PROOF·EDGE

ATTESTATION OF FORECAST INTEGRITY · Nº 0002

SOMNIA SHANNON · TESTNET 50312

VERIFIABLE FORECAST RECORDER / V1 + V2

Forecast preimages that cannot change after anchoring.

Bring any estimator. Proof·Edge freezes its probability and a contemporaneous market snapshot, seals both under a salted Keccak-256 commitment, and anchors the batch before on-chain expiry. There is still no minimum lead-time rule; the margin is measured, published and warned about instead of enforced. Pending and resolved records are published together, so losses remain beside wins.

Production root0x243125583bf2dd8d7c8a33710bc8c1377e5c86f959ae6c288a0236a1203c5543
Anchor transaction0x69b5e331…0beb2afb ↗
Model hash0x253a60a7…1dd16417
Estimatordreamdex-ec-oracle-follow-strike-adapter
Ledgerpublished/forecast-events.jsonl
Anchored late0 WINDOWS · 0 ROOTS · EXCLUDED FROM PROOF + SCORING
Anchor leadMIN 31 S · 31s · MEDIAN 285 S · 4m45s · 13 OF 3881 UNDER 90 S · LOW_LEAD IS A WARNING, NOT A VERDICT
Pending resolution16 WINDOWS · PUBLISHED, NOT YET SCORED
Orphan ledger events1 EVENTS · EXCLUDED FROM CANONICAL CHAIN · SEE LEDGER INTEGRITY REPORT
LicenseMIT

§ 1 Resolve & score

CURRENT MODEL0x253a60a726a063c0e14acd10d7a206a0b82308a8bc703ced5304c79a1dd16417

Primary reading · sealed version only

All evaluated windows

Current model only; PASS and VETO together.

CURRENT PRODUCTION
SKILL SCORE

−0.269N = 3648

95% CI [−0.300, −0.240]
BRIER · AGENT

0.301N = 3648

LOWER IS BETTER
BRIER · MARKET AT COMMIT

0.237N = 3648

FROZEN, NEVER REFRESHED
MEAN p_AGENT

0.488N = 3648

MEAN p_MARKET

0.490N = 3648

AGENT − MARKET

−0.002N = 3648

Risk-gate passed for execution

Current model only; first recorded gate ruling was PASS.

CURRENT PRODUCTION
SKILL SCORE

−0.020N = 667

95% CI [−0.042, +0.001]
BRIER · AGENT

0.227N = 667

LOWER IS BETTER
BRIER · MARKET AT COMMIT

0.222N = 667

FROZEN, NEVER REFRESHED
MEAN p_AGENT

0.483N = 667

MEAN p_MARKET

0.487N = 667

AGENT − MARKET

−0.004N = 667

Reliability of the sealed probabilities

Predicted p(YES) against the frequency the market actually resolved YES. N = 3648.

MARKET [0.0–0.1) · n=16 · predicted 0.0672 · observed 0.0625 · 95% Wilson [0.0111, 0.2833]MARKET [0.1–0.2) · n=84 · predicted 0.1577 · observed 0.1786 · 95% Wilson [0.1113, 0.2739]MARKET [0.2–0.3) · n=195 · predicted 0.2551 · observed 0.2564 · 95% Wilson [0.2002, 0.3220]MARKET [0.3–0.4) · n=506 · predicted 0.3560 · observed 0.3775 · 95% Wilson [0.3363, 0.4205]MARKET [0.4–0.5) · n=1038 · predicted 0.4569 · observed 0.4836 · 95% Wilson [0.4533, 0.5140]MARKET [0.5–0.6) · n=1165 · predicted 0.5374 · observed 0.5150 · 95% Wilson [0.4863, 0.5436]MARKET [0.6–0.7) · n=432 · predicted 0.6416 · observed 0.6181 · 95% Wilson [0.5714, 0.6626]MARKET [0.7–0.8) · n=138 · predicted 0.7393 · observed 0.6739 · 95% Wilson [0.5919, 0.7465]MARKET [0.8–0.9) · n=56 · predicted 0.8472 · observed 0.7679 · 95% Wilson [0.6423, 0.8590]MARKET [0.9–1.0] · n=18 · predicted 0.9426 · observed 0.8889 · 95% Wilson [0.6720, 0.9690]AGENT [0.0–0.1) · n=504 · predicted 0.0628 · observed 0.3730 · 95% Wilson [0.3319, 0.4160]AGENT [0.1–0.2) · n=427 · predicted 0.1464 · observed 0.4215 · 95% Wilson [0.3756, 0.4689]AGENT [0.2–0.3) · n=307 · predicted 0.2489 · observed 0.4528 · 95% Wilson [0.3980, 0.5087]AGENT [0.3–0.4) · n=299 · predicted 0.3498 · observed 0.4615 · 95% Wilson [0.4059, 0.5182]AGENT [0.4–0.5) · n=321 · predicted 0.4515 · observed 0.4984 · 95% Wilson [0.4441, 0.5528]AGENT [0.5–0.6) · n=355 · predicted 0.5490 · observed 0.4845 · 95% Wilson [0.4330, 0.5364]AGENT [0.6–0.7) · n=315 · predicted 0.6487 · observed 0.5333 · 95% Wilson [0.4782, 0.5877]AGENT [0.7–0.8) · n=310 · predicted 0.7471 · observed 0.5419 · 95% Wilson [0.4863, 0.5965]AGENT [0.8–0.9) · n=338 · predicted 0.8516 · observed 0.5030 · 95% Wilson [0.4499, 0.5559]AGENT [0.9–1.0] · n=472 · predicted 0.9381 · observed 0.6250 · 95% Wilson [0.5805, 0.6675]
agent · 95% Wilson bars market midpoint at commit · baseline perfect calibrationpoint radius ∝ √n · sealed model 0x253a60a7…

Every point is a probability that was anchored on Somnia before its market expired and scored against the outcome that arrived afterwards — the windows the current-model cards above score. No point here could be added, removed or reselected after the fact: the bins are whatever the sealed bytes already contained. Bars are 95% Wilson intervals on the observed frequency. All ten bins hold at least one forecast. (dashboard/app/forecast-data.json, key resolve_score.calibration).

Immutable model versions

Each row is derived from the model_hash already sealed inside its forecast payload.

7 VERSIONS · CURRENT FIRST
MODEL HASHSAMPLENMEAN p_AGENTMEAN p_MARKETBRIER A / MSKILL95% BOOTSTRAP CI
0x253a60a7…1dd16417CURRENT PRODUCTIONALL EVALUATED3648USABLE0.4880.4900.301 / 0.237−0.269[−0.300, −0.240]
RISK-GATE PASS667USABLE0.4830.4870.227 / 0.222−0.020[−0.042, +0.001]
0x5b5b26d8…687b9ecbHISTORICAL VERSIONALL EVALUATED6TOO SMALL0.4140.3620.174 / 0.147−0.189[−0.875, +0.120]
RISK-GATE PASS3TOO SMALL0.3890.3060.153 / 0.095−0.601[−0.885, −0.476]
0x6a7015d6…99257755HISTORICAL VERSIONALL EVALUATED4TOO SMALL0.2130.2120.186 / 0.144−0.297[−1.560, +0.125]
RISK-GATE PASS3TOO SMALL0.2100.1750.046 / 0.039−0.186[−1.645, +0.189]
0xe43b1820…0e8e5639HISTORICAL VERSIONALL EVALUATED9TOO SMALL0.3740.2800.206 / 0.180−0.144[−0.980, +0.261]
RISK-GATE PASS2TOO SMALL0.4470.4610.332 / 0.254−0.308[−0.341, −0.263]
0x914a3008…d73c3b34HISTORICAL VERSIONALL EVALUATED8TOO SMALL0.4160.5100.344 / 0.232−0.481[−1.378, +0.198]
RISK-GATE PASS4TOO SMALL0.4860.4770.216 / 0.249+0.131[−0.177, +0.304]
0x61b7d39b…a7c8bb41HISTORICAL VERSIONALL EVALUATED46TOO SMALL0.5040.5070.390 / 0.258−0.514[−0.769, −0.245]
RISK-GATE PASS10TOO SMALL0.5760.5470.299 / 0.273−0.095[−0.241, +0.073]
0xeb49a182…9e93d7c9HISTORICAL VERSIONALL EVALUATED144USABLE0.4670.4970.299 / 0.235−0.270[−0.437, −0.103]
RISK-GATE PASS30TOO SMALL0.5140.5190.213 / 0.221+0.033[−0.073, +0.135]

Mixed-model historical total

Context only. Never compare this total with a single model version.

SECONDARY · 7 MODELS COMBINED
ALL EVALUATED

−0.272N = 3865

MEAN p_AGENT 0.487 · p_MARKET 0.490 · BRIER A/M 0.301 / 0.237
RISK-GATE PASS

−0.020N = 719

MEAN p_AGENT 0.484 · p_MARKET 0.487 · BRIER A/M 0.227 / 0.222
EXCLUDED FROM SCORINGEvery reason stays visible; counts may overlap.
0ANCHORED LATE
0UNANCHORED
0VOIDED
16UNRESOLVED / UNREVEALED
0RESOLVED WITHOUT SCORE
0MISSING GATE RULING

* Scores update from append-only resolution events. They use only the sealed agent probability and the market probability captured at commit; there is no backfill or historical repricing. The interval is a deterministic 1,000-resample, 95% bootstrap.

§ 2 Ledger of sealed forecasts

agent  market · risk gating never filters the recorded sample

WINDOWCOMMITTED p(YES)AGENTMARKETGATEOUTCOMEBRIER A / M

EXHIBIT § 2.1 — SELECTED WINDOW

ETH / 5M 0x0000…2132

× TRADE VETOED
DIVERGENCE TOWARDNO
EDGE AT COMMIT18.70 pp
OUTCOMEPENDING
RISK RULINGdisagreement above ceiling — recorded, not traded

EVIDENCE DIGEST0x02fa337dc51e614ffcd79c2b975242ca09c6250e4f6121cce9a6dd922b737fa9

§ 3 Proof chain

  1. 01Snapshotmarket midpoint frozen✓ DONE
  2. 02Commitsalted Keccak-256 preimage✓ DONE
  3. 03AnchorMerkle root on Shannon✓ DONE
  4. 04ScoreBrier vs market baseline✓ DONE

§ 4 Independent verification

Do not trust this document.
Recompute it.

A clean clone reproduces commitments and Merkle proofs, checks every production root from the submitter, reads expiry and outcome from Shannon, and matches each receipt, block time, emitter, root, and leaf count. New-format anchors also bind the preceding ledger head. Late and pending records stay visible but unscored.

ROOT0x243125583bf2dd8d7c8a33710bc8c1377e5c86f959ae6c288a0236a1203c5543

ANCHOR TX0x69b5e331c87151395c7db05be090aac22e9d7a77e7885b825c001ac20beb2afb ↗

ENCLOSURE A — proof-edge / clean clone
git clone --recurse-submodules https://github.com/Vastargazing/proof-edge.git
cd proof-edge
npm install --no-save
npm run check
npm run verify:log
npm run verify:chain
npm run verify:completeness
npm run verify:all
EXPECTED3859 / 3881 PROVABLE · 1678 / 1678 ON-TIME ROOTS · 0 FAILURES · 1 PENDING AFTER WATERMARK
ENCLOSURE B — in-browser verifier · no install

The five checks below are the ones npm run verify prints. This panel imports the same src/evidence-verifier.ts the CLI uses and reads Somnia Shannon directly over public JSON-RPC; only the transport differs.