CNTB — Connect Biopharma Holdings Limited

Company context · prepared 2026-08-13 · company sweep 2026-08-13 · USD · framework v5.6.1

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. Two optional rows are recorded honestly: one returned nothing, one was refused. The regulatory tier was not called at all, which is correct rather than a gap — 02-connectors.md makes that tier conditional on a regulatory catalyst (a PDUFA date, an advisory committee, or another decision by a regulator), and the only catalyst analysed under this ticker is a clinical-trial readout.

C.1 What the company is

Connect Biopharma Holdings Limited is a clinical-stage biopharmaceutical company. It has no approved product and no product revenue; its only revenue is milestone and cost-sharing income from a licensing partner, which was $2.8 million in the second quarter of 2026 [VERIFIED — exhibit 99.1 to Form 8-K, 2026-08-12]. It is headquartered in San Diego, California, having moved its centre of gravity from Taicang, China, and it lists nine programs in the third-party pipeline feed [VERIFIED — BPIQ fetch_company_drugs 2026-08-13].

It is effectively a single-asset company. Five of the nine pipeline rows are one molecule, rademikibart (formerly CBP-201), an antibody against the interleukin-4 receptor alpha subunit. Of the four non-rademikibart rows, two are the same shelved molecule (icanbelimod, one of them marked Failed Discontinued), and the remaining two are a discontinued in-licensed compound and a Phase 1 asset with no update since 2022. The company’s own description of itself is narrower still than its pipeline: “Connect Biopharma is a clinical-stage biopharmaceutical company dedicated to transforming care for asthma and COPD” [VERIFIED — exhibit 99.1 to Form 8-K, 2026-08-12].

Share price on the day of this sweep: $2.33 [VERIFIED — BPIQ fetch_company_info 2026-08-13]. This is the spot price every program document under this ticker measures its event against.

C.2 Pipeline — every program

Every row returned by fetch_company_drugs on 2026-08-13 is listed, including the two marked Failed. The catalyst column reports catalyst_date_text exactly as the feed discloses it; the “date to a day?” column is yes only when that text names a day (01-rules.md rule 23). No row on this ticker names a day.

What the shut programs say about the company. Two rows are marked Failed, and three more are alive only on paper. That is a company that has narrowed hard rather than one that has run out of ideas: the rhinosinusitis study was terminated for pandemic operational reasons, the pruritus in-licence was handed back “due to strategic prioritization”, and icanbelimod was parked while the company looked for a partner. The consequence for a reader is arithmetic rather than moral — with five of nine rows on one molecule and only one dated readout pair, the equity carries almost no diversification against the September 2026 result.

C.3 Financial position

The two runway figures disagree by roughly eight months, and this document does not resolve the disagreement. Both are reported because both are true statements about different things. The company’s “at least one year” is a forward statement about a plan; the 5.9-month recomputation is backward-looking arithmetic that assumes the first half of 2026’s spending rate simply continues. The reconciling facts, stated so a reader can judge for themselves rather than take either figure on faith:

• Research and development expense was $16.3 million in the second quarter of 2026 alone, against $8.8 million in the same quarter of 2025, and the company attributes the increase to the two Seabreeze STAT studies [VERIFIED — exhibit 99.1 to Form 8-K, 2026-08-12]. Both studies finished enrolling in June 2026 and both reached study completion by 2026-08-10, so the largest single line in the burn is winding down by construction, not by intention.
• Against that, the company started a third study — Seabreeze STAT IV, 40 participants — in August 2026 [VERIFIED — same release; NCT07705737], which puts some spending back.
• The partner milestone stream is real but not scheduled: up to approximately $99 million of remaining development, regulatory and commercial milestones, plus tiered royalties up to low double-digit percentages on Greater China net sales [VERIFIED — same release]. $2.8 million of it landed in the second quarter of 2026. None of it is committed to a date.

Which figure to use for what. For the question “does the cash reach the September 2026 readout?”, both readings answer yes with months to spare, so the answer does not depend on the disagreement. For the question “when is the next raise?”, the two readings differ materially, and the program document’s run-up section treats the runway as tight rather than comfortable for that reason — the conservative reading of two that disagree.

Dilution history. One raise in the last 24 months, and no withdrawn offering.

The share count corroborates that this is the only raise: 55,349,208 shares at 2025-03-21 rose only to 56,521,282 by 2026-03-26 — routine equity compensation — then jumped to 62,963,853 by 2026-04-30, which the 6,130,000-share placement almost exactly accounts for [VERIFIED — EDGAR XBRL dei:EntityCommonStockSharesOutstanding, four filings].

Two share-registration statements are live, and they are different animals. Getting this distinction right matters, because one is dilution capacity and the other is supply overhang:

1. A $300,000,000 universal shelf on Form F-3, filed 2025-06-10, amended by a post-effective amendment on 2025-09-10 and effective 2025-09-12, carrying forward unsold capacity from a 2022 registration (file 333-264340) [VERIFIED — EDGAR F-3 filing-fee exhibit, accession 0001193125-25-138482]. This is primary issuance capacity: the company can sell newly issued shares off it. Whether it can sell the full amount in the next twelve months is a separate question this sweep did not settle — a registrant whose non-affiliate public float is under $75 million is limited to one third of that float per year, and CNTB’s float sits near that boundary [UNVERIFIED — the affiliate share count was not confirmed, so the limit was not tested].
2. A resale registration on Form F-3 for exactly 6,130,000 ordinary shares, filed 2026-05-15 and effective 2026-06-01 [VERIFIED — EDGAR F-3 filing-fee exhibit, accession 0001193125-26-225436, which registers 6,130,000 shares at an assumed $2.37]. The number is the private placement’s own share count, so this registers the March 2026 buyers’ shares for resale rather than creating new ones. It is not dilution. It does free 9.7% of the shares outstanding to be sold into a stock that trades roughly $600,000 a day (C.6), which is a real overhang.

Is a raise likely before the next catalyst? Probably not, and probably soon after. The cash reaches September on either runway reading, the company has just asserted a twelve-month runway in a filed exhibit, and raising immediately before a binary readout would be an unusual choice. Raising into a positive result, off a shelf that is already effective, would be the ordinary one.

C.4 Valuation frame

Why this row was computed by hand rather than from the price cache. 03a-company-spec.md C.4 normally derives the 52-week range and the position within it from data/prices/<TICKER>.json with lib/prices.mjs. That file does not exist for CNTB — the committed cache covers twelve tickers and this is not one of them. C.4 states the exception for exactly this case (“Recompute … by hand only for a ticker the cache does not cover”), so the range is taken from BPIQ’s own fifty_two_week_high / fifty_two_week_low and the position is computed with positionInRange, the same library function the cache path would have called on its last step. Creating the cache is scripts/fetch-prices.mjs’s job and writes outside this run’s lane; it is recorded as an open item in C.8 row 8.

One incidental finding: BPIQ’s max_52_week_position of 0.42471 agrees with the last-price computation here to five decimal places, not with the previous-close computation (($2.24 − $1.23) ÷ $2.59 = 0.390) that 02-connectors.md’s documented bug would predict. The field is not read either way, but the bug did not reproduce on this ticker.

C.5 Ownership and flow

Short interest has nearly doubled into the readout window. Four consecutive settlements: 1,061,303 shares (2026-06-15) → 1,327,642 (2026-06-30, +25.1%) → 2,062,435 (2026-07-15, +55.4%) → 2,060,037 (2026-07-31, flat) [VERIFIED — FINRA consolidatedShortInterest, three calls]. The jump lands in the fortnight after the 2026-06-17 announcement that put a September date on the readout, which is when a short position ahead of a binary event becomes a dated trade rather than an open-ended one. Days-to-cover moved from 2.12 to 8.02 across the same four settlements, mostly because average daily volume fell from 627,641 shares to 256,793 rather than because the position grew — a thinning tape under a growing short position.

Insider flow: genuine open-market buying, read from filings rather than from the connector. BPIQ’s insider connector returned zero rows on two attempts, so the flow below comes from the Form 4 XML on EDGAR. Three transactions in the most recent window, and the distinction between them is the point:

[VERIFIED — EDGAR Form 4 XML, accessions 0001193125-26-138012, 0001193125-26-253607, 0001835268-26-000022]

Both purchase prices sit above the $2.33 spot on the day of this sweep — $3.45 is 48% above it and $2.4774 is 6% above it. That is one director, so it is concentration rather than consensus, and a director’s own view is not evidence about the trial. It is, however, the cleanest available statement that inside money has been buying rather than distributing, and there is no offsetting open-market insider sale in the window.

Fund positioning: four holders, two of them new, one adding. From the most recent filing block: Ikarian Capital 5,388,486 shares (up 600,000 from the prior block), ADAR1 Capital Management 742,400 (new), Perceptive Advisors 792,696 (unchanged), Stonepine Capital Management 310,000 (new). Together 7,233,582 shares, or 11.5% of the shares outstanding [VERIFIED — BPIQ fetch_company_hedge_fund_holdings 2026-08-13, share counts only]. Independent corroboration for one of them: the press feed records “ADAR1 Capital Management LLC Invests $1.95 Million in Connect Biopharma” on 2026-07-28, and $1,945,000 is the exact value in the block, so that holding is a 2026-06-30 position reported in late July [VERIFIED — press feed 2026-07-28 cross-checked against the connector's own value].

Concentration versus diffusion, and one exit. Ownership is concentrated and strategic rather than diffuse: Schedule 13D and 13G filings over the last year name Wubin Pan and BioFortune at 10.8%, an insider-linked 9.6% stake, Wei Zheng at 6.2%, and Panacea adding 1.16 million shares [VERIFIED — EDGAR submissions and press-feed filing headlines, 2026-02 to 2026-05]. Against that, BML Capital Management sold 554,058 shares (2026-02-23) and a further 858,613 (2026-05-06), and BioFortune filed a Form 144 notice to sell 500,000 shares on 2025-11-17 [VERIFIED — press-feed filing headlines]. One holder distributing while a director buys and two new funds arrive.

One passive flow worth naming because it is not a view on anything: CNTB was added to the Russell Microcap Index on 2026-06-29 [VERIFIED — press feed 2026-06-29], which mechanically adds index holders and helps explain institutional ownership without implying anyone chose the name.

C.6 Options chain and its readability

Plain takeaway: the chain cannot price this event, and it cannot even bracket it tightly. There is no listed strike near the $2.33 spot — the ladder is $2.50, $5.00 and $7.50, so the increment is larger than a 7% move in the stock and there is no strike below spot at all. Total volume across every September contract on the read date was 22 lots. The put side of the money is quoted 0.55 bid against 3.50 ask, a six-fold spread that is not a market. Any figure derived from this chain is an artifact of where the quotes happen to sit, not a statement about expected volatility, and the independent Yahoo cross-check that would normally test that judgement was refused this sweep (C.0).

The bracket, since a point estimate is not available. The 2026-09-18 $2.50 call is quoted $0.35 bid / $0.60 ask on a $2.33 stock. Paying 15% to 26% of spot for a five-week call struck 7% above the money is consistent with an expected absolute move somewhere in the region of 35% to 65% around the event [UNVERIFIED — derived from a chain this same section records as unusable, and offered only as a bracket]. The program document treats this as a sanity check on its own scenario ranges and never as an input to them.

C.7 Reaction to past catalysts

Intraday moves from fetch_company_historical_catalysts, every one cross-checked against the press-release feed for the same date. One row overturns the naive reading entirely.

What this table supports and what it does not. It supports a range of roughly +8% to +13% for a clean, positive, data-driven single-day move on this ticker — but every one of those moves was a presentation of data already known to be positive, never a blind binary readout. It contains no clean example of a negative surprise: the one large decline is a bundled placement. A reader should therefore treat the downside side of this table as unpopulated rather than as evidence that this stock does not fall on bad news.

C.8 Company data-quality flags

1. market_cap is computed off a stale share count, not a stale price. BPIQ reports $146,706,000, which divided by the last price of $2.33 implies 62,963,853 shares — exactly the 2026-04-30 count from the first-quarter 10-Q, one quarter old. This is not the previous-close bug 02-connectors.md documents (that would have used $2.24 and produced a different figure). Recomputed in C.4 on the EDGAR-filed 62,974,309 at 2026-07-31. The correction is small in this instance ($146.73M against $146.71M) and the point is the mechanism, not the size.

2. enterprise_value cannot be reconciled with any cash figure. BPIQ reports $90,031,283, which against its own market cap implies subtracting $56.675M of cash. That matches neither its own cash field ($27.015M), nor cash plus short-term investments ($31.486M), nor either figure at any prior quarter-end (2026-03-31: $46.034M; 2025-12-31: $44.342M). Discarded and recomputed in C.4.

3. max_52_week_position did not reproduce the documented bug on this ticker. BPIQ’s 0.42471 agrees with the last-price computation, not the previous-close one. The field is not read regardless — see C.4 — but the finding is recorded because it means the bug is not universal.

4. short_float is unreadable and was discarded. BPIQ reports 0.10992. Against the authoritative FINRA count that is 3.4× too high whether read as a share of the 62.97M shares outstanding (3.27%) or of the estimated non-affiliate float (4.80%). C.5 uses FINRA divided by the EDGAR count, whose units are known.

5. fetch_company_insider_transactions returned zero rows on two attempts while six Form 4s are on file. This is the bug 02-connectors.md records against TLSA, now reproduced on a second ticker — so it is a connector property rather than a TLSA quirk. The call completes rather than erroring, so the state is CALLED with an empty result, and the flow in C.5 is read from EDGAR Form 4 XML instead.

6. fetch_company_earnings_transcript returned no output. Optional row, no consequence: the second-quarter release was read in full from the SEC as exhibit 99.1 to the 2026-08-12 Form 8-K, which is a stronger source than a transcript anyway.

7. fetch_company_hedge_fund_holdings block labels could not be resolved to a quarter-end close, and the CGEM filing-quarter bug did not reproduce as described. 02-connectors.md says to divide value by shares and match the result against the quarter-end close. Here the four funds in the “2026Q2” block give $2.619, $2.619, $2.619 and $2.434 — three agreeing and one not, so a single quarter-end close does not fit the block. Independently, ADAR1’s $1,945,000 value matches a holding publicly reported on 2026-07-28 as a 2026-06-30 position, which makes “2026Q2” the holdings quarter on this ticker, not the filing quarter the CGEM finding describes. The labelling convention is therefore not consistent across tickers. C.5 uses share counts, which are internally consistent, and does not use the dollar values.

8. The committed price cache now exists, and it disagrees with BPIQ about the spot price. Amended 2026-08-13 by the rademikibart-acute-copd-exacerbation run; the original text recorded that data/prices/CNTB.json did not exist. It exists now — 1,255 daily bars from 2021-08-13 to 2026-08-13, fetched by scripts/fetch-prices.mjs, source Yahoo Finance chart API v8 — so the consequence this row previously carried forward, that no run-up settlement was possible on this ticker, no longer holds.

   Two findings from the cache, neither of which changes a figure elsewhere in this document:

   • It confirms C.4’s 52-week range exactly. lib/prices.mjs range52w over the cache, as of 2026-08-13, returns a low of $1.23 and a high of $3.82 — the same figures C.4 took from BPIQ’s fifty_two_week_high / fifty_two_week_low. The hand computation C.4 performed under its own stated exception was correct.
   • It disagrees with BPIQ on the spot price for this very day, and the disagreement is left open. The cache’s bar for 2026-08-13 opens at $2.33, reaches $2.64 and closes at $2.56. C.1’s $2.33 is BPIQ’s last_price read mid-session during this sweep, which is the day’s opening print rather than its close. C.1 and C.4 are deliberately left unchanged: the rademikibart-acute-asthma-exacerbation program locked its prediction against $2.33, and rewriting a figure a locked record cites would corrupt that record rather than correct it. The rademikibart-acute-copd-exacerbation program locks against the $2.56 close instead, and says why in its Market and timing § Spot: 05-prediction-protocol.md requires a run-up settlement to read the committed cache and never a fresh fetch, so an entry price absent from that cache could never be settled against it. Neither figure is averaged and neither is picked over the other. Consequence a reader must carry: the two program documents under this ticker are locked against different spot prices on the same day, so their percentage figures are not directly comparable.

9. Drug names are dirty, as documented. One molecule appears as "Rademikibart /CBP-201 " (id 13433, trailing space), "Rademikibart /CBP-201" (id 18937), "Rademikibart (CBP-201) (Th2 cell modulator)" (ids 16030 and 14215), "Rademikibart (CBP-201) (Th2 Cell Modulator)" (id 19715) and "Rademikibart (CBP-201) (IL-4Rα Antibody)" (id 19716). Every reference in this repository keys on the integer id.

10. Two rows carry has_catalyst: true against a delivered event. Ids 19715 and 19716 are both flagged as having a catalyst with catalyst_date: null and catalyst_date_text: "TBA", but the Phase 1 intravenous data they describe were reported on 2026-03-30 and reiterated on 2026-05-12. They look like stale flags on an event that has already happened. They contribute no window to the catalyst-clustering computation (a null date yields no range), so they do not distort the attribution finding in the program document — but a reader counting “four catalysts on this ticker” off the feed would be counting two that no longer exist.

11. Source conflict left open: the runway. The company’s filed “at least one year” and this document’s 5.9-month recomputation are both in C.3, unreconciled and both tagged. Neither is averaged and neither is picked.