# WITHDRAWN 2026-08-10: V1-M Polymarket data bundle

**`v1m-polymarket-bundle.tar.gz` has been withdrawn and removed.** This file is
the notice that replaces it. Nothing here is a redirect to a newer version;
there is no newer version.

**Withdrawn artifact:** `v1m-polymarket-bundle.tar.gz`
**sha256 (as published):** `f22d9829737eaf155bed68d834d3fd9a874ada8c1d21663e5d4361f83195f718`
**Published:** 2026-04-27. **Withdrawn:** 2026-08-10.
**Contents at withdrawal:** `wallet_analysis_20260425_201800.csv` (8,997 rows)
and `wallet_all_positions_20260425_201800.csv` (542,241 rows).

## Why it was withdrawn

Three measured reasons, all verified against the bundle's own bytes on the day
of withdrawal.

**1. Per-wallet statistics computed on truncated samples.** The extractor
capped each wallet's trade history at 2,000 order-book fills. Of the 7,805
wallets in the bundle clearing the five-resolved-position floor, **2,644, or
33.9 percent, sit at exactly that cap**, meaning their histories were cut short.
The bundle carries no per-row flag saying which rows those are, so a reader
cannot tell a complete record from a window.

**2. A second, larger truncation.** The bundle predates the 2026-07-30 fix to
the resolution lookup, which requested 50 event slugs per call against a server
default of 20 and silently dropped every fill in the unanswered slugs. Capped
rows lost a further median 48.9 percent of their fills before scoring.

**3. Identifiable accounts joined to performance statistics.** The per-wallet
CSV carries a `name` column, and **8,766 of its rows hold a Polymarket display
handle** rather than an address, published beside that account's Brier score,
win rate and profit. The bundle's previous README stated the handles were "not
anonymized in this bundle by design." On evidence that cannot support a
per-account statement, that combination should not have shipped.

## Why this notice exists rather than a quiet deletion

This is the same defect class that retired the top-100 whale calibration audit
on 2026-08-09, at roughly ninety times the scale and with names attached. That
audit was withdrawn because 85 of its 100 rows sat at the same cap. Applying
that reasoning to 100 accounts and not to 8,997 would have been a decision
about how visible the problem was, not about whether it was real.

The withdrawal is recorded in the audit-chain verifier at `/research/verify`,
which reports this path as deliberately withdrawn rather than as a missing
file, so a broken link and a retraction stay distinguishable.

## What this does and does not invalidate

It does **not** retract the V1-M cross-venue paper's Manifold half. The
Manifold bundle, `v1m-data-bundle.tar.gz` (15,106 aggregated users, hashed
identifiers), is unaffected and remains available.

It does mean that no public artifact currently reproduces any Polymarket
per-wallet result. See the dated correction under Reproducibility at
`/research/edge-score-methodology-v1`, which sets out the same limitation for
the V1 paper.

Any future re-cut must ship without display handles, with a per-row truncation
flag and a per-row skipped-fill share, and must not reuse this filename or this
sha256.

---

## Correction appended 2026-09-04

The paragraph above under "What this does and does not invalidate" says the
Manifold bundle `v1m-data-bundle.tar.gz` "is unaffected and remains available".
That was true when it was written and is no longer true. On 2026-09-04 the
Manifold bundle and its unpacked per-user file were withheld pending
Manifold's written position on data rights. The notice, with the sha256 each
carried as published, is at `/research/v1m/README-manifold-bundle-withheld.md`.

Appended rather than edited in place, because this is a published withdrawal
notice and a reader who cited it should be able to see what it said when they
read it. The sentence above stands as written; this correction supersedes it.

The rest of this notice is unchanged and still stands. In particular, the
2026-08-10 withdrawal remains a retraction on evidence grounds, and the
2026-09-04 withholding is not: no Manifold result has been retracted.