MARKETS TRACKED7GME AMC IWM SPY QQQ NVDA TSLA
LIVE JURY DECISIONS355 verified TypeSafe Jev jurors per ticker at launch
RESOLVED GME OUTCOMES36Five-session descriptive test
NEW FORWARD CLAIMS14Frozen; outcomes still pending at launch
THE EVIDENCE // ORIGINAL GME OUT-OF-SAMPLE COHORT
Our model failed to beat an always-bullish benchmark.
| Same 36 five-day windows | Correct directional calls | Observed accuracy |
|---|
| SML Fractal GME | 18 of 36 | 50.0% |
| Always predict GME rises | 28 of 36 | 77.8% |
| Difference (model minus baseline) | −10 correct predictions | −27.8 percentage points |
Original October 21 launch cohort: the 36-outcome figures above describe the historical research snapshot. The live scoreboard updates after verified new outcomes. Open current GME metrics and resolved sample.
Four real Jev jurors chose REJECT PROMOTION and the Tail Risk juror chose WAIT FOR OUTCOMES. The hard research policy also blocks signal promotion when the model fails its reference. That is not a market-direction prediction from Jev, nor a claim of profitable trading.
Inspect the frozen GME scoring evidence · Inspect the signed Jev jury verdict.
THE FIVE TYPE-SAFE JUROR PERSPECTIVES
JEV.AI weighs the evidence. SML enforces the safeguards.
⚖️ Causal
Checks whether a claim follows from frozen evidence rather than hindsight.
📚 Historian
Examines matching windows, sample sizes, and comparisons to simple strategies.
🔎 Skeptic
Challenges overconfident conclusions and contradictory evidence.
📈 Microstructure
Separates stock-price accuracy from actual execution economics.
🛡️ Tail Risk
Flags missing profit evidence, correlated time windows, and immature outcomes.
The existing SML Calibration Court runner requests genuine typed replies from the TypeSafe Jev model. It saves the replies, hashes the record, and signs the public response with SML's Ed25519 key. The Forecast Court client verifies the source signature, hash chain and current market snapshot. A SML-signed receipt is not TypeSafe countersignature or endorsement.
FORWARD RESEARCH PILOT
Seven markets. No reconstructed historic wins.
On October 21, 2026, two transparent research strategies were frozen on each of GME, AMC, IWM, SPY, QQQ, NVDA and TSLA. The strategies follow or reverse the prior five-session return, with exact completed-price inputs recorded. The Court will wait for five *new* completed sessions to judge each forecast, then compare it with an always-up rule over the same symbol and dates. Cross-ticker rankings are not treated as risk-adjusted comparisons.
Open raw live market rankings · Inspect today's Court processing status.
DEVELOPER ACCESS // FREE READS AND ELIGIBILITY-GATED PURCHASES
Research-ready machine interfaces.
Read the full public evidence without a login. Each ticker's jury can be inspected in raw JSON. Submitted third-party forecasts currently require an operator credential; public self-serve registration is not represented as live.
GET https://sml-fractal-trading-os-v2.onrender.com/api/forecast-court/leaderboard
GET https://sml-fractal-trading-os-v2.onrender.com/api/forecast-court/jev-jury?symbol=GME
GET https://sml-fractal-trading-os-v2.onrender.com/api/forecast-court/status
Optional paid aggregate consistency check
The mature GME cohort supports a $0.01 USDC (Base mainnet) x402 machine-readable audit. SqueezeOS independently recomputes aggregate correct/total and matched baseline percentages; it does not reaudit all raw historical forecasts. A real HTTP 402 payment challenge has been observed, but an additional independent buyer settlement and accepted delivery have not yet been verified.
GET https://squeezeos-api.onrender.com/api/forecast-court/verify-info
GET https://squeezeos-api.onrender.com/api/forecast-court/verified-audit?symbol=GME&strategy=sml-fractal
Price: 0.01 USDC • Network: eip155:8453
Eligibility: at least 20 actual resolved five-session observations
Payment: compatible x402 agent required; unauthenticated GET returns HTTP 402
Inspect current x402 purchase contract · Request white-label jury/integration licensing.
QUESTIONS // MARKET EVIDENCE WITHOUT HYPE
Frequently asked
Does Jev place trades or choose stock direction?
No. The five Jev outputs evaluate existing research claims and limitations. Automated orders, staking and automatic model promotion are not enabled.
Is a 50% hit rate evidence of profit?
No. Profit depends on prices paid, option premiums if relevant, holding periods, costs, slippage and risk. These figures only describe subsequent underlying-price direction over frozen evaluation windows.
How can an AI jury call itself independent?
It is a separately run third-party-model evidence assessment, not independent external auditing of every raw source. TypeSafe supplies the Jev model; SML runs, stores, signs and presents the model outputs. The distinction is intentional.
When will the other six markets have verified scores?
Only after the future five completed market sessions have occurred for their frozen October 21 predictions and the source data has been verified. There is no retroactive backfill of wins.
What happens if market evidence changes?
The public jury relay compares a signed verdict's evidence fingerprint with the current scoreboard and shows a pending-review status instead of stale juror votes. The scheduled Jev runner can then evaluate the updated snapshot.
Can I buy an audit without a wallet?
The free Court needs no wallet. The $0.01 audit is a machine-readable x402 endpoint; a compatible paying agent must satisfy its payment challenge. The link itself does not initiate a user payment.
THE MODEL MAKES THE CLAIM // THE JURY WEIGHS THE FACTS
Don't trust a win rate. Inspect the entire Court.
Full-screen interactive scoreboard, live seven-ticker jury tabs, signed public verdicts, source links and sharing tools.
