Every material platform change, on the record.

Beta 0.8

AI-native access and paid evaluation

• Added stateful live-decision alerts that silently baseline each immutable forecast and notify saved-race members plus globally matching Signal Scouts on reliable WATCH/VALUE threshold crossings
• Exposed the paid locked-versus-current decision engine through a derived-only Syndicate REST resource and read-only MCP tool with fail-closed card compatibility and WATCH/VALUE trigger prices
• Added a bearer-authenticated, read-only MCP endpoint so Syndicate customers can bring current signals, derived prediction records, filtered market tapes, and prospective trust evidence into compatible AI agents
• Turned one-time API-key creation into copyable generic/OpenAI MCP connection packs with a live authenticated protocol handshake
• Added a $9 non-recurring Race Day Pass with exact-price Checkout validation, idempotent 24-hour access, delayed-payment handling, and refund/dispute revocation
• Extended pass access through interactive and background Overlay workflows, privacy controls, monthly conversion, and operator revenue telemetry
• Corrected public evidence to count one terminal revision per race and contract, then added a chronological holdout ensemble calibration lab that can never alter production weights automatically
• Closed the Syndicate webhook lifecycle with an exact derived prediction resource so every settlement event can be fetched directly without paginating history
• Turned append-only changed quotes into a paid 72-hour filtered consensus market tape for racecards, Syndicate API workflows, and signal webhook follow-up
• Added a timestamp-safe forecast read to the market tape so members can see whether the locked model-versus-consensus value gap expanded, eroded, stayed steady, or had no reliable later move
• Made decision alerts revision-aware: members hear when a qualified signal appears, strengthens, weakens, or disappears, while unchanged scheduled refreshes stay silent
• Added a paid channel-agreement lens and derived API diagnostics for the visible spread across market, ratings, and qualitative AI probabilities

Beta 0.7

Racecard-safe personal decision evidence

• Locked field and condition fingerprints into every current forecast and failed current-decision surfaces closed after racecard drift
• Added immutable, server-validated pre-race research receipts with member intent, model evidence, accepted quote provenance, and settled quote-versus-SP review
• Closed the receipt lifecycle with deduplicated card-drift and official-settlement notifications while keeping email opt-in explicit
• Made each official race-result import, forecast settlement, and receipt outcome notification one repeatable-read transaction
• Included research receipts in private account export and deletion while excluding them from prediction generation
• Added intent-split personal receipt diagnostics for calibration gap, binary Brier error, and quote-versus-SP discipline with bounded-window and non-independence disclosures
• Bounded every model call to 6,000 output tokens and added an AI-only list-MRR contribution view that includes scheduled shared generation cost
• Added an idempotent subscription observation ledger and bounded operator lifecycle view for starts, cancellations, recoveries, plan movement, intervention queues, and structured cancellation feedback
• Added verified, endpoint-signed Syndicate webhooks for derived signal updates and official settlement with bounded retry and delivery operations
• Added signed production-path webhook tests and cursor-paginated delivery diagnostics for Syndicate integrations
• Added durable current-month Syndicate API usage visibility by resource for customers, response headers, and operator demand evidence without inventing a monthly cap
• Expanded the prospective ledger into a fixed model trust map across region, surface, race type, field size, and input confidence with paired uncertainty and minimum-sample direction gates
• Added per-race trust context to the paid next-24-hours briefing without allowing retrospective evidence to create, suppress, or reorder signals

Beta 0.6

Evidence-grade market and performance reporting

• Rejected market quotes without a verifiable source timestamp
• Attached bookmaker and quote time to each accepted best observed price
• Prevented AI from treating jockey or trainer names alone as performance evidence
• Added a paid upcoming decision board and machine-ready Syndicate signals endpoint
• Added separately selectable price-move and intelligence-ready alerts with idempotent email delivery
• Added performance evidence tiers, 95% strike-rate intervals, and maximum paper drawdown
• Added matched-race Brier skill reporting against the locked de-vigged market
• Added region and data-confidence diagnostics to expose hidden performance regimes
• Restricted shared forecasts and live signals to a 30-minute fresh-market window
• Added a durable member activation checklist and deduplicated commercial activation metrics
• Extended Signal Scout with price-band, accepted-book-count, and maximum-spread rules applied consistently to live and settled evidence
• Added an evidence-cited AI pace map with explicit unknown and confidence states
• Turned locked pace styles into a coverage-gated field-relative pressure and setup diagnosis for racecards and Syndicate records
• Added optional pace-pressure rules consistently across Signal Scout, alerts, briefings, review, evidence, decision board, and Syndicate signals
• Turned channel agreement into a saved fail-closed rule across the decision board, alerts, briefings, settled evidence, REST, and MCP
• Added a paid Signal Scout rule audit with overlapping rejection pressure and strictly rejected one-rule near misses
• Added a paid same-contract forecast revision trail for probability, rank, signal, pace, component movement, and qualitative-generation lineage
• Added a private Scenario Lab for reweighting the locked market, ratings, and qualitative AI components
• Added an automated finite-grid weight stress test with probability, rank, and leader-stability ranges
• Projected weight robustness onto the paid decision board and added a stable-only Syndicate signals filter
• Extended Signal Scout with a fail-closed rank-stability rule shared by live, alert, email, and settled-evidence workflows
• Added worst-grid EV and probability-edge ranges plus a fail-closed robust-edge rule across member and API workflows
• Added a paid live decision monitor that reprices locked probabilities against the newest reliable market and explains signal persistence
• Added paid top-two/top-three Harville projections and the same derived finish bands to the Syndicate prediction API
• Made Syndicate history self-scoring with settlement filters and derived official outcomes in JSON and CSV
• Published the same aggregate connection-combination record through a rate-limited, derived-only Syndicate API contract
• Added prospective top-two/top-three Brier scoring and calibration from complete official finishing-position ledgers
• Added a private stable-ID runner tracker with upcoming declarations plus consent-separated result intelligence linked to the latest compatible locked forecast
• Added paid longitudinal runner profiles joining stored results, rating movement, matching context, and current-contract locked forecast history
• Added paid stable-ID trainer and jockey profiles with stored form, upcoming entries, context splits, and locked-model expectation diagnostics
• Added a 100-connection trainer/jockey tracker with grouped declaration alerts, separately opted-in email, and privacy lifecycle coverage
• Added a paid racecard trainer–jockey combination lens with recent/context records and locked expected-versus-observed diagnostics
• Added a personalized next-24-hours dashboard briefing and separately opted-in, local-time, exactly-once daily email delivery
• Added a seven-day post-race review joining the newest locked revision to saved cards, tracked runners, current Scout rules, and official outcomes
• Separated qualitative-generation identity from full forecast identity so market-only refreshes reuse validated AI work with lineage and zero new model tokens
• Clarified that the public ledger is platform-controlled rather than independently proofed

Beta 0.5

Independent forecast channels and reliable settlement

• Separated qualitative AI inputs from the numerical ratings component to prevent double counting
• Added paginated three-day official-result catch-up and winner-gated settlement
• Added private launch operations, audited recovery, and Stripe entitlement reconciliation
• Added exact-once usage accounting, AI cost controls, and a clean production dependency audit

Beta 0.4

Decision-grade market intelligence

• Added a deterministic median-book, de-vigged market baseline and expected-value states
• Separated non-price AI analysis from the deterministic market channel and added a bounded hybrid forecast
• Added public Brier scoring and forecast-band calibration after settlement
• Added paid prediction entitlements with one rotating live preview
• Added a commercial launch kill switch for data-rights and legal approval

Beta 0.3

AI-generated visual identity

• Introduced the thoroughbred-and-data brand mark
• Added optimized navigation, favicon, and Apple icon assets

Beta 0.2

Syndicate delivery surface

• Added hashed API-key lifecycle
• Added derived prediction API
• Added derived-only CSV ledger and account exports

Beta 0.1

Provider-backed platform foundation

• Replaced fictional production races and performance claims
• Added racecard, odds, result, and settlement ingestion
• Added immutable structured predictions and public ledger
• Added account tiers, quotas, billing, and in-app price alerts