WI-060 · Platform & Contracts · ranking-replay-calibration

Ranking replay & calibration tooling (historical replay, policy comparison, offline outcome metrics)

queued P2 low risk   Owner: Paul   Reviewer: Paul   0% · 0/1 tasks complete

Gates: P4 Requirements: Updated: 2026-06-22

At a glance

CodeWI-060
PhasePlatform & Contracts
Order21 of 93
Story points3
Primary surfaceOffline replay + policy-comparison tooling (no live calls)
Retires
Depends onCandidate ranking engine — deterministic features, fixed-point scoring, priority bands, tie-break, explainability, canonical hash
Blocks
Evidence
queued (P4 tooling)

Goals

Let humans evaluate and evolve ranking policies safely: replay recorded trusted results through current/proposed policies, compare versions with explained movements, and gather offline outcome metrics — never auto-mutating production weights.

Implementation — what to build

Replay

  • Replay tooling over recorded trusted results (no provider calls); compare two policy versions / alternate intents; explain every rank/score/band movement + top-K changes.

Calibration

  • Offline outcome metrics (selection/conversion/manual-override/rank-position); proposed-policy workflow: analysis → replay → human review → explicit publication. No self-modifying live ranker.

Checklist

  • P4 Deterministic candidate ranking: the model returns evidence/component scores; TroveSnap computes the final ranking.

Gates & testing

  • P4 Deterministic candidate ranking: the model returns evidence/component scores; TroveSnap computes the final ranking.
    CODE TEST

Requirements

No linked requirements.

Verify (done when)

Historical replay runs recorded trusted P3 results through current/proposed policy or alternate intent and reports rank/score/band/top-K movement with a reason per movement; policy-comparison report works; outcome/learning-loop metrics collected offline only (publication of any new weight set stays an explicit human step). Per spec §29/§32.13/§35 (tooling rows).

Cost triggers

No paid API / infra spend triggered by this item.

Status & editing

queued   Sprint: P&C Wave 3: Vision Pipeline & Cost

Edit status / sprint on the ★ Live Board → — changes are logged live with who / when / why.

Human-in-the-loop

No human tasks linked.

Findings

None recorded yet.

Concerns

None recorded yet.

Risks

None recorded yet.

Tech debt

None recorded yet.

Tools used

No tool calls recorded.

Files & artifacts

No files / artifacts recorded.

Update log

No log entries yet.