← Trading Hub

Trust scoreboard

Calibration, hallucination, override, and decision-telemetry metrics. Public, refreshed periodically (not a fixed cadence — see per-metric dates below). Per Execution Bible §15.68 — published BEFORE the metrics are flattering, because the alternative (closed black-box commentary) is the entire reason institutional research can't be trusted.

Trust commitments

1. Every score is decomposable

Every output cites the factors that drove it (top contributors with values + units). No black-box numbers — every score reverses to its components. Click any score on /security?ticker=X to see the factor stack.

2. Calibration is measured, not assumed

Forecasts are p10/p50/p90 cones, not point estimates. Realised vs forecast tracked weekly via Brier score (per §15.1). When the engine says "30% chance of -8%," roughly 30% of those events should hit -8%. Calibration drift triggers a /internal/calibration alert.

3. Overrides are first-class

Every recommendation overridable in one click. Overrides + reasons feed back into factor weights (closed loop per §14.5). Disagreements with the engine are signal, not noise.

4. Sources are cited

Every external data point shows its source + freshness badge (per §15.6). Two feeds disagreeing by >5% surface as "feeds in conflict" — never silent-averaged.

5. Mistakes are documented

Every postmortem ships within 7 days of an incident (per §15.70). Listed in /changelog under "Security" or "Fixed". No silent rollbacks.

Operational metrics

Tests passing
322/322
100% green · last run 2026-07-01
Console errors (mobile)
0
measured 2026-04-26 (8 root causes resolved) · not re-verified since
SEO score
88
measured 2026-04-26 (canonical, OG, JSON-LD, sitemap, robots) · not re-verified since
Mobile TBT
170ms
measured 2026-04-26 (-90% via bundle) · not re-verified since
Calibration tracked
v104
scorecard engine shipped · awaiting 30d Brier sample
Decision telemetry
draft
localStorage-only · server-side ai_decisions deferred

Per-engine calibration

v76 Layer-conditional scoring (Tier 1 macro factors) awaiting sample
Prescience expected-move (14 asset classes) 14/14 classes
v77 Crypto-perp funding/OI/liquidations live
v95 News-velocity sentiment awaiting validation
v110 Prescient ensemble v2 (15 streams) scaffolded

Methodology: every signal logged with predicted distribution + realised outcome. Brier score computed weekly. Engines below 0.55 reliability flagged for re-weighting per §15.1.

Source attribution snapshot

Every external feed used in scoring, with usage scope + freshness contract.

CoinGeckoCrypto spot prices · TTL 5min · used by v77/v79
Binance FuturesFunding/OI/Liquidations · TTL 5min · v77 only
Yahoo FinanceEquity prices/FX/rates · TTL 60s · all classes
FREDMacro / yields / DXY · TTL 1h · v83
SEC EDGARFilings · TTL 1h · v80 event-driven
FDADrug approvals · TTL 1h · v81 pharma
EIACommodity stocks/inventories · TTL 6h · v84
IMFFX reserves/balances · TTL 24h · v85
DeFi-LlamaOn-chain TVL/protocol · TTL 1h · v79
TradierOptions flow/IV · TTL 5min · v86
COTFutures positioning · TTL 7d · v78
Coinbase ExchangeBTC/ETH order ladder · TTL 30s · v77

Audit trail

DateEventOutcome
2026-04-26Bible-aligned pass: trust + status + changelog + legal pages addedShipped
2026-04-26Expected-move audit: 14-class profile table, leverage drag, contango, fundingShipped
2026-04-26Alerts pipeline: silent no-op since inception → reads scores+regime correctlyFixed
2026-04-26Console errors: 70+ → 0 (8 root causes incl IDB compound key, fetchWithRetry sentinel)Fixed
2026-04-26Mobile TBT 1760ms → 170ms via bundle (184 scripts → 1)Shipped
2026-04-26SEO score 38 → 88 via canonical/OG/JSON-LD/sitemap/robotsShipped
2026-07-01Comprehensive audit (security/data-integrity/cron-health/engine/ethics/code-quality/SOTA) — 36 confirmed findings, Critical/High/Medium/Low fixes applied + deployedFixed
2026-07-01Trust page KPI honesty pass: test count re-verified live (322/322); console/SEO/TBT KPIs now dated per-metric instead of one blanket "last run" implying weekly refresh that wasn't happeningFixed

Full release history at /changelog. Incident postmortems land here per §15.70 within 7 days of the incident.

Page last refreshed: 2026-07-01 (test count re-verified live; console/SEO/TBT KPIs carry their own last-verified dates above) · Next review: 2026-10-01 (quarterly cadence per §21.6)