registry read and is stated as such in that section.
3. No conflict-of-interest statement was available for any named investigator. Neither the PubMed metadata payload nor the PubMed Central full-text render of PMID 38861293 / PMC11509068 carries a financial-disclosure section. Every “no competitor relationship found” entry in A.5b therefore rests on a search with nothing to search, which 02-connectors.md § KOL sources is explicit is not evidence of absence.
4. The registry’s primary_completion_date (2026-12-01) postdates the catalyst (2026-10-24), and this is not a source conflict. The primary endpoint was analysed in April 2025 and the survival analysis in April 2026; the registry field tracks last-patient completion, not data presentation. Recorded rather than resolved, per rule 24, because a reader comparing the two dates will otherwise assume one of them is wrong.
5. No peer-reviewed publication of any tovecimig clinical result exists. Every efficacy and safety figure in this document traces to a company press release or a congress abstract. PubMed returns four papers for this molecule in total, three of them the originator’s own preclinical or pharmacokinetic work and one a protocol description. For a registration-stage asset that is a thin evidence base, and it is the reason A.5’s publication-quality score is Low and A.5b’s judgement is UNKNOWN.
6. The licence terms with ABL Bio are unverified. The royalty and milestone figures used in B.3b come from consistent public summaries of the November 2018 agreement, not from the agreement or an SEC exhibit. They are used only to establish that B.3a’s peak-sales figures are gross rather than net to Compass, which does not depend on the exact rate.
7. No patent estate could be located for tovecimig. No composition-of-matter patent number, filing date or expiry was found by the exclusivity web search. B.1’s IP row is scored on the verified seven-year orphan-drug exclusivity alone. This is a gap in the analysis, not a finding that no patent exists.
8. The 2026-04-27 BPIQ historical-catalyst row’s intra_day_price_change_percent (+1.13%) is misleading on its own and the open_price_gap_percent (−64.81%) is the real move, confirmed against six same-day press items. 02-connectors.md advises preferring the intraday field over the gap field; that advice is wrong on this row, and the press-feed cross-check is what settles it. Recorded here because the same row is the anchor for the miss scenario above.