IRON — Disc Medicine, Inc.

Company context · prepared 2026-08-20 · company sweep 2026-08-20 · USD · framework v5.11.0

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.

This is a full company-tier re-sweep, replacing the 2026-08-13 version. The delta window reported throughout is 2026-08-13 → 2026-08-20.

C.0 Company-tier coverage — CLEARED

One row per mandatory tool in the company-tier table in framework/02-connectors.md, in the order that file lists them, followed by its optional rows.

No mandatory row reads NOT CALLED, so the company tier is CLEARED. The three optional rows that were not called are optional by 02-connectors.md’s own table and do not bear on the gate.

C.1 What the company is

Disc Medicine is a clinical-stage biopharmaceutical company in Watertown, Massachusetts, working entirely on serious diseases of the blood [VERIFIED — BPIQ fetch_company_info 2026-08-20]. Its whole portfolio is built on two biological levers: how the body builds heme (the iron-carrying part of haemoglobin, the molecule that makes blood red and carries oxygen), and how the body controls where iron goes.

It runs four clinical programs. Bitopertin is an oral drug for erythropoietic protoporphyria, a rare inherited disease in which sunlight causes severe pain; it received a Complete Response Letter — a formal FDA refusal to approve on the current package — on 2026-02-13, and its Phase 3 APOLLO trial reads out in Q4 2026, with a Complete Response Letter response and an FDA decision now guided to mid-2027 [VERIFIED — exhibit 99.1 to the Form 8-K filed 2026-07-30]. Selcodebart (DISC-0974) is an antibody that lowers hepcidin to free up iron, aimed at anaemia in myelofibrosis and in inflammatory bowel disease. DISC-3405 is an antibody that does the opposite — it raises hepcidin to restrict iron — and is being tested in polycythemia vera and in sickle cell disease. DISC-0998 is preclinical. Two older eye programs under the GEM103 name are recorded as failed and discontinued.

The company has no revenue of any kind. It is not a single-asset company: bitopertin dominates the equity story today, but three of the four clinical programs carry a catalyst inside the next eighteen months, on two independent mechanisms.

Share price on the sweep date was $79.51, and that figure needs a warning label. It is the value carried in data/prices/IRON.json for 2026-08-20, read through lib/prices.mjs’s closeOnOrBefore [VERIFIED — data/prices/IRON.json, read through lib/prices.mjs 2026-08-20]. It is not a closing price: it is a partial intraday bar, written while the session was still open, on 106,841 shares against a typical full-day volume of roughly 335,000 (C.4). BPIQ read a last price of $78.92 on 101,303 shares the same day [VERIFIED — BPIQ fetch_company_info 2026-08-20] — a second intraday reading, 0.7% lower. The last complete daily bar in the cache that is free of the defect described in C.8 is 2026-08-18 at $80.20. Every figure below that needs a spot price uses $79.51, because that is what a settlement replayed against the committed cache will see; the 0.7% uncertainty around it is real and is stated here rather than buried.

C.2 Pipeline — every program

Every row returned by fetch_company_drugs on 2026-08-20, including the discontinued and failed ones. The catalyst is reported as catalyst_date_text; the “to a day?” column is yes only when that text names a day (01-rules.md rule 23). Not one of the nine rows names a day, so every catalyst_date on this ticker is a synthesized period-end placeholder.

Five rows carry has_catalyst: true. All five contribute a date range to the catalyst-clustering computation in the program document’s Attribution section, whether or not they have a document of their own.

C.3 Financial position

A named correction carried from the previous version. The 2026-08-13 sweep gave the recomputed runway as “about 2030-09” and “about 2029-08”. Those dates counted the 48.7 and 36.2 months forward from the sweep date rather than from the 2026-06-30 balance-sheet date the cash figure belongs to, which double-counts six weeks of runway the company had already spent. The month counts were right; the dates were about two months too generous. Corrected above. Nothing downstream depended on it — every reading still clears every catalyst in C.2 by years — but the error is recorded rather than silently fixed.

Reading the runway rows together. Three answers exist and they do not agree: the company says “into 2029”, BPIQ’s own cash_runway_end_date says 2030-06-30, and the arithmetic gives anything from 2029-07 to 2030-07 depending on which burn number is used. The gap is not a contradiction — it is the difference between cash actually leaving the building and the accounting loss, and between today’s spend and the higher spend a company running two programs into pivotal trials in 2027 expects. The conservative reading, the company’s own, is used everywhere below. On any of the three readings the runway clears every catalyst in C.2 by years. This company does not need to raise before a readout, which is the question the row exists to answer.

Dilution history. Every raise in the last 24 months, plus the June 2024 offering, which falls just outside that window but is retained because it is the same-day counterpart to this ticker’s single most informative past catalyst (C.7). Every line below was re-read from the Form 10-Q filed 2026-07-30 for this sweep, replacing the FY2025 10-K as the source; all figures agree.

Is a raise likely before the next catalyst? No, and the answer is unusually clean here. The company holds $717.7M against a stated runway into 2029 and a monthly cash burn of $14.7M. It has not filed a registration statement in 2026. Its remaining capacity is real but idle: the Cantor at-the-market facility runs to $200.0M under a Form S-3 shelf filed in August 2024, of which $10.1M gross has ever been drawn, leaving roughly $190M undrawn and nothing at all drawn in the first half of 2026 [VERIFIED — Form 10-Q filed 2026-07-30, "At-the-Market ('ATM') Program"], and the company extended rather than drew on its Hercules credit facility in June 2026. The one dilution mechanism that fires without warning is therefore the at-the-market facility, and this company has used it for $9.8M in two years while raising $626.9M net through three underwritten offerings — it is not how Disc raises money.

Two live share counts, both confirmed afresh this sweep. 38,392,844 common shares were outstanding at 2026-07-23 [VERIFIED — EDGAR XBRL dei:EntityCommonStockSharesOutstanding, 10-Q cover page], and 38,345,666 at the 2026-06-30 balance-sheet date [VERIFIED — 10-Q note 9]. Separately, 445,422 pre-funded warrants remained unexercised at 2026-06-30 — 59,523 from the October 2025 offering, 181,818 from January 2025 and 204,081 from June 2023, each confirmed individually in the 10-Q with the words “none of the … pre-funded warrants had been exercised”. A pre-funded warrant is a share that has already been paid for and can be converted for essentially nothing, so it is economically a share today. The economic share count is therefore 38,838,266, and that is the denominator used for cash per share below rather than the reported count. The previous version took this figure from the FY2025 10-K at 2025-12-31; the 10-Q confirms the same three tranches six months later, so the figure is unchanged and now rests on a fresher source.

C.4 Valuation frame

Average daily volume, because a run-up driver reads it. Over the 30 complete trading days ending 2026-08-18, average share volume was 335,063 and average dollar volume $25.77M [VERIFIED — computed in node from data/prices/IRON.json, read through lib/prices.mjs]. The comparable figures on 2026-08-13 were 409,543 shares and $31.50M. Volume has cooled by about a fifth in a week, and that matters mechanically rather than rhetorically: the run-up priority formula buckets average dollar volume at $30M, so the fall crosses a threshold. The window ends 2026-08-18 rather than today on purpose — the 2026-08-13, 2026-08-19 and 2026-08-20 bars carry partial intraday volumes (C.8) and would drag any average that included them.

What the range is made of, because both ends were set by one program. The $40.00 low is the intraday low of 2026-02-13, the session the FDA’s Complete Response Letter on bitopertin was disclosed; the shares closed that day at $55.95, so the low is a wick rather than a level anything held at. The $99.50 high is the intraday high of 2025-12-08, two days after positive Phase 2 data on selcodebart at the American Society of Hematology meeting. Both ends therefore belong to programs other than the one analysed here, which is the first quantitative statement of this ticker’s materiality structure and is why the program document does not reach for either end as a scenario anchor without saying so.

Cash per economic share: $18.48 ($717.746M ÷ 38,838,266). Net of the $59.3M Hercules debt, $16.95. Spot is 4.3× cash per share, so cash is nowhere near a floor on this equity — a fact worth stating plainly, because on most tickers in this corpus it is the single most load-bearing downside anchor and here it does nothing at all.

C.5 Ownership and flow

The short position is large, it grew fast into the readout, and this sweep adds no new reading. Across the four most recent published FINRA settlements the short count went 5,668,145 (2026-05-15) → 5,999,027 (2026-05-29) → 6,010,116 (2026-06-15) → 6,363,314 (2026-06-30) → 7,290,442 (2026-07-15) → 7,394,606 (2026-07-31): up 41% between mid-April and the end of July, with the largest single jump (+14.6%, +927,128 shares) in the first half of July. Days-to-cover of 20.9 means that at the average daily volume FINRA measured, closing the whole short position would take roughly a month of trading. The 2026-08-14 settlement has not published yet — FINRA runs about two weeks behind, so it should appear around 2026-08-26 — which means nobody can yet see whether the shorts kept adding through the first half of August. That is a stated blind spot, not an inference of stability. BPIQ’s own short_float field says 22.4% (it said 22.8% last week), which sits close to the FINRA-derived 20.9% of estimated float rather than contradicting it — but the FINRA figure is the one used here, because its units are known and BPIQ’s are not.

Insider flow, read from EDGAR because the connector could not see it. BPIQ’s insider endpoint returned zero rows on three attempts (C.0). The EDGAR feed for 2026 now contains 36 Form 4 filings and 20 Form 144 notices, against 34 and 19 at the previous sweep. The three new Form 4s were read in full and every one is the same shape — an option exercise (transaction code M) immediately followed by a sale (code S), under a Rule 10b5-1 plan adopted in March 2026 [VERIFIED — EDGAR Form 4 filings 0001193125-26-355841, 0001193125-26-352823 and 0001193125-26-340543, read 2026-08-20]:

The distinction rule 25 asks for: this is routine, not conviction — and the plan dates are what prove it. A chief executive selling $2.6M of stock three weeks before a guided data readout looks like a signal and is not one: the sale was executed under a trading plan adopted on 2026-03-12, five months earlier and before the readout was even pulled forward into Q3. A plan adopted in March cannot express a view formed in August. That is the whole reason Rule 10b5-1 plans exist, and it is why this row is recorded as noise rather than evidence.

What remains genuinely informative is the absence on the other side. There is not one open-market purchase by any insider anywhere in the 2026 filing record, across 36 Form 4s. At a price 20% below the 52-week high and days away from a Phase 2 readout the company pulled forward, that absence is a small negative signal, and it is recorded here as one — the same reading the previous version gave, now with two more weeks of filings behind it.

Concentration. Thirteen hedge funds hold the name and BPIQ classifies it as a “pick” on that basis [VERIFIED — BPIQ fetch_company_info hf_holding_count 13, in_top10_hfs true, 2026-08-20]. FMR LLC filed a Schedule 13G/A on 2026-08-06 [VERIFIED — EDGAR submissions feed], reported in the press feed as a 12.8% position [WEB ESTIMATE — press-feed summary of the 13G/A, 2026-08-06; the filing itself was not read this sweep]. The press feed over the delta window also carries three routine institutional-position reports — GSA Capital Partners opening a $1.19M position (2026-08-18), Sei Investments opening a position (2026-08-13) and Manufacturers Life reducing (2026-08-10) — all 13F-derived aggregator copy, none large enough to move the ownership picture, and none read from a filing. No large holder was identified as exiting.

C.6 Options chain and its readability

Plain-English takeaway first. This chain is unusually readable for a company this size — there are strikes above and below spot, real open interest, and implied volatility values that are not sitting on the connector’s floor or ceiling artifact. And what it says is the single clearest statement of materiality on this page, unchanged in a week: the market is pricing the December event at roughly double the volatility of the September one. Implied volatility for at-the-money options runs about 45% for the 2026-09-18 expiry, 54% for 2026-10-16, and 97–99% for 2026-11-20 and 2026-12-18. The only thing that changes between October and November is that the November and December expiries contain bitopertin’s Q4 APOLLO topline. Traders are pricing DISC-3405’s Q3 readout as a real but ordinary event and APOLLO as the one that decides the equity.

What did change is the size of the near-term move. The 2026-09-18 at-the-money straddle has fallen from 12.0% of spot to 10.4% in seven days, on a front implied volatility that eased from 46.4% to 45.4% and a week less time to expiry. The market is pricing a smaller move into mid-September than it was, not a larger one.

C.7 Reaction to past catalysts

Every row from fetch_company_historical_catalysts, each cross-checked against the same-day press feed and against the price cache. No new row has appeared since the 2026-08-13 sweep — the newest catalyst in the feed is still 2026-06-12 — so this table is a re-verification rather than a revision. The “intraday move” column reports BPIQ’s own intra_day_price_change_percent; the “read” column reports the close-to-close move computed from data/prices/IRON.json, because on this ticker the two frequently disagree and the close-to-close figure is the one a position would actually have realised.

The pattern across the record, stated once because every scenario range in the program document leans on it. Three things are true here at the same time. First, BPIQ’s intraday figure is directionally wrong on this ticker more often than it is right — it disagrees in sign with the close-to-close move on five of the nine usable rows, so it is never used below. Second, this company reliably sells into its own good news: three of the largest positive intraday prints (2025-10-20, 2025-01-21, 2024-06-14) coincide with an equity offering priced off that same day’s strength, and a fourth (2024-12-09) simply faded. Third, and most important for the program below, only bitopertin has ever moved this stock more than 10%. Every hepcidin-pathway readout in the record — three separate positive datasets on DISC-0974 — produced a move between −5.5% and +0.02%. The one exception is 2024-06-14, and that exception is the molecule analysed here.

C.8 Company data-quality flags

• The committed price cache carries three defective bars, and one of them was locked against. scripts/fetch-prices.mjs never overwrites an existing non-null value, so a bar first written mid-session stays an intraday snapshot for ever. On this ticker that has happened three times: 2026-08-13 is stored as $79.50 on 66,796 shares, while today’s fetch reported $80.01 on 341,000 shares for the same date and discarded it; 2026-08-19 is stored as $80.58 on 79,634 shares against a discarded $80.32 on 457,200; and 2026-08-20, today, is a partial bar by construction at $79.51 on 106,841 shares. The 2026-08-19 case is independently confirmed: BPIQ’s prvious_close field for 2026-08-20 reads $80.32, matching the discarded value and not the stored one. The consequence for this ticker’s own record is specific: the $79.50 “close” the 2026-08-13 analysis locked at was an intraday price, not that day’s close of $80.01. Every company figure the previous version derived from it — market capitalisation, enterprise value, the 52-week position — was therefore built on a price 0.6% below the real one. All of them are recomputed above from today’s reading, and today’s reading carries the same defect, which is why C.1 and C.4 label it rather than presenting it as a close.
• max_52_week_position fires the previous-close bug on this ticker, again. BPIQ reports 0.67765. ($80.32 − $40.00) ÷ ($99.50 − $40.00) = 0.67765 exactly, and $80.32 is its own reported previous close. The last-price computation gives 0.66403. C.4 derives the figure from the price cache instead, so the bug is designed around rather than corrected for. Second consecutive sweep on which it has fired here.
• market_cap fires the same bug, on a current share count. $3,083,713,280 ÷ 38,392,844 EDGAR-filed shares = $80.32 to the cent, again the previous close. This is the MNOV failure mode (stale price, current shares), not the CNTB one (current price, stale shares) — the two are distinguishable here only because the EDGAR share count was fetched independently.
• enterprise_value has become nearly reconcilable, and the change is worth recording. On 2026-08-13 BPIQ’s $2,348,507,828 against its own market cap implied subtracting $745,571,404, which matched no balance-sheet figure this repository could identify. This sweep its $2,426,445,301 against its own $3,083,713,280 market cap implies subtracting $657,267,979, against an actual net cash position of $658,406,000 (cash $717.746M less debt $59.340M) — an unexplained residual of $1.14M, or 0.05%, where a week ago there was an $87M hole. The most likely reading is that BPIQ has picked up the June 2026 debt drawdown; that is an inference, not a fact BPIQ states. Either way the field remains wrong on its price basis (previous close, as above), so C.4 still recomputes rather than quotes it.
• fetch_company_insider_transactions returned zero rows on three attempts while 36 Form 4s sit in the 2026 EDGAR feed, three of them filed inside the delta window. Second consecutive sweep on this ticker, and the fourth ticker overall after TLSA and CNTB. The call completes without error, so nothing announces the gap. Flow was read from EDGAR, and the three new filings were parsed individually.
• tute_own is unreadable and was discarded. 1.14916 is either 114.9% of shares outstanding or 1.1%; neither is credible. It has moved from 1.14464 in a week, so the field is live and still uninterpretable. C.5 records the gap rather than picking a reading.
• The press feed still carries misdated aggregator copies of a 2025 financing as 2026 events. The two MSN items — “Disc Medicine announces public offering of common stock and pre-funded warrants” (2026-03-20) and “Disc Medicine prices upsized securities offering to raise total $250M” (2026-04-26) — describe the October 2025 offering and are still present in the feed. There is no 424B5, S-3 or S-1 anywhere in the 2026 EDGAR feed, and the share count moved only from 38,160,065 (2026-02-19) to 38,392,844 (2026-07-23), which no $250M raise could produce at these prices. Taken at face value these two items would have put a phantom 2026 raise into the dilution table and into every scenario anchor built on it. The EDGAR filing feed is what settles it, and the press feed alone would not have.
• A recurring plaintiff-firm solicitation is in the feed and there is no litigation behind it. GlobeNewswire items headed “If You Have Suffered Losses in Disc Medicine, Inc. … Contact The Rosen Law Firm” appear on 2026-04-03, 2026-04-04, 2026-05-26, 2026-06-03, 2026-06-10 and 2026-08-15 — six times since the February 2026 Complete Response Letter. The 2026-08-15 item is new inside this delta window and reads, at a glance, like a fresh legal event. It is not. The Form 10-Q filed 2026-07-30 states verbatim that the company “was not subject to any material legal proceedings during the six months ended June 30, 2026 and 2025 and, to the best of its knowledge, no material legal proceedings are currently pending or threatened” [VERIFIED — Form 10-Q filed 2026-07-30, commitments and contingencies note]. These are advertisements soliciting a class that does not yet exist. Recorded because the previous version did not carry them and a reader meeting the 2026-08-15 headline cold would reasonably misread it.
• BPIQ carries duplicate rows on both sides of this ticker. The same discontinued GEM103 asset appears as two pipeline rows (13662, 14204) under two indication strings; the 2024-12-09 ASH presentation appears as two historical-catalyst rows with identical figures (4411, 4423); the 2026-06-12 EHA session appears twice (7865, 7867); and the National Priority Voucher appears twice (6003 dated 2025-10-20 with figures, 5771 dated 2025-10-19 with nulls). Any count taken off row counts on this ticker is inflated.
• The 2026-02-13 Complete Response Letter is absent from fetch_company_historical_catalysts entirely. The largest single-day move in this company’s history — −21.9%, twelve times normal volume, and the session that set the 52-week low — is not in the catalyst feed at all. It was found in the press feed and confirmed against the price cache. A reaction history read only from the catalyst connector would have missed the most important row in C.7.
• FINRA short interest is one settlement stale through no fault of the connector. The newest published settlement is 2026-07-31; the 2026-08-14 settlement publishes around 2026-08-26. Short interest through the first half of August is therefore unobservable at this sweep, and C.5 says so rather than treating the July figure as current.