calibration and reliability command center
Calibration command center
KELLY is allowed to learn from forecasts only after reality closes the outcome. This page shows the scoring path: forecast, settlement, Brier score, reliability posture, replay, and memory eligibility.
Settled
0
Pending
0
Average Brier
waiting
calibration contract
What must happen before learning?
reliability audit path
KELLY does not get credit for unclosed predictions.
The calibration surface is for settled, replayable outcomes. It separates pending forecast debt from scored evidence so the engine cannot inflate progress with unobserved claims.
Settled outcomes only
Calibration starts after an approved reality source closes the forecast.
Probability versus outcome
The score compares the forecast probability against the observed result.
Brier and reliability
Lower Brier is better; reliability asks whether 70% claims happen about 70% of the time.
Replay before memory
Calibration is useful for learning only when proof and replay reconstruct the chain.
score interpretation
Brier score is squared forecast error on observed outcomes.
A 0.70 forecast that happens earns less error than a 0.30 forecast that happens. Pending forecasts stay outside the score until reality resolves them.
no fake calibration
Calibration records should appear only after settlement. Replay and proof links are the audit trail that keeps scored learning separate from text confidence. This remains decision-support telemetry, not a command to act.
Live proof mapchecking
Follow KELLY's current governed learning lane from source to observation, receipt, settlement, calibration, replay, and governed memory. The headline metrics come from the live runtime; archived bootstrap lineage is kept only for traceability.
Sources
0
Observations
0
Hypotheses
0
Closures
0
Visible cells
0
Replayable
0
Total Neon cells
0
Visible replay receipts
0
Next candidates
0
Velocity
unknown
Source: unknown. Blocker: none. Public users cannot directly write ground truth; archived ledger backlog is monitored separately from current cell replay receipts.
none
Loop state: waiting
Waiting for public-safe proof-map evidence.
Last checked: not yet. Public user writes: blocked.
Calibration SnapshotLive API
Settled: 0
Pending: 0
Average Brier: —
See calibration drive cards below for domain / horizon breakdown.
Reality calibration by domainUnavailable
Neon persistence: not wired on this API host
No domain calibration snapshots yet — settle forecasts first.
Sealed Proof EvidenceSealed proof evidence
FULL PASS / GO
Authority: KELLY_SEALED_PROOF_EVIDENCE.json
Council: council_b2eb6556-3f10-49cb-b…
Proof: proof_council_b2eb6556-3f10-…
Chain reconstructed: true
Open calibration proof windowReplay archived bootstrap receiptSealed council reference (proof_council_b2eb6556-3f10-…) is historical lineage metadata. Live proof receipts are listed on Receipts and Glass Lab.
Operator calibration summaryUnavailable
Summary unavailable. Sealed reference verdict: FULL PASS / GO
Unsettled debtUnavailable
Domain / horizon breakdown0/30 settled
Live drive: accumulating. Trend cards appear when spine data is available.