CGEM / zipalertinib-1l-egfr-ex20ins-nsclc — Zipalertinib for previously untreated, locally advanced or metastatic non-squamous non-small cell lung cancer with an EGFR exon 20 insertion mutation (first line)

Program analysis · bpiq_drug_id 17466 · prepared 2026-08 (refresh of the 2026-08-05 analysis) · USD · framework v5.6.0 · 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): An oral tablet designed to shut off a lung-cancer-driving EGFR mutation that ordinary EGFR tablets cannot reach, tested on top of chemotherapy in previously untreated patients.
• The event and when (as disclosed): Top-line REZILIENT3 Part B results — progression-free survival by blinded independent central review — “expected by the end of 2026” per Taiho via Cullinan’s Q2 2026 release (2026-08-06), reiterating the same guidance for the third time. ClinicalTrials.gov still lists primary completion five months later (2027-05-27); the disagreement is reported in the Readout section, not resolved.
• The main reason it could work: The identical experiment already succeeded — PAPILLON, same mutation, same line, same control arm, same endpoint, hazard ratio 0.395 — and zipalertinib carries a peer-reviewed, centrally reviewed 35.2% response rate from a harder, pretreated population, against a control arm (chemotherapy alone) worth only 6–7 months of PFS.
• The main risk: For the readout, an effect size below PAPILLON’s: zipalertinib hits one target where amivantamab hits two, the trial appears not to stratify by insertion location (which sank poziotinib’s far-loop patients), and the statistical assumptions are not public. For the trade, attribution: three autoimmune readouts on this ticker land in the same or the preceding quarter and have historically moved the stock harder than any zipalertinib event.
• What it means for the stock: A meaningful-but-not-dominant program (see ../company.md C.2): a win is worth roughly 15–30% of enterprise value on the modelling here, and the shares already sit within 8% of their 52-week high with a ~23%-of-float short base. The direction call is up, at low confidence, inside a window contaminated by the company’s own other catalysts.

0. Program-tier coverage — CLEARED

Company-tier coverage is in ../company.md C.0. The catalyst is a trial data readout, so the conditional regulatory tier in 02-connectors.md does not bind this sweep.

A. Scientific & clinical assessment

A.1 Mechanism of action and scientific rationale

Zipalertinib is an oral, irreversible (covalent) small-molecule inhibitor of EGFR built on a pyrrolopyrimidine core. An EGFR exon 20 insertion wedges a few extra amino acids into the kinase domain, jamming the alpha-C helix in its active orientation so the receptor signals continuously without any growth factor. The same jam narrows the drug-binding pocket, which is why first-, second- and third-generation EGFR tablets do not work against it. Zipalertinib is shaped to fit the narrowed pocket and carries an acrylamide warhead that bonds permanently to cysteine 797, and it is selective for mutant over wild-type EGFR, which keeps the dose-limiting rash and diarrhoea low [VERIFIED — PubMed: PNAS crystallography, DOI 10.1073/pnas.2417144121; Cancers 2026 review, DOI 10.3390/cancers18020323]. In REZILIENT3 it is added to platinum doublet chemotherapy (pemetrexed plus carboplatin or cisplatin), which kills dividing cells by an entirely separate mechanism.

Target validation is strong by the standard that matters: three drugs against this exact mutation are already approved (amivantamab first-line, sunvozertinib and now-withdrawn mobocertinib second-line), and independent academic crystallography at Dana-Farber/Harvard shows TAS6417 (zipalertinib) inhibits the two commonest insertion variants, insASV and insSVD, in a mutant-selective manner [VERIFIED — DOI 10.1073/pnas.2417144121]. The step the readout must prove is narrower than the biology: that adding zipalertinib to chemotherapy delays progression longer than chemotherapy alone, in a randomised trial, under blinded central review.

The honest scientific risk is the effect size, through one specific mechanism: in vitro, zipalertinib inhibits near-loop insertions more potently than far-loop ones, and that distinction was decisive for poziotinib (PFS 11.1 months near-loop versus 3.5 far-loop) [VERIFIED — Nat Commun, DOI 10.1038/s41467-025-61817-8]. The REZILIENT3 registry record shows no stratification by insertion location. A far-loop-heavy randomisation imbalance, or a genuinely weaker far-loop effect, is the specific way this trial comes in below PAPILLON — and it is the main reason the probability band below is 23 points wide. Zipalertinib is also inactive against C797S, the mutation that removes its covalent anchor [VERIFIED — DOI 10.3390/cancers18020323] — a resistance question, not a readout question.

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

Trials

REZILIENT3 design detail that bears on the readout. Primary endpoints are PFS by BICR (the efficacy event) and the rate/severity of treatment-emergent adverse events, with Part A’s dose-limiting-toxicity endpoint already cleared. Eleven secondary endpoints include ORR, DoR, intracranial response measures, overall survival (Part B), and three patient-reported instruments — which are unblinded in an open-label trial and therefore soft; central review protects the PFS number specifically. Control-arm patients may cross over to zipalertinib monotherapy after BICR-confirmed progression: ethically correct and analytically unhelpful, because it dilutes any overall-survival difference — OS should not be expected to carry the result. An independent data monitoring committee watches interim safety. Key exclusions: any prior ex20ins-targeted therapy, any history of interstitial lung disease or pneumonitis of any grade, QTcF above 470 ms, strong or moderate CYP3A4 inducers/inhibitors within 7 days [VERIFIED — CT.gov get_trial_details, 2026-08-10].

The statistical assumptions are not public. The assumed hazard ratio, the target event count that triggers the analysis, the alpha allocation across the co-primary endpoints, whether an interim efficacy analysis exists, and whether randomisation is stratified by insertion location were not found in the registry, the protocol paper, or any search — re-checked this sweep. The stratification question most affects the modelled probability and remains unanswered [UNVERIFIED — absence across all sources].

Feasibility matrix

Resourcing sufficiency. Effectively a non-issue: the trial is fully enrolled and Taiho carries the operational load; development costs are shared 50/50. Cullinan holds $356.0M against a $12.74M monthly burn (../company.md C.3) — runway past the readout by roughly two years.

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

A.3a Target label. First-line treatment of locally advanced or metastatic non-squamous NSCLC with a documented EGFR exon 20 insertion mutation, in combination with pemetrexed and a platinum agent.

A.3b Target product profile

A.3c Strategic Go/No-Go questions (pre-Phase-III / registration set, as the asset sits at a Phase 3 readout):

A.3d Regulatory designations.

• Breakthrough Therapy designation (FDA, zipalertinib): grants intensive FDA guidance on trial design, senior agency engagement, and rolling/priority review eligibility. Awarded on preliminary evidence of substantial improvement over available therapy [WEB ESTIMATE — company 10-K summary and NDA press materials; the filing itself was not opened].
• No orphan designation, and none expected: lung cancer is far too common and the ex20ins subset has not been carved out [VERIFIED as a negative search across the press feed and web searches].
• No Special Protocol Assessment disclosed for REZILIENT3 [UNVERIFIED — absence of evidence].

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.

A.5 Evidence quality and endpoints

Endpoints

A.5b Key opinion leaders

Panel as of. 2026-08-10 — the date the CT.gov search_investigators and PubMed searches below were run.

Investigators

The REZILIENT3 registry record lists no overall officials and no named site investigators, so the paid-by-trial panel below is identified from the programme’s own publications, where authorship on the trial’s dataset is the relationship. Being paid by the trial is a relationship with the sponsor by construction; Conflicts is for relationships beyond that one. No conflicts search below reached a journal disclosure appendix — PubMed metadata does not carry them and the full-text appendices were not opened this sweep — and per rule 40 the column records that limit rather than certifying a clean record.

Independent voices

Judgement

Dissent

B. Commercial assessment

B.0 Current treatment algorithm

1. Diagnosis and molecular testing. A real filter: exon 20 insertions are structurally diverse and older PCR panels miss many of them, while next-generation sequencing finds them. An undetected patient never enters this algorithm and is lost to every drug in the class equally.
2. First line, United States, since March 2024: amivantamab plus carboplatin and pemetrexed. Works well and is hard on patients — infusion reactions on first exposure in the majority, rash, paronychia, raised clot risk.
3. First line where amivantamab is unavailable or not tolerated: platinum doublet chemotherapy alone. ORR 19–47%, PFS 6–7 months. This is REZILIENT3’s control arm.
4. Second line and later: sunvozertinib (oral, US accelerated approval 2025, ORR 46%, DoR 11.1 months), or zipalertinib once approved — the pending 2027-02-27 decision covers exactly this slot.
5. After that: further chemotherapy, clinical trials, or supportive care.

Zipalertinib-plus-chemotherapy would compete for step 2, directly against amivantamab plus chemotherapy. It opens no new line of therapy and creates no new patient pool — these patients already exist and are already treated. Its case is substitution on the strength of being a tablet rather than a drip.

B.1 Competitive landscape and positioning

Where this asset wins, and the single fact the thesis rests on. Every competing first-line option in the United States involves a drip; zipalertinib is a tablet with low rates of the classic EGFR toxicities, in a disease where patients take treatment until it stops working. The structural problem is equally plain: REZILIENT3 compares against chemotherapy alone (step 3) while competing for the amivantamab slot (step 2). When the trial began enrolling in December 2023 chemotherapy alone was the standard; PAPILLON’s approval three months later moved the goalposts mid-trial. Regulators accept the comparator and in many countries chemotherapy alone remains the practical standard, but a statistically successful trial can be commercially ambiguous, and no head-to-head against amivantamab is running or planned. A 2026 peer-reviewed review states it plainly: the relative impact of monotherapies versus chemotherapy combinations in this mutation “currently still lacks robust evidence” [VERIFIED — DOI 10.3390/cancers18020323]. Against sunvozertinib and furmonertinib, both oral, the route-of-administration differentiation largely disappears.

B.2 Addressable market

B.3 Value and feasibility

B.3a Expected peak sales. Built bottom-up because no analyst forecast specific to zipalertinib exists — re-verified this sweep; aggregate “EGFR inhibitor market” reports were found and again not passed through (rule 6). All scenarios are conditional on REZILIENT3 succeeding and first-line approval following, cover the United States only (Cullinan has no other economics), exclude the separate second-line opportunity (bpiq_drug_id 16438), and exclude the milestone payments, counted separately in B.3b.

The Cullinan share applies 50% of an assumed 50–60% pre-tax margin to brand sales; neither party discloses a margin, so that conversion is the least verifiable input [UNVERIFIED — modelled]. The least defensible operating input is the peak share: it depends on a cross-trial tolerability comparison no trial will run, and the 25–55% spread is the honest width of that uncertainty.

B.3b Feasibility

B.4 Product-development risk

Framing questions.

1. Does the development plan support the target product profile claims? Mostly. The randomised Phase 3 supports the efficacy-versus-chemotherapy claim and the safety claim; nothing in the plan supports the claim that matters commercially — superiority or parity against amivantamab — because no such comparison exists or is planned.
2. Will the identified risks affect the target product profile? The near-loop/far-loop sensitivity question can shrink the effect size that the profile’s efficacy row promises; the pneumonitis exclusion means the trial population is systematically healthier-lunged than the real-world population the label would cover.
3. If a risk cannot be mitigated, is the asset still differentiated from competitors? Against infused amivantamab, yes — oral dosing and tolerability survive any plausible readout short of a miss. Against oral sunvozertinib and furmonertinib, no — the differentiation is the data, and the data are not head-to-head.

Readout

Sources

The modelled estimate. The registry’s primary_completion_date (2027-05-27) plus rule 34’s stated default of two to four months to database lock and analysis gives 2027-07 to 2027-09. data/benchmarks/readout-lag.json holds no comparable observation, so the default applies and the tag says so. This arithmetic keys on the one field this sweep judges unmaintained (see the ctgov row), and REZILIENT3’s top-line is an event-driven PFS analysis that can trigger well before primary completion — no event count is public, so no independent event-side arithmetic is possible. The modelled row therefore bounds the late tail rather than centring the estimate, and the window below leans on the thrice-repeated guidance instead.

Window

Basis. Earliest is the opening of the guided quarter — the March guidance said “by Q4 2026”, so nothing supports a Q3 release. Likeliest sits late in the guided window because the guidance is deadline-shaped (“by the end of 2026”, stated three times without narrowing), and an event-driven analysis run by a partner who guides to a deadline tends to land near it rather than early in it. Latest allows one quarter of slip beyond the deadline: the registry’s 2027-05-27 primary completion is judged unmaintained, but the plainer reading — that the readout slips into 2027 — is not excluded, and one quarter is the typical size of a first slip. Precision is PERIOD because no source names a month; confidence is MEDIUM because the guidance has been repeated identically three times with zero slips and enrolment completed on schedule, but it is a partner’s deadline rather than a disclosed date, and the registry points later.

Disagreement. UNRESOLVED. Company/partner guidance (three statements, 2026-03-10 through 2026-08-06) says by end of 2026; ClinicalTrials.gov primary completion says 2027-05-27, five months after the guided window closes. Reported, never averaged, never resolved by picking one (rule 24). The likeliest reconciliation — Taiho guides to a pre-specified event-driven analysis while the registry tracks (and neglects) the last-visit field — is an [UNVERIFIED] inference.

Date slippage. Zero slips on the readout guidance. Zero is a finding, with a caveat: the guidance is only five months old, so a clean record here is weaker evidence than one spanning years. (The programme’s enrolment guidance was also met: “complete enrollment by H1 2026” guided 2025-08-07, reported complete 2026-03-10.)

Attribution

Status. CONTAMINATED

Computed 2026-08-10 with lib/clustering.mjs’s attributionFor over this ticker’s pipeline for bpiq_drug_id 17466, CATALYST_CLUSTER_MIN_MONTHS = 6, using this document’s own readout window for this program’s range; transcribed, not estimated.

Conflicts

Note. Required because the status is CONTAMINATED. The confirmed conflict is this drug’s own second-line PDUFA (2027-02-27, an exact disclosed day) landing inside this window’s latest edge — and it is the smaller problem, because the stock-call window below deliberately closes on 2027-01-31, before it. The larger, unconfirmed-but-real problem is the four T cell engager readouts sharing the window: CLN-978 rheumatoid arthritis (Q3 2026, immediately before), CLN-978 lupus and Sjögren’s and velinotamig lupus (all Q4 2026, simultaneous) — one more simultaneous readout than the 2026-08-05 analysis faced, velinotamig having joined the catalyst list since. Company.md C.7 shows the engager programs move this stock harder than any zipalertinib event (+14.23%, +18.87% against +5% to +13%), and C.2 marks two of them dominant against this program’s meaningful. Any share move inside this window is therefore presumptively attributable to the autoimmune franchise first, and the scenario ranges below cannot be read as this event’s isolated effect. The stock-direction call carries this caveat explicitly.

Market and timing for this event

• Plain takeaway. The stock sits within 8% of its 52-week high, up roughly 3× from the October 2025 low, with ~23% of the float sold short and four data readouts due on the ticker in one quarter. The market is positioned for the autoimmune programs; this readout is the cheaper, better-understood event riding in the same window.
• Months to this catalyst. About 2 to 5 from 2026-08-10 against the readout window (earliest 2026-10-01, likeliest 2026-11-30) — a PERIOD-precision window, so entry/exit timing carries a quarter’s worth of slack, not a date.
• Expected move around this event. Not readable from the chain: ../company.md C.6 records the December 2026 expiry (the right one — it expires before the 2027-02-27 PDUFA) carrying 1,673 contracts of open interest against zero traded on the read date, with near-money spreads wider than their own mids. Bracket from the ticker’s own reaction function instead: roughly +8% to +20% on a clear positive, −10% to −25% on a miss, from C.7’s zipalertinib rows (+5.04% to +12.85% on efficacy wins at lower bases) widened for the larger base and the short interest.
• Nearest comparable past reaction. C.7’s 2025-01-29 row: REZILIENT1 met its primary ORR endpoint, +12.85% intraday, clean attribution. Comparable: same drug, same disease, an efficacy result on a primary endpoint. Not comparable: the shares were ~$5.50 then (a third of today’s), REZILIENT1 was single-arm while this is randomised and competitive, and the autoimmune programs had not yet become the reason people own the stock. Most sobering row: the NDA acceptance, −4.40%.
• Materiality. “Meaningful, not dominant” per ../company.md C.2 — a win is worth roughly 12–33% of enterprise value on B.3b’s modelling, capped by the 50% US-only economics, the stale comparator, and two dominant-materiality CLN-978 rows in the same pipeline table. The stock call below is sized to that, per rule 27.
• Date slippage. Zero slips across three statements (Readout, above); guidance five months old.

Spot. $18.03, read 2026-08-10 intraday (../company.md C.4; up 1.67% on the day, four days after the Q2 2026 results).

Scenario prices

Expected value. 0.78 × $21.75 (positive midpoint) + 0.22 × $14.875 (miss midpoint) = $20.24, +12.2% against spot $18.03. Probability applied to the midpoint of each range. Arithmetic, not advice, and not a price target.

Run-up

Entry and exit

Predicted move and peak

Priority score drivers

Five of the seven transcribed from lib/runup.mjs’s scoreDriver (computed 2026-08-10); the two judgement drivers read from this document.

Priority score. Priority score 8 · formula_version 1.0.0 — from lib/runup.mjs’s priorityScore, never hand-computed. The PERIOD-precision gate (0.20×) and the maxed clustering risk sink an otherwise decent base: this is the formula saying “the readout is probably good and the trade around it is bad”, which is also this document’s prose conclusion.

Settlement. Left null at lock time; settled only on an explicit user request against the committed price cache, per 05-prediction-protocol.md § Run-up settlement.

Verdict

What I would do. Watch. Unchanged from 2026-08-05, with the trade now slightly worse than the science.

Why. A positive REZILIENT3 remains the most likely single outcome by a wide margin: the identical experiment succeeded emphatically (PAPILLON, HR 0.395), zipalertinib carries a peer-reviewed 35.2% centrally reviewed response rate from a harder population, the control arm is chemotherapy alone, and the partner has now repeated “top-line by end of 2026” three times without moving. But everything that made this a poor trade five days ago got incrementally stronger this sweep: velinotamig’s lupus data joined the same quarter (four readouts in Q4 now, plus CLN-978 rheumatoid arthritis in Q3), the shares sit 7% below the 52-week high after a 3× run, and the filed share count shows the preferred quietly converting into the strength. Cullinan keeps only half of United States profits and nothing elsewhere on a drug whose composition-of-matter patent expires in 2034; the trial beats a comparator that stopped being the US standard in March 2024; and this ticker’s own record prices zipalertinib news at +5% to +13% on wins and −4% on regulatory progress while paying +14% to +19% for the T cell engager story.

What would change this. Upward: disclosure of REZILIENT3’s statistical assumptions (hazard ratio, event count, stratification by insertion location), which would turn the central efficacy risk into arithmetic; or Taiho narrowing “by end of 2026” to a month. Downward: the guidance moving for the first time; ClinicalTrials.gov moving primary completion beyond 2027-05-27; a pneumonitis signal anywhere in the programme; or a complete response letter at the 2027-02-27 second-line decision, which reads directly across to a first-line filing built on the same dataset.

What to watch.

• Q3 2026 (by 2026-09-30): CLN-978 rheumatoid arthritis multi-dose data (id 18971) — lands before this readout and resets the price level it is measured against; also the registrational CLN-049 Phase 2 start (id 13562).
• Any date from 2026-10-01: the REZILIENT3 top-line itself, from Taiho rather than Cullinan. Read the hazard ratio and median PFS against PAPILLON’s 0.395 / 11.4 months before reading the p-value.
• Q4 2026: CLN-978 lupus (18188) and Sjögren’s (18907) data, and velinotamig lupus data (19423) — the three simultaneous contaminants of this window’s attribution.
• ~2026-11-06 (next quarterly release): whether “by end of 2026” survives a fourth statement; the first cash print after this analysis; whether the ClinicalTrials.gov primary completion date moves.
• FINRA settlements (semi-monthly): whether the 16% short base builds or covers into the quarter.
• 2027-02-27: the second-line PDUFA (id 16438) — outside this prediction’s window by design.

Locked prediction

• Outcome-direction (will REZILIENT3 Part B meet its primary PFS endpoint?): positive — probability 78%, band 65–88% [UNVERIFIED — modelled]. Above the ~35–40% Phase 3 oncology base rate for four verified reasons (exact successful precedent in the same design; centrally reviewed 35.2% ORR from a harder population; a chemotherapy-alone comparator worth 6–7 months PFS; full enrolment with the dose cleared) and held below ~90 by three unresolved hazards (single-target mechanism vs PAPILLON’s dual; BICR strictness; no public statistical assumptions, including insertion-location stratification).
• Stock-direction (which way do the shares move?): up — confidence low — window 2026-10-01 to 2027-01-31, basis: the guided readout window plus one month to absorb it, closing before the 2027-02-27 PDUFA so this call is not scored on that event. Materiality per ../company.md C.2 is meaningful-not-dominant, and attribution is CONTAMINATED (four same-window catalysts) — the direction is retained rather than declined to no-edge because it agrees with the expected value, but any move inside the window may not be attributable to this event.
• Scenario prices: positive $19.50–$24.00 · miss $13.50–$16.25 (anchors in the table above).
• Expected value: $20.24, +12.2% against spot $18.03 (2026-08-10). Arithmetic, not advice.
• Run-up: entry $18.03 on 2026-08-10, exit rule T-5 trading days before readout.window.earliest (2026-10-01) — predicted move −10% to +18%, predicted peak 0% to +28% around 2026-09 — priority score 8, formula_version 1.0.0.
• Settles on: Taiho and/or Cullinan report REZILIENT3 (NCT05973773) Part B met its primary endpoint of progression-free survival by blinded independent central review, with a statistically significant improvement for zipalertinib plus chemotherapy over chemotherapy alone. A release stating the endpoint was not met, or futility, settles as a miss. A delay is not a miss on the outcome side; the stock side settles on its window regardless.
• Locked: yes · Settled: no

Program data-quality flags

• Open Targets search_entities BLOCKED, verbatim Rate limit exceeded for client: global, four attempts. No independent genetics-based target validation obtained; mitigated by three approved drugs against the mutation and independent crystallography.
• The catalyst date still disagrees between sources by five months (guidance “by end of 2026” vs registry primary completion 2027-05-27) and is reported rather than reconciled (rule 24). The registry’s primary-completion and overall-completion dates remain identical, against outcome time frames of “approximately 5 years” — internally inconsistent, so the registry field is treated as unmaintained and the modelled readout estimate that keys on it as a late bound only.
• has_results is false on NCT05973773; no REZILIENT3 efficacy data exist. Everything quantitative about zipalertinib comes from REZILIENT1 — a different trial, in a different line, with no control arm.
• REZILIENT3’s statistical assumptions are not public: assumed hazard ratio, target event count, alpha allocation, interim analysis existence, and stratification by insertion location were again not found. The stratification question most affects the modelled probability.
• REZILIENT3 is open-label; central review protects the PFS endpoint specifically, the three quality-of-life instruments are unblinded and soft, and the optional crossover structurally dilutes overall survival.
• Part A’s combination-safety conclusion rests on six patients.
• PubMed returned 27 records (>15), metadata retrieved on 14; the 13 unopened are older and no claim rests on pre-2024 literature not directly read.
• ChEMBL route flags remain wrong (oral: false for an oral drug) and remain unused; ChEMBL returns no bioactivity data, so selectivity claims rest on PubMed.
• The patent estate and Breakthrough Therapy designation remain a WEB ESTIMATE from annual-report summaries; the filings were not opened, and the family count differs between report years (six vs seven) — reported, not reconciled.
• PAPILLON, sunvozertinib and furmonertinib figures come from peer-reviewed reviews and web search, not the primary trial publications.
• No zipalertinib-specific analyst peak-sales forecast exists; every B.3a figure is modelled bottom-up, and aggregate “EGFR market” forecasts were deliberately not passed through (rule 6).
• The 50–60% pre-tax margin converting brand sales to Cullinan revenue is assumed, not disclosed — the least verifiable input in section B.
• The REZILIENT3 registry lists no overall officials or site investigators, so the KOL investigators table is built from trial publications; no conflicts search reached a journal disclosure appendix, and the tables say so rather than certify clean records (rule 40).
• Company-level flags live in ../company.md C.8 — most consequentially this sweep: the BPIQ 13F quarter-labelling bug, the changed composition of the BPIQ cash field, and the stale market-cap share count corrected from EDGAR XBRL.