01
Evidence
Sealed point-in-time observations, each with an observation time, an availability time, and a source. The corpus is the run’s only view of the world.
Domain reference
One kernel, three domain modules.
A domain module supplies typed evidence, a belief shape, gates, and a settlement rule. The lifecycle, the receipt, and the refusals belong to the kernel, so adding a domain adds meaning rather than a second decision engine.
live:falsesimulation-onlyno custody
Same submission shape
What every run submits.
Whatever the domain, a run presents the same five things. The domain module decides what counts as evidence and which gates a belief has to clear.
02
Candidate
The versioned model or strategy under test. It proposes; it never authorises, and it can never reach an adapter on its own.
03
Baseline
The thing the candidate has to beat, evaluated on the same sealed cases so the comparison is like for like.
04
Mandate
The authority to act: objective, permitted effects, exposure or stake bounds, and an expiry read at the decision clock.
05
Domain module
The pack that gives the run its meaning: which evidence is admissible, how a belief is shaped, which gates apply, and what settlement is.
Full-field, pre-off, source-bound
Sport and racing
Runs bind a sealed point-in-time racing corpus, produce a ranked belief over the full field, and gate every wager through stake bounds and ruin policy before the decision is emitted. Settlement is the published result; learning records whether the belief or the price was wrong.
What you supply
- Chronological, pre-off history sealed with source authority and digest
- Quote snapshots per runner, with availability times
- Published results used only for settlement, never for inference
Gates and what they mean
- Stake bounds and maximum-loss policy per runThe most a single run may risk is declared before the decision, and the run cannot exceed it.
- Ruin policy evaluated before any stake is requestedThe worst-case drawdown for the declared bankroll is evaluated before any stake is requested.
- Out-of-sample holdout required before a candidate can actA candidate that has only seen development cases never reaches a decision.
Settlement
The published result resolves the run and binds to it. Learning records whether the belief or the price was the binding constraint.
Never, in this module
- No account, wager, or funds reach a sportsbook or exchange
- No claim of profit, edge, or strike rate
- No outcome data available before the off
Out-of-sample, cost-aware, merge-gated
Quant research
Distributed training merges candidate parameters across shards, then evaluation compares the candidate against the baseline on the same sealed out-of-sample cases. Promotion requires beating the floor after deflation; in-sample fit alone can never promote a candidate.
What you supply
- Immutable point-in-time market and research corpus
- Artifact and parameter lineage per shard
- Development and out-of-sample case identities kept disjoint
Gates and what they mean
- Out-of-sample coverage floor with abstentions countedCoverage is measured on held-out cases with abstentions counted, so declining a hard case does not raise the figure.
- Deflated performance floor after selection costPromotion requires beating the floor after selection cost is subtracted, not merely looking strong in sample.
- Capital-risk admission before any position intentThe declared capital-risk ceiling is checked before any position intent exists.
Settlement
The horizon closes and the realised return is recorded against the simulated intent, with the deflated score and coverage kept together.
Never, in this module
- No broker, exchange, or capital path is contacted
- No promotion from in-sample fit alone
- No synthetic digest standing in for real lineage
Knowledge-dated, restatement-aware
Financial analysis
Facts are admitted at the date the market could have known them, not the date the report was printed. Restatements and delistings are modelled rather than papered over, and the run ends in an analysis stance with a report projection — never an order.
What you supply
- Knowledge-date admission for every reported figure
- Restatement and survivorship treatment recorded per fact
- Provenance chain from filing to belief
Gates and what they mean
- Mandate-safe stance under declared analysis boundsThe stance has to sit inside the declared analysis bounds, with the mandate as the only authority.
- Settlement and evaluation bound to the same runEvaluation and settlement bind to the same run, so a stance cannot be scored against another decision.
- Corrections append rather than rewrite historyA correction adds a record; the amended and original figures both stay readable.
Settlement
The outcome is recorded against the decision run, including when the run abstained, and evaluation keeps the abstention inside coverage.
Never, in this module
- No order path and no execution adapter
- No recommendation presented as advice
- No figure admitted before it was knowable
Side by side
Sport, quant, and financial analysis.
Same lifecycle, different meaning. The differences live in what the evidence is, how the belief is shaped, which gates bind, and what settlement resolves against.
| Domain | What you supply | Belief | Gates before a decision | Effect today | Settlement |
|---|---|---|---|---|---|
| Sport and racing | Sealed pre-off history and quote snapshots, a full-field candidate and baseline, a mandate with stake and ruin bounds. | Ranked belief across the whole field, renormalised after non-runners, priced against reachable quotes. | Stake bound, ruin policy, out-of-sample coverage floor. | A simulated wager under an idempotency key. No account, wager, or funds reach a sportsbook or exchange. | Published result binds to the run; learning separates belief error from price error. |
| Quant research | Immutable point-in-time market corpus, feature snapshot with per-feature availability, shard lineage, a capital-risk mandate. | Portfolio-level probability with an interval, aggregated from per-name scores. | Coverage floor with abstentions counted, deflated floor after selection cost, capital-risk admission. | A simulated position intent. No broker, exchange, or capital path is contacted. | Horizon closes; the realised return is recorded against the simulated intent with its deflated score. |
| Financial analysis | Knowledge-dated filings, restatement and survivorship treatment, a peer set, an analysis-only mandate. | An analytical stance with an interval that widens when an amendment lands after the belief was formed. | Mandate-safe stance under declared analysis bounds, restatement awareness, analysis-only effect class. | An analysis stance only. There is no order path and no execution adapter. | Outcome recorded against the run, abstentions stay in coverage, and corrections append. |
The same kernel also carries forecast packs where the outcome is a probability rather than a price: Threshold, Weather, Sports event, Analysis.
Next
Gates refuse. That is the point.
Every module above can refuse the run it was given: a missing mandate, a coverage floor, a ruin policy, a prohibited effect. The refusal reference lists each one with the token that shows it.
PostureWhat is not live today
No module here contacts a broker, sportsbook, exchange, provider, or any other counterparty, and none holds funds or accounts. Effects are simulated, and the hosted service is not released. Trust and posture.