Research Loop | PredictCoin

Demonstration fixture — not production data.

This frozen exercise reuses the real scoring engine and receipt view, but the market and prices are demonstration values. It is isolated from public performance statistics and never reads live feeds. Audit hash-chain verification runs in the test suite, not in the browser preview.

Market (frozen fixture)

Question: Will BTC/USD close at or above $100,000 at 12:15 UTC on 2026-10-22? (demonstration fixture)
Venue: kalshi
Ticker: DEMO-BTC-2026-OCT22-FIXTURE
Source: demonstration-fixture
Status: finalized
Outcome: YES

Evidence — timestamped observations

Checkpoint +3mcaptured
scheduled 2026-10-22T12:03:00Z · captured 2026-10-22T11:59:30Z
YES 69.0% · NO 31.0%
Checkpoint +6mcaptured
scheduled 2026-10-22T12:06:00Z · captured 2026-10-22T12:06:12Z
YES 71.0% · NO 29.0%

Forecast receipt (issued before close)

Initial forecastIntegrity not verified

Will BTC/USD close at or above $100,000 at 12:15 UTC on 2026-10-22? (demonstration fixture)

kalshi · demo-contract-yes · demo-pred-0001

Estimated YES probability

70.0%

Raw model estimate: 70.0%
Calibrated estimate: Unavailable

Explanation as recorded

Thesis: Unavailable
Strongest counterargument: Unavailable

Explanations are shown as recorded at forecast time, not regenerated.

Timing

Forecast issued: 2026-10-22T12:00:00Z
Server receipt time: Unavailable
Market closes: 2026-10-22T12:15:00Z
Actual lead time: 15.0 min

Decision and data health

Eligibility status: research_only
Data health: passed
Model version: research-fixture-v0
Calibrator version: Unavailable
Policy version: prediction-eval-v1

Reason codes: research_model_not_ported

Recorded sources

Snapshot manifest
digest: fixture-snapshot-digest-0001

Resolution

Resolution status: resolved
Resolution source: demonstration-fixture-settlement
Outcome: YES
Individual Brier loss: 0.0900

Pending, disputed, and void outcomes are not scored. One resolved event is not proof of calibration.

The research forecasting model is not ported in this build, so the forecast is honestly labeled research-only. Its quality is still scored once resolved — forecast quality is separate from trading eligibility.

Resolution + reproducible evaluation

Resolution status: resolved
Outcome: YES
Resolution source: demonstration-fixture-settlement

Evaluation report

Status: complete
Scored forecasts: 1
Unique events: 1
Brier loss: 0.0900
Exclusions: later_revision_not_scored
Policy version: prediction-eval-v1
Scoring version: 2.0.0

Replaying the same frozen fixture reproduces this exact selection and score every time. One resolved event is not proof of calibration.

Bad-case acceptance exercises

stale_data_blocks_eligibilitydegraded
data_health=degraded; eligibility=abstained; reasonCodes=[STALE_DATA]
missing_source_shows_unavailabledegraded
receipt valid; recorded source artifacts: none
conflicting_settlement_blocks_evaluationblocked
evaluation.status=blocked; conflicting_event_records
duplicate_request_rejectedrejected
idempotency_key already recorded; duplicate write refused
cross_tenant_access_denieddenied
forged tenant/owner field does not authorize access; expected 403/404 with no record content
audit_failure_fails_closedrolled_back
requireAudit throws on persistence failure; protected write is not committed (verified in node test suite)

Each case reuses real production logic and reports the honest outcome — none fabricate a success.