Historical model audit

Forecast Replay

Return to a historical cutoff and reconstruct what 1stSeen could have predicted—using only evidence the system actually knew by then. Every run is computed by the worker’s leak-safe backtest when you ask for it.

Future evidence is sealed

The held-out opening and every post-cutoff observation are scoring data only. They cannot enter the forecasting component.

What can be replayed

Held-out targets with a defensible actual interval

Roles with two or more recorded openings: 22
Excluded: only observed_by evidence: 0
Replay candidates: 22
held out on an exact opening: 22
held out on a bounded opening: 0

An archive capture proves a role was visible by then, not that it opened then, so an observed_by opening is never scored as ground truth.

Latest persisted backtest

Why targets could not be scored

No backtest has been persisted yet. Run firstseen backtest to record which targets the leak-safe evaluator can score and why the rest are skipped.

Choose a held-out target

Showing 21–22 of 22

Palantir
Software Engineer Intern Defense Tech
Held outAug 5, 2024exactexact
No backtest persisted
Palantir
Software Engineer Intern Defense Tech
Held outAug 5, 2024exactexact
No backtest persisted

Choose the historical vantage point

Palantir · Software Engineer Intern Defense Tech · held out Aug 5, 2024 exact

Forecast cutoff