Biotech Intelligence Briefing · Annexon · Vonaprument · Tanruprubart · 2027–2040

Nineteen events, one verdict — the market prices better-than-even odds on the DMC; we make it 42%.

ARCHER's vision signal (4 of 89 vs 15 of 89 patients losing 15 letters, strict definition) is real enough to test and too small to bank: it came from a secondary analysis after the primary missed [ga_p2_strict, ga_p2_primary]. ARCHER II is the fair test. Base odds: 42% that the DMC calls Month 15 met, 25% for US approval (it needs two sub-studies), 33% for the EU [ed_two_substudies]. Base-case per share: $5.69 if "met", $1.15 if "continue", $0.52 on futility; $2.96 weighted against $4.19. At spot the market implies ~68% odds on the model's economics, roughly the 55–65% argued in the post on X that started the debate (§02) [co_px_last].

Contact the author

Disclosure · The author holds no position in ANNX. Investment research, not investment advice.

Price as of publication · The spot price is the $4.19 close of 2 October 2026, the last close before this briefing was published on 3 October 2026. It is not updated afterwards, so every comparison with spot describes the market on that date.

ARCHER Phase 2 vision results are sponsor-presented secondary analyses; primary endpoint missed; no peer-reviewed efficacy paper. Q4 2026 is a DMC verdict with no patient-level data.

Company Annexon, Inc. · Nasdaq ANNX
Assets vonaprument (ANX007, GA) · tanruprubart (ANX005, GBS) · ANX1502 (oral, CAD)
Status ARCHER II DMC verdict Q4 2026 · GBS BLA targeted Q4 2026 · EU MAA under review
Engine model/annx.js · engine · loading · HANDOFF-ANNX 2026-10-03 · built from a9d931b (2026-10-05) · inputs 001afe2a5af9 · spot $4.19 (2026-10-02) · facts 212 (22 unverifiable)
Data vintage ARCHER II design 2026-09-25 · price 2026-10-02, close $4.19
Research cut-off 2026-10-03 · spot dated 2026-10-02 close
Author holds no position in ANNX
Scenario
Scenario differences
01

The DMC verdict: three states, one share price

Live for the active scenario. Probabilities come from the trial-odds engine (section 03); each state carries its own value per diluted share.
What Q4 2026 brings. One overall verdict from the independent data monitoring committee on the Month 15 primary: met, continue to Month 24, or futility. No patient-level data are released either way [ed_dmc_event, ga_a2_dmc]. The two sub-study results, which the US filing needs, follow in Q1 2027, and only if the DMC finds Month 15 met [ed_dmc_event, ed_two_substudies]. On "continue", the Month 24 analysis completes in Q3 2027 [ga_a2_m24].

Your odds, the model's economics

Move P(DMC met) with everything else held: value = P × "met" value + (1 − P) × value if not met (the continue / futility blend). The amber line is spot; the market-implied P(met) is where the line crosses it (§8.3).

Table: value per share by P(met)

All eight states

Not advice. Investment research for information only; not investment advice or a recommendation to buy or sell any security. Every probability is a model output under stated assumptions, and every value is an output of mostly [A] inputs, not a price target. Not medical advice. Clinical data are summarised for valuation context only. Research cut-off 2026-10-03; spot $4.19 is the 2026-10-02 close. The author holds no position in ANNX.
02

The X debate, adjudicated

Paraphrased claims, by role only. Evidence is the verified fact base; the verdict is in plain words.

The post that started it

In the run-up to the DMC verdict, an investor who had bought ANNX at $3.80 as a "lotto position" set out the bull case in a long post on X. Other investors replied and challenged it. On this page, "the OP" (original poster) means that post's author, and "the critic" the main respondent. Their claims are paraphrased, not quoted; the post is linked so that readers can check the paraphrase against the original.

The OP's case, in summary

  • Two discounted bets. ANNX is a GA bet and a GBS bet, both priced low.
  • GA, vonaprument. The bet is not on slowing lesion growth (about 6% in Phase 2) but on fewer patients losing 15 letters of vision: about 4–5 of 89 on monthly dosing against 15–19 of 89 on sham, depending on the definition. The proposed biology is that C1q drives photoreceptor and synapse loss in parallel with the atrophy that shows as lesion growth, a mechanism Annexon described before the Phase 2 data. Monthly dosing beat every-other-month dosing. ARCHER II's 659 patients give near-certain power at the Phase 2 effect, so the risk is that the effect shrinks.
  • GBS, tanruprubart. 30 mg/kg won the Phase 3 and 75 mg/kg missed, which the OP reads as a possible treatment window. The open questions are whether a trial weighted to the AMAN subtype in Asia carries over to AIDP-weighted Western patients, and whether the FDA accepts the bridge.
  • The OP's odds. GA: 55–65% at Month 15, 65–75% at Month 15 or 24. GBS: 60–65% on the current or bridged US path, 75–85% eventually.
  • The OP's price ladder. Both fail ~$1–2; GBS only $5–8; GA only $8–13; both succeed $15–25; a strong GA franchise plus GBS $25–35+. Insider purchases by the CEO (~$100K) and a director (~$3.3M) noted as alignment, not as a reason to own the stock.

The replies, in summary

  • The critic: the company fitted its theory to the data after the fact; the full distribution of vision change in Phase 2 was never shown, only responder cuts; sham event rates looked unusually high; the Month 15 / Month 24 alpha split is undisclosed, so the two looks cannot be given separate odds; and moving the endpoint from 12 to 15 months, then adding 24, signals low confidence.
  • The OP's answers: a timeline of Annexon's C1q photoreceptor statements back to 2018; the disclosed mean vision trend (dose-ordered but not significant), read as a tail effect; lampalizumab's Phase 3 sham rates as a benchmark; and event accrual, low dropout and over-enrolment as the reason for the timing changes. The OP agreed that the two looks are correlated.

The table below takes each claim in turn against the verified fact base. Section 10 sets the OP's odds and price ladder against the model.

truehalf right / supportivewrong or too highunproven / open

The sham-rate dispute: labels swapped, arithmetic consistent

The thread quoted 17.8% "foveal" and 25% "subfoveal". The deck has 25.0% foveal (11 of 44) and 17.8% non-foveal (8 of 45), lenient definition; they add to the 19 of 89 lenient sham events [ga_p2_foveal, ga_p2_lenient]. Strict: 10 of 44 + 5 of 45 = 15 of 89 [ga_p2_strict]. Post hoc, small n.

Verdict, in plain words

03

The odds, computed: prior, haircut, sham rate, alpha

Deterministic quadrature over the true effect, the sham rate and the two sub-studies, recomputed in your browser. Moving a control here sets a Custom case for the whole page.

Controls

Prior: a spike at "no effect" plus a slab around HR 0.70, updated with the Phase 2 strict monthly HR 0.272 (0.090–0.819) [ga_p2_hr]. Haircut: the share of the posterior log HR that carries into ARCHER II. Sham: shifts all four sham-grid values. Alpha: the Month 15 / Month 24 split is undisclosed [ga_a2_alpha]; the slider keeps α15 + α24 = 0.05, the presets include the correlation-adjusted 0.035 / 0.035 [A].

Live odds

Event tree, live

Conditional probability in the middle column, joint on the right. Sub-studies are tested at two-sided 0.05 each; "met" is the overall z beyond α15; futility is an observed effect in the wrong direction [A].

Prior and posterior on the Phase 3 effect

Table: posterior summary

Power grid at a fixed effect and sham rate (α15 live)

P(DMC met at Month 15) if the true HR and sham rate were known; P(both sub-studies | met) in small type. The model averages over this grid with the prior. Management's ">90%" power claim fits sham 18–20% with RRR ≥ 50%; it does not hold at 35% [ga_a2_power].

Analytic power at 2:1 (439:220, two-sided α 0.05, normal approximation)

Reproduces the coordinator's check [ga_a2_powercalc]. Each sub-study is half the trial; "both" is the square of "one".

Alpha-split scenarios (prior live)

The split moves the DMC headline more than the value: alpha shifts when success shows, not much whether.

computing…

P(met at Month 15) by prior and haircut

Rows: prior P(no effect). Columns: Phase 2 → 3 haircut. Sham prior and alpha as active. The cell nearest your setting is outlined.

computing…

04

ARCHER Phase 2, by arm and definition

Two event definitions exist. The toggle swaps the whole panel; numbers from the two are never mixed.

Primary missedGA lesion growth by FAF at Month 12

≥15-letter loss, definition

Patients losing ≥15 letters, by arm

Event rate by arm

More of the Phase 2 record

Table: ocular safety by arm (sham / monthly / EOM)
05

ARCHER II: what is known, what is not

Unknowns are drawn empty and named; the model's stand-in for each is [A].

Known design

Not disclosed

The timepoint moved three times

Masked pooled accrual is "in line with projections" [ga_a2_exec]; masked accrual cannot separate a working drug from a low sham rate.

    After the verdict

    06

    Sham-rate benchmarks and the model's prior

    The sham rate drives power. The amber ticks are the model's four sham-grid values for the active case (weights in the label).

    Share of sham eyes losing ≥15 letters

    Table: benchmarks with window, rule and source
    07

    The GA market: Syfovre and Izervay, and what vonaprument would have to win

    No GA drug has a vision claim. Revenue per injection is falling and the competitor is monthly too.

    Quarterly sales, 2026 ($M)

    Izervay $169M in April–June 2026, now ahead of Syfovre (~$162M) [gm_izv_leadership]. US class ~$1.3B annualised [A arithmetic].

    Table: annual sales (different fiscal calendars)

    Price per injection

    Launch list prices against today's Part B limits and Syfovre's net per commercial dose. The model's net price is the amber tick [validated].

    No vision claim anywhere

    Patients and injection burden

    Pipeline endpoints

    Market views

    Revenue build, live

    Table: risked revenue by year ($M)
    08

    Tanruprubart in GBS: one positive dose, one population, no IVIg comparison

    The Phase 3 ran in Bangladesh and the Philippines against placebo infusion plus supportive care.

    Phase 3 (GBS-02) by dose

    GBS-DS 3–5, onset ≤10 days, no IVIg or plasma exchange; 9 sites in the Philippines, 2 in Bangladesh [gb02]. Comparator: masked placebo infusion plus supportive care; no randomised comparison against or on top of IVIg/PE [gb18]. The 75 mg/kg dose missed. No peer-reviewed Phase 3 paper [gb17].

    Secondary and subgroup results (30 mg/kg vs placebo, nominal p-values)

    Subtype mix: trial vs Western GBS

    The trial is AMAN-dominant; Western GBS is mostly demyelinating [gb12, gb13]. The AIDP subgroup that looks strongest is 16 vs 18 patients [gb14].

    Table: subtype shares

    Ventilation: incidence vs duration

    The same number of patients needed ventilation in each arm; among them, time on the ventilator was shorter on drug [gb09, gb08].

    FORWARD · open-label, n=10 so far

    What the FDA has said

    Approval paths, live

    EU and US probabilities are [A] (section 16). The US-after-an-added-study path is shown as an option, not a probability: the engine drops it when its value is below the study's cost [A].

    Incidence, care and price anchors

    Competition

    09

    ANX1502: an option, not a forecast

    An oral classical-pathway inhibitor in cold agglutinin disease; patient count undisclosed.

    The record

    10

    The OP's numbers against the model

    The OP's odds on the model's economics, and the OP's price ladder against the model's states.

    Odds

    The OP's odds priced on the model's states

    The OP's scenario ladder vs the model's states

    The coordinator's notes keep only the "both fail ~$1–2" label; the other buckets are mapped by level.

    11

    Sum of parts: assets, costs, cash and debt

    Probability-weighted equity across the eight states, then shares. The Oxford loan is deducted at face plus the final payment.

    Equity bridge, $M

    Table: the bridge and the share count

    Oxford facility: the tranche ladder

    Balance sheet and share facts

    Insiders, short interest, price path, sell-side
    Not advice. The per-share value is the output of stated assumptions, most of them [A]; it is not a price target. Sell-side targets do not validate it. Not medical advice. Spot $4.19 = 2026-10-02 close; research cut-off 2026-10-03. The author holds no position in ANNX.
    12

    Catalysts, 2026 to 2027

    Solid: company guidance. Dotted: registry estimate. Dashed italic: model timing [A]. Faded: passed. ★ = moves the value most.

    13

    Failure-point monitor

    Live where the engine can read a parameter; hollow where an event is awaited. A pending trigger is never drawn green.
    14

    Which lever moves the share price?

    One at a time: each lever swung from its bear to its bull value with everything else at the active case. Not a confidence interval.

    computing…

    Table: lever values
    15

    Monte Carlo on value per share

    Each draw re-runs the trial odds, so this takes a few seconds per hundred draws in your browser.

    Click the button to populate the histogram · P10 / P50 / P90 appear here
    Table: histogram bins
    16

    Adjust the assumptions

    Every slider carries its source and a confidence flag: [H] hard datum · [M] management or filing · [A] analyst assumption · [validated] an [A] cross-checked against an independent benchmark. State serialises to the URL.
    17

    Method, self-tests and caveats

    Trial odds. Spike-and-slab prior on the ARCHER II hazard ratio: P(no effect) π0 (base 0.50 [A]) plus a slab with median HR 0.70 and SD 0.35 on log HR [A], updated with the Phase 2 strict monthly HR 0.272 (SE 0.5633 on log HR from the 95% CI) [ga_p2_hr]. The strict HR is used, not the lenient p=0.0021, because it is the more conservative definition the company now presents and a time-to-event HR maps onto the ARCHER II cumulative endpoint. The posterior slab is shrunk by the transfer haircut τ on log HR (base 0.85 [A]). The true HR is the spike plus the slab on a fixed 49-node grid (z = −6 … +6, step 0.25); the sham Month 15 rate takes four grid values (shifted by sham_shift); Month 24 sham = Month 15 × 1.6 [validated] [ga_bm_anegondi]; drug-arm risk = 1 − (1 − sham)^HR. Each arm is split into two exact halves (the sub-studies; definition undisclosed, halves [A]) [ga_a2_substudy]. In each half the observed Month 15 and Month 24 risk differences are bivariate normal with the covariance of cumulative events in the same patients; half-sum and half-difference are independent, so every state probability is a one-dimensional Gauss-Legendre integral. Rules: met = overall z at Month 15 beyond two-sided α15; futility = observed z < 0 [A]; continue otherwise, then Month 24 met at α24; each sub-study at two-sided 0.05. Quadrature error < 1e-9; the gap to the retired 400,000-trial patient-count Monte Carlo is within 0.009, from event-count discreteness, not quadrature.
    Regulatory layer. P(US) = P(US | both sub-studies) 0.85 × P(both) + P(US | one) 0.25 × P(one), with a 2-year delay on the one-sub-study path; P(EU) = 0.60 × P(met at either look) [A]. Base: 0.85 × 0.1937 + 0.25 × 0.3254 = 0.246; 0.60 × 0.5549 = 0.333.
    Valuation. Per state: GA value given its launch years and approval odds (US launch 2028 on the Month 15 path, +1 year on the Month 24 path, +2 on one sub-study; EU one year after US), peak US = patients on therapy × share × injections × net price; ramp to peak over ga_ramp years; loss of exclusivity 2040 with 60% then 35% retained; COGS, 2% Stanford royalty [co_stanford], SG&A; ex-US partnered at a royalty plus an upfront (base); taxed at 21% when positive (NOLs ignored); launch spend discretionary (state GA value floored at zero). GBS: EU, US on the current package, US after an added study (dropped when not value-accretive). ANX1502 a fixed option value. Corporate costs by state; in success states the engine also values a wind-down sale of GA rights at ga_sale_frac and takes the higher. Equity per state = GA + GBS + ANX1502 + corporate (cash est. 9/30 − debt at face − R&D − G&A). Discount (1 + r)^(t − 2026), horizon 2026–2045.
    Shares, raises, TSM. 204,444,933 common + pre-funded warrants and 2,388,473 RSUs [co_shares, co_rsu]. The pre-verdict ATM (base $75M) is priced at spot × (1 − raise discount) in every state (series convention) [co_atm]. The post-success raise (base $300M, US-path states only) is priced at the solved state value × (1 − discount): a company that hears "met" raises at the post-verdict price, one that hears "futility" does not raise. Options 23,190,315 at WAEP $6.40 by the treasury-stock method when the state value exceeds $6.40 [co_options]. Per share = Σ p(state) × value(state); event-conditional values average within each verdict group.
    Scenario weights. Bear 0.30, base 0.50, bull 0.20 [A]; they drive the scenario-weighted tile only, which uses preset values for every key the presets change.
    Self-tests. running…
    Unverified list — facts that matter but could not be verified; mentioned only with [unverified], never anchors

    Superseded — what was corrected before publication
    Could not be sourced — absences shown empty on this page, never imputed
      Caveat. [A]-flagged parameters are analyst assumptions; stress them before citing. Commercial share matters more than the trial prior: the DMC decides whether there is a product; share decides what it is worth. Revenue does not scale with the observed effect size in this model [A]. Model outputs are not forecasts, price targets or recommendations.
      Built 2026-10-03 · research cut-off 2026-10-03 · spot $4.19, 2026-10-02 close · Chart.js and the site badge are served from this site; the only external request is Google Fonts.
      18

      What this page does not claim

      Shown in full.
      19

      Disclaimers and compliance