You're betting a capture budget on what PursuitIQ tells you. So the engine is built to be checkable, not just persuasive — every figure traces to a source, every recommendation cites its evidence, and the score is held to its real outcomes.
Scores, runway, balances — all produced by code and traceable to a source and timestamp. The model explains the numbers; it never invents them.
Each verdict points to the computed signals and the solicitation text behind it. An ungrounded claim doesn't ship.
We track win-rate by score band against real results and monitor calibration — and we show you when a band drifts.
Observed win-rate by Pursuit Score band. Calibration is the proof that the score means what it says.
illustrative — bands reconcile to logged outcomes once the flywheel has volume
Funding figures and scores are computed in code — the model explains, never invents. Recommendations cite evidence; ungrounded claims don't ship. Nothing deploys without passing offline evals, and calibration is tracked against real outcomes.
Every tenant row is scoped and isolated; cross-org access is a tested invariant, not a hope.
Encryption in transit and at rest, secrets in a manager, an immutable audit log, and export/delete support.
Commercial AWS today; SOC 2 underway; GovCloud / FedRAMP only when demand pulls for it.
Bring a pursuit you already understand and check our figures against what you know.