Context belongs in your research stack.
Explore the engine’s internal research interfaces and the direction for a future public API. Preserved evidence and explicit availability are central to the design.
A contract that can say
“not available.”
Missing evidence must survive the trip from data source to interface. Null values represent unavailable qualification; they are never silently converted into zeros.
The example is a proposed public presentation contract. It is not the existing engine response shape.
Register developer interest →{
"symbol": "BTC-USD",
"mode": "illustrative_proposed_contract",
"regime": "TransitionUncertain",
"fmsi": null,
"persistence": null,
"liquidity": null,
"riskScore": null,
"opportunityScore": null,
"analyticalUseAllowed": false,
"recommendationsEnabled": false,
"timestamp": "sample"
}Phase 8 internal research contracts
Implemented local engine routes, protected by its origin, loopback and configuration guards. These are documentation, not endpoints hosted by this website.
/api/shadow/statusRead monitoring history and the separate qualification state. Current qualification is NotStarted; monitoring is not qualification credit.
/api/shadow/paper/{id}Read an immutable paper portfolio version and its retained source bindings. Simulated balances are separate from owner accounting.
/api/shadow/researchRead saved daily research captures with exact known-at cutoffs, observed features and unavailable conditions.
/api/shadow/forecasts/{id}Read a frozen forecast population with discovery coverage, exclusions and actual freeze time.
/api/shadow/forecastsFreeze server-resolved evidence using id, researchId and instrumentId. Caller-supplied prices, fits and historical freeze times are not accepted.
/api/shadow/cost-populationsRegister explicit account, notional and friction assumptions against a forecast and matching saved fee assumption. Registration uses actual server time.
/api/shadow/calibrations/{id}Read retained calibration evidence or explicit blockers. Missing or noncausal inputs do not become calibrated probabilities.
Existing internal regime contracts
Implemented in the source application. Paths are documentation, not hosted marketing-site endpoints.
/api/features/regimes/fits/{productId}Preview a fit with trainingEndsAt and knownAt cutoffs.
/api/features/regimes/fitsSave productId, trainingEndsAt, knownAt, and expectedRunId.
/api/features/regimes/evaluate/{fitId}Evaluate a saved fit using startsAt, decisionDays (1–30), and knownAt.
/api/features/regimes/runsSave fitId, startsAt, decisionDays, knownAt, and expectedRunId.
/api/features/fmh/{productId}Preview experimental FMH diagnostics for 256, 512, or 1,024 returns.
/api/features/runs/{id}Read a retained feature record at a registry knowledge cutoff.
Explicit temporal context
Input knowledge, training completion, and registry visibility are distinct. A historical date evaluated with today’s knowledge is retrospective research.
Preserved numerical meaning
Source financial decimals and many derived quantities are strings in engine JSON. Clients must preserve these contracts; there is no generic float coercion.
Bounded, verifiable requests
Regime saves reject client-supplied results and unknown fields. Changed previews return 409; inconsistent evidence can return 422.