CMPX — Compass Therapeutics, Inc.

Company context · prepared 2026-08-19 · company sweep 2026-08-19 · USD · framework v5.10.3

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

Program analyses for this company are the sub-folders listed in C.2.

C.0 Company-tier coverage — CLEARED

Every mandatory company-tier row was called. The regulatory tier was not called at all, and that is correct rather than a gap: 02-connectors.md makes that tier conditional on a regulatory catalyst (a Prescription Drug User Fee Act target date, an advisory committee, or another decision by a regulator), and the catalyst analysed under this ticker is a congress data presentation. A Biologics License Application (BLA — the marketing application a biological drug needs before it can be sold) is planned for later in 2026 but has not been submitted, so there is no regulatory docket to query yet.

One connector could not be used at all, and it is not a sweep row. data/prices/CMPX.json does not exist: this ticker has never been fetched by scripts/fetch-prices.mjs, and that script writes outside this analysis’s own folder, so this run could not create it. 02-connectors.md § The gate is explicit that the price cache is not a coverage row, and 03a-company-spec.md C.4 gives the by-hand fallback for a ticker the cache does not cover. C.4 below therefore recomputes the 52-week position by hand from BPIQ’s own high and low. The consequence is recorded in C.8 row 4 and is carried into the program document’s run-up section, where one priority-score driver reads the same range.

C.1 What the company is

Compass Therapeutics is a clinical-stage cancer company in Boston, Massachusetts, founded in 2014, that makes antibody drugs — proteins engineered to stick to a chosen target on or around a tumour [VERIFIED — BPIQ fetch_company_info 2026-08-19]. It runs four molecules across six pipeline rows, and it has no approved product and no product revenue: the 2025 Form 10-K reported $0.0M of revenue against a $66.5M net loss [VERIFIED — press feed, 10-K summary 2026-03-05].

It is close to a single-asset company in practice, though not on paper. One molecule, tovecimig (also called CTX-009), accounts for the only programme that has finished a randomised trial, the only planned marketing application, and essentially all of the sell-side valuation. The other three molecules — CTX-471, CTX-8371 and CTX-10726 — are all in Phase 1 or awaiting a Phase 2 start.

Tovecimig is not wholly owned. Compass licensed it from the South Korean biotechnology company ABL Bio in November 2018, holds rights worldwide except Korea and China, and owes ABL Bio tiered single-digit royalties on net sales plus up to roughly $399M of development, regulatory and commercial milestone payments in oncology [WEB ESTIMATE — public summaries of the ABL Bio / Compass licence, retrieved 2026-08-19]. The programme document treats that as an economic haircut on peak sales rather than as a verified contract term, because the licence itself was not read this sweep.

C.2 Pipeline — every program

Every row returned by fetch_company_drugs on 2026-08-19 is listed. “Date to a day?” is yes only where catalyst_date_text names a day (01-rules.md rule 23); every “no” row carries a BPIQ period-end placeholder in catalyst_date that is never used for timing anywhere in this repository.

Three rows carry has_catalyst: true and all three land in the fourth quarter of 2026. That concentration is what the programme document’s Attribution section is about; it is a real feature of this pipeline, not a placeholder artifact, because one of the three dates is disclosed to a day.

C.3 Financial position

The recomputed runway and the company’s own guidance disagree by roughly eighteen months, and the company’s is the one to use. Cash and marketable securities fell from $209M at 2025-12-31 to $180M at 2026-06-30 — $29M over six months, or $4.83M a month, which is what makes the BPIQ burn field credible. But a company that is hiring a Chief Commercial Officer (appointed 2026-01-05), preparing a marketing application and building launch capability does not keep burning at its clinical-stage rate. “Into 2028” is management telling you the burn rate is about to rise. The recomputed figure is shown because the template asks for it, not because it is a better estimate. BPIQ’s own screening text carries a third figure, a cash runway to 2029-01-08 [WEB ESTIMATE — BPIQ bpiq_pick_criteria, undated]; it is unsourced and undated and is recorded here only so a reader who sees it elsewhere knows it was considered and set aside.

Dilution history.

The share count confirms the August 2025 raise: 138,282,498 shares at 2025-08-04 became 177,862,102 at 2025-10-30 [VERIFIED — EDGAR XBRL dei:EntityCommonStockSharesOutstanding], an increase of 39.6M shares, which at the $120M gross figure implies roughly $3.03 a share. There has been no withdrawn offering in the window. A raise before the next catalyst is plausible rather than likely. The $400M shelf is effective and completely unused, the company has 36 months of clinical cash and says it has 18, and the natural moment to draw on a shelf is into strength after a good congress presentation rather than before one at a share price 64% below the February high. The programme document’s run-up section treats this as a risk that grows after the October event, not one that threatens the position before it.

C.4 Valuation frame

This is the run-up baseline for every programme under this ticker. The shares sit near the bottom of a range whose top was set on 2026-02-03, before the overall-survival readout; the whole distance between $6.88 and $2.48 was travelled on one programme’s data.

C.5 Ownership and flow

Insider flow is grants and tax sales, with two small genuine purchases and nothing since April. Nearly every acquisition row is a stock-option grant at a $0.00 stated price (2026-01-02: 2,000,000 options to the chief executive, 1,000,000 each to the chief commercial and chief medical officers, 720,000 to the general counsel, 670,000 to the chief financial officer, 90,000 to each of six directors). The disposals cluster on the anniversaries of vesting — the chief executive sold 46,777 shares at $5.35 on 2026-01-09 and 36,687 at $6.52 on 2026-02-08, both consistent with withholding to cover tax on vested restricted stock rather than a view on the shares.

The only recent open-market-style acquisitions are small and, importantly, before the April collapse in one case and just after it in the other: 15,000 shares at $1.89 by the chief accounting officer and 25,000 shares at $1.89 by the general counsel, both dated 2026-04-29, two days after the overall-survival readout. Together those are $75,600 — a gesture, not a signal, and the press feed labels the second one an equity grant rather than a purchase, which the $1.89 price contradicts. It is recorded as ambiguous.

Nothing has been filed since. EDGAR shows no Form 4 after 2026-04-30. Four months of a 64%-lower share price with no insider buying is the finding, and it cuts against the bullish read of the two April purchases.

Concentration is high and the register turned over hard across the April readout. BPIQ labels each block by filing quarter, so the block labelled 2026Q3 holds positions as at 2026-06-30 and the one labelled 2026Q2 holds positions as at 2026-03-31 — see C.8 row 5 for how that was established rather than assumed. Between those two dates, which is to say across the 2026-04-27 collapse, the register went from ten funds to six: BVF (7.76M shares), Sofinnova (2.06M), Logos (3.15M) and RTW (4.19M) all disappear; Tang Capital adds (16.79M → 17.99M); Palo Alto Investors adds heavily (1.51M → 5.83M); Vivo cuts (9.55M → 7.76M). OrbiMed holds 15,219,994 shares unchanged in every block back to the one labelled 2025Q3. Four dedicated healthcare funds left through the readout and two bought the other side of it. [VERIFIED — BPIQ fetch_company_hedge_fund_holdings 2026-08-19, share counts]

Two law firms — Bronstein, Gewirtz & Grossman and Johnson Fistel — announced investigations of the company on 2026-04-27, 2026-05-05, 2026-05-07 and 2026-07-02 [VERIFIED — press feed]. No securities-class-action complaint appears in the EDGAR feed and none is recorded as filed; solicitation announcements are not litigation, and they are noted here only because a reader will see them.

C.6 Options chain and its readability

Plain takeaway: the chain cannot price this event, and the reason is structural rather than merely noisy. There is no listed strike at or near the $2.48 spot — the strikes jump from $2.00 straight to $3.00 — so an at-the-money straddle (buying a call and a put at the same strike, the standard way to read what the market expects a stock to move) cannot be built at all. What can be read is that speculative activity is real and one-sided: 20,548 contracts of the November $3.00 call changed hands on the sweep day against an open interest of 2,278, so the day’s volume was nine times the entire pre-existing position in that contract.

The programme document’s expected-move figure is therefore taken as a bracket calibrated against framework/07-benchmarks.md, not from this chain.

C.7 Reaction to past catalysts

Every row cross-checked against the press feed for the same date.

What the table says as a whole. This ticker moves violently on tovecimig hypothesis tests (+49.7% on the primary hit, −64.8% on the survival miss) and modestly on congress presentations of data already in hand (+4.3% on the 2024 Society for Immunotherapy of Cancer poster). The October 2026 event is the second kind, and the programme document’s scenario ranges are built on that distinction rather than on the two dramatic rows.

C.8 Company data-quality flags

1. market_cap is computed off the previous close. $430,410,144 ÷ 180,087,915 filed shares = $2.3900 exactly, which is previous_close, against a $2.48 last_price. This is the MNOV-pattern failure 02-connectors.md records, not the CNTB stale-share-count pattern: the share count is current (2026-07-28, from the 10-Q filed eight business days before the sweep) and the price is stale. C.4 recomputes from the filed count and the last price.
2. max_52_week_position is computed off the previous close. 0.14801 = (2.39 − 1.61) ÷ (6.88 − 1.61) to five decimals. The bug fires on this ticker; C.4 recomputes to 16.5%.
3. enterprise_value cannot be reconciled against the same payload’s cash field. $245,757,083 against a $430,410,144 market capitalisation implies subtracting $184,653,061. The payload’s own cash is $179,879,000, leaving $4,774,061 unexplained, and no debt line or prior quarter-end cash figure in the sweep accounts for it. C.4 recomputes.
4. data/prices/CMPX.json does not exist, so every figure that lib/prices.mjs normally owns was derived by the C.4 fallback instead. This is not a coverage gap (the price cache is not a sweep row) but it is a real limitation: the 52-week range here is BPIQ’s own high and low rather than the cache’s own intraday extremes, no drift-since-last-catalyst can be computed, and the programme document’s run-up priced_in driver reads the same substituted range. Running node scripts/fetch-prices.mjs CMPX outside an analysis run would close this.
5. Hedge-fund block labels are filing quarters on this ticker — the standard convention, but checked rather than assumed. Dividing value by shares gives an implied price that is internally consistent inside each block: the 2026Q3 block returns $2.190, $2.186, $2.190, $2.167, $2.190 and $2.190; the 2026Q2 block returns $5.29 on all ten rows; the 2026Q1 block returns $5.37. Only one reading fits: $2.19 is a post-collapse price and $5.29 and $5.37 are pre-collapse prices, so the 2026Q3 block holds 2026-06-30 positions, the 2026Q2 block holds 2026-03-31 positions, and each block’s holdings are one quarter older than its label. 02-connectors.md is explicit that this convention is not consistent across tickers and must be matched before use; it was matched here, and the C.5 narrative is dated accordingly. Share counts are unaffected either way and are what the narrative rests on. The exact 2026-06-30 close could not be confirmed against the price cache because the cache does not cover this ticker (row 4); the argument above rests on the three implied prices sitting on opposite sides of a 64% collapse, which does not need the close.
6. The note field is a dated slip changelog on four of six rows, exactly as documented. CTX-471’s Phase 2 start has been re-guided four times (H2 2025 → Q1 2026 → mid-2026 → H2 2026 → Q3 2026) and CTX-10726’s first data twice. The tovecimig row’s slips are counted in the programme document rather than here.
7. short_float disagrees with FINRA and is not used. BPIQ reports 0.30106 with short_data_updated_at 2026-07-31. The FINRA 2026-07-15 settlement gives 39,298,468 shares, which is 21.8% of shares outstanding or 24.9% of estimated non-affiliate float — neither of which is 30.1%. The gap could be genuine (short interest was rising ~5% per settlement, so a 2026-07-31 reading near 47M shares would reach 30% of float) or it could be the units problem 02-connectors.md flags. It could not be resolved this sweep because the 2026-07-31 settlement request timed out twice. C.5 reports the FINRA figure with its own denominator stated.
8. A “Revenue ($25.25M)” figure circulates in third-party summaries of the second-quarter 10-Q and is rejected under 01-rules.md rule 6. The company’s own 2025 Form 10-K reports $0.0M of revenue and the 10-Q reports a $25.154M net loss for the quarter; $25.25M is a loss or expense line mislabelled as revenue by an automated summariser. Compass has no product and no revenue.
9. Published analyst-target aggregates conflict badly and the stale ones are excluded. One aggregator reports a $14.57 average with a $30 high; the actual post-second-quarter targets are Stifel $6 (maintained 2026-08-14), Piper Sandler $6 (raised from $4, 2026-08-07) and Leerink $7 (raised from $4, 2026-08-07). The high figures are pre-2026-04-27 targets that were never withdrawn. Only the three dated, post-collapse targets are used as anchors in the programme document.