Shadow Ledger

Sealed before the outcome. Verifiable after.

Name: Forum Atlas Shadow Ledger
Creator: Forum Atlas

Forum Atlas puts its forecasts on the record against open prediction markets — Kalshi and Polymarket — using a commit–reveal scheme. The probability is cryptographically sealed before the market settles, and revealed after, so we cannot quietly edit a bad call. You do not have to trust us: every row below is checkable by anyone with a shell.

Correction — 2026-08-01

A classifier defect (bare-substring keyword matching, plus an out-of-scope denylist that could never be completed) admitted entertainment, sports and electoral markets into an industry-sector book and mislabelled their sectors.

15 never-in-scope rows (14 Emmys award markets and 1 football transfer market) were still sealed — no probability, salt, outcome or Brier had ever been published for them — and were removed before reveal, the same remedy and disclosure as the 8-row pre-merge curation of 2026-07-20. 2 Formula-1 rows had ALREADY settled and been Brier-scored; those are NOT retracted and NOT hidden — they remain published and verifiable, annotated, and excluded from headline accuracy only. 25 in-scope rows kept their commitment and hash and had only their sector label corrected; sector is not an input to the commitment hash, so every published hash still recomputes byte-for-byte.

Internal audit reference WS-K-001 / WS-F0-004 / WS-F0-005 / WS-F0-006. Rows removed before reveal: 15. Already-revealed rows annotated and excluded from headline accuracy (never removed): 2. Sector labels corrected with commitment and hash untouched: 25. See our corrections policy.

179

forecasts committed & sealed

resolved & revealed to date

Jul 2026

first settlements expected late July 2026

Counts and rows queried live from the production database at page render. The scoreboard starts empty on purpose — an honest empty scoreboard beats a fake one.

How commit–reveal works, in plain language.

Commit. When we take a position on an open market, we do not publish the probability. We publish a commitment hash instead: sha256("<market_source>:<market_id>|<probability, 4 decimal places>|<salt>") — the market's source and ID, our probability formatted to four decimal places, and a random salt, run through SHA-256. The hash pins the number without disclosing it; the salt prevents recovering it by brute force.
Reveal. When the market settles, we publish the probability and the salt for that row. Anyone can then recompute the hash and confirm it matches the one committed before the outcome was known. A changed number produces a different hash — there is no way to revise history undetected.
Outcomes are not ours to declare. The outcome of every forecast comes only from the market's own settlement — never from our judgment.
Scores accumulate in public. Each revealed row gets a Brier score (lower is better), alongside the market's own Brier score computed from its price at our commit time — so you can see not just whether we were right, but whether we beat the market.

Check a revealed row yourself.

Take any revealed row's source, market ID, probability and salt, and recompute the hash in one line — it must equal the row's commit hash, byte for byte:

echo -n "polymarket:123|0.6200|<salt>" | shasum -a 256

Machine-readable copy of this entire ledger: /ledger.json (schema forum.shadow_ledger.v1).

Independent timestamps. This page is anchored in the Internet Archive, which keeps snapshots we cannot edit after the fact. Compare any commit hash against the earliest capture at archive.org/web/*/forumatlas.com/ledger to verify it existed — unchanged — before settlement.

The ledger.

All 179 rows, newest first. Sealed rows show only the commitment; probability and salt appear when the market settles. Click a hash to select it.

Committed (UTC)	Source	Question	Sector	Closes (UTC)	Commit hash	FA p	Salt	Outcome	Brier (FA · mkt)
2026-08-01 14:06 UTC	polymarket	Will the number of Republican House members who retire in 2026 be between 40 and 43 inclusive? correctedSector label corrected: the original label came from a bare-substring keyword match ('health' matched inside a word, not as a word). The commitment, its hash and any Brier score are unchanged — sector is not an input to the commitment hash. Audit WS-K-001/WS-F0-004/WS-F0-005/WS-F0-006.	—	2026-08-31 00:00 UTC	`a89d5e97b4a52a33303256e046df16db888e1103ae8265589d66a7c18c697386`	sealed until settlement
2026-08-01 14:06 UTC	polymarket	Will Mark Sanford be the new republican nominee for Senate in South Carolina? correctedSector label corrected: the original label came from a bare-substring keyword match ('auto' matched inside a word, not as a word). The commitment, its hash and any Brier score are unchanged — sector is not an input to the commitment hash. Audit WS-K-001/WS-F0-004/WS-F0-005/WS-F0-006.

Related: the accuracy scoreboard for platform forecasts, and the transparency page for how Forum Atlas discloses what is and isn't AI-generated.