The Terminal
Today's slate.On your desk.Before kickoff.
The signal as a ready-to-run app for your trading desk: every model call fully revealed — side, line, edge, EV — with the audit receipt behind it. White-labeled for your book, distributed privately to your team, never the public store.
Live
§ 01Inside the app
The whole story, on the desk.
6 sports, one pass per window, every pick the system publishes — and a verifiable receipt for every call we make. The whole app is built around the same proof loop the website reports against.
Intelligence
Insights
Track record
§ 03The receipts
Called before kickoff. Verifiable after.
Every pick ships with its receipt — SHA-256 content hash, anchor timestamp, conformal pass/fail, and lead time before kickoff. Tap through to verify against the public ledger.
Sealed in an immutable ledger
Append-only by database trigger. We can't edit or delete it — not us, not an admin, not a compromised credential.
Anchored to Bitcoin
During the alpha, each day's manifest was salted, hashed, committed to the public ledger before kickoff, and upgraded to a Bitcoin block proof. The record is sealed; the official ledger continues the protocol from its first anchor.
Verify in two commands
Clone the public repo, run verify.py. Pure-stdlib Python, no toolchain, no accounts. Hand the link to anyone who asks "is this real?"
Sample receipt
What every pick ships with.
Each prediction is hashed, anchored, and stamped with the lead time before kickoff. The fields below are the receipt's actual structure — the same fields the public verifier checks. Values shown are illustrative; the real receipt for any anchored pick is retrievable via the audit_trail repo.
Most recent anchor · Jun 25 · 12:00 UTC
- Content hash
- 0x95a29c4ccbaded85…
- Anchor timestamp
- Jun 25 · 12:00 UTC
- Kickoff
- Jun 25 · 23:45 UTC
- Lead before kickoff
- —
- Conformal status
- 2-outcome setConformal · FAIL
- Model
- 1147456614
Open the receipt; open the ledger.
§ 04Live insights
Per-model performance. Live, from the ledger.
Each of the 16 production models gets its row — live win rate and sample size, read straight from the immutable record. Equity curve underneath. Live only — backtest is never blended in.
§ 05Market mapping
Every pick. Matched to its market.
Trade tab matches each live prediction to its live Kalshi market — the model's calibrated probability next to the market's implied price, the edge per contract, and an order ready to send from your own Kalshi account.
Edge over the market
Model probability minus market-implied probability — surfaced per contract alongside EV after Kalshi's fees. Tiered Strong / Fair / Slim so the gap is scannable, not buried.
Sized to confidence
Side, line, edge, EV, Kelly and confidence — every signal the model uses to size, surfaced together on the same screen as the order.
Mapped, not executed
The engine matches each pick to its tradable market and exposes the mapping through the feed. Execution stays wherever your book already does it — we ship signal and mapping, never orders.
Market matching ships in the feed: every pick carries its exchange-market mapping where one exists.
§ 06SplitWinnerGPT · in-app
SplitWinnerGPT — ask the model directly.
A purpose-built LLM grounded in today's audited slate. Ask why a pick looks the way it does, how a model has been performing, what the audit receipt actually proves, or where the Kalshi edge sits — every answer cites the exact public ledger entry it came from.
- Scoped to today's picks, the models behind them, and the live audited record — no hallucination from outside scope.
- Every numerical answer carries a citation pointer back to the audit_trail row it was computed from — auditable, not generated.
- Powered by Anthropic; the grounding layer is ours. Provenance, citations, and refusal behaviour disclosed in-app.
Want more scope?
The standalone SplitWinnerGPT product opens the same engine to the full audited record — any sport, any model, any period — not just today's slate. Same citation discipline, no scope cap.
About the standalone SplitWinnerGPT§ 08Questions
Before you deploy.
Plain answers to what a buyer asks first.
How does my desk actually run this?
Private distribution, never the public store: TestFlight or Ad Hoc for pilots, Apple Business Manager Custom Apps pushed through your MDM for real deployments. Your branding on the shell, our signal underneath, your seats managed by you.
How is it licensed?
Per-desk under the 2026 beta — a fixed seasonal price in writing covers the feed and the terminal together. No per-seat metering games during the beta; bring the traders who need it.
What can we do with the data internally?
Use it on your desk: price against it, alert on it, store your own copy of what you consumed. No redistribution or resale outside your book, no training models on the feed — the contract states it plainly and the ledger keeps both sides honest.
Is the terminal separate from the API?
Same engine, same hashes, same anchors. The terminal is a rendered view of the feed — anything it shows, your quants can pull raw from /v1/* and verify against the public ledger.
What if a game is cancelled or postponed?
If the model issued a prediction for a game that doesn't get played, the row stays on the ledger but doesn't score — it can't help or hurt the model's record. Your pass keeps access to every other pick the system publishes during your window.
How does the in-app GPT differ from the standalone?
The in-app chat is scoped to today's pick, the models behind it, and the live record — it's distilled. The standalone SplitWinnerGPT product opens the same engine to the full audited record: ask about any sport, any model, any period the public ledger covers. Same citation discipline, no scope cap.
§ 09Build on it
SplitWinner API.
The same hardened feed the consumer app reads, exposed as a contract. By inquiry — limited beta, hand-selected integrators.
Enterprise · by inquiry
Audited engine, on a contract.
Tamper-evident prediction provenance as an API. Daily, pre-game predictions in JSON with cryptographic receipts your security team can verify against the public audit_trail repo. For regulators, sportsbook compliance, integrity vendors, and regulated sports-trading desks.
Signed, versioned /v1/* REST surface — Bearer + HMAC on every request.
Daily, pre-game JSON predictions with audit_trail row ids + SHA-256 content hashes.
99.5% uptime, p95 < 500 ms on prediction endpoints.
Public verifier ships with every prediction — clone github.com/SplitWinner/audit_trail_alpha, run verify.py, check.
OpenAPI spec as the contract. RFC 7807 errors. X-RateLimit-* headers.
Reach the team at sales@splitwinner.com.
Talk to enterpriseUpdates
Methodology, product, and pricing changes — short notes, not the daily drop. Unsubscribe in one click.
How it gets there
Open methodology. Real numbers.
The numbers on this site come from the production pipeline. Tap through for held-out performance, calibration drift, and conformal coverage.
Sandbox first. Then the beta.
A key in minutes — the full API surface, mocked plausibly-live, no card, no NDA. When it holds up on your desk, the 2026 beta cohort is a direct conversation: a limited set of desks, a fixed seasonal price in writing.
No card · no NDA · fixed price in writing