feat(scoring): per-repo time-decay hyperparameters + go live (#703)

[JSONbored]

What

Completes #703: time-decay now resolves each repo's own curve and goes live (SCORING_TIME_DECAY_ENABLED=true).

• The curve comes from the code — defaults are parsed live from upstream's constants.py into the scoring snapshot (the TIME_DECAY_* constants are now modeled).
• …but changes per repo — each repo's scoring.time_decay overrides overlay the defaults per-field, mirroring upstream's resolve_time_decay. GITTENSOR_REGISTRY_URL is master_repositories.json, so these are the same overrides upstream reads → gittensory's projection matches upstream's actual scoring.

How

• registry/normalize.ts: parse the nested scoring.time_decay block per repo (grace_period_hours, sigmoid_midpoint_days, sigmoid_steepness, min_multiplier) → RegistryRepoConfig.timeDecay; absent/invalid fields stay null (→ fall back to the default at resolve time).
• scoring/preview.ts: resolveTimeDecay(constants, overrides) overlays per-field (override ?? default); calculateTimeDecay(prAgeHours, constants, overrides?) applies the resolved per-repo curve; computeScoreCore passes repo.registryConfig.timeDecay.
• wrangler.jsonc: SCORING_TIME_DECAY_ENABLED=true — the go-live switch (deploys on merge via Workers Builds).

Real example

JSONbored/gittensory already sets scoring.time_decay: { grace_period_hours: 24, sigmoid_midpoint_days: 10, min_multiplier: 0.05 } upstream — so on this repo an 18h-old PR is still fresh (24h grace), where a default-config repo (12h grace) would have started decaying. sigmoid_steepness (unset) falls back to the upstream default 0.4.

Safety

A fresh PR is unaffected (decay 1.0), and no caller passes prAgeHours today — so going live changes nothing until a projection explicitly asks for an aged-PR score. Per-repo curves then apply automatically.

Tests

scoring.time_decay parsing (partial / empty-object / absent / array), resolveTimeDecay per-field overlay, calculateTimeDecay with overrides (grace + midpoint), and the preview applying each repo's resolved curve. 97% coverage held; typecheck + UI lint + OpenAPI clean.

[ghost]

Note

Gittensory Gate skipped

PR closed before full evaluation. No late first comment was created.

Signal	Result	Evidence	Action
Gate result	⚠️ Skipped	#733 is no longer open.	No action.

---

💰 Earn for open-source contributions like this. Gittensor lets GitHub contributors earn for the work they already do — register to start earning →.

Checked by Gittensory, a quiet PR intelligence layer for OSS maintainers.

[ghost]

gittensory · advisory review

Reviewed 6 changed file(s) — two independent AI reviewers.

Suggested action: ✅ Safe to merge — both reviewers found no blocking issues.

Reviewer A · gpt-oss-120b — recommends ✅ merge
This PR introduces per‑repository time‑decay hyper‑parameters, adds parsing of those overrides from the registry, updates the scoring preview to apply them, and includes comprehensive unit tests and an env‑var switch. The changes are well‑scoped and type‑safe.

Suggestions

• Confirm that the config variable used in computeScoreCore is the repository config that now includes timeDecay; if its name differs, adjust the reference.
• Add a unit test verifying that the SCORING_TIME_DECAY_ENABLED environment flag correctly disables the decay logic in previews.
• Run the full test suite to ensure no other modules are affected by the new optional timeDecay field in RegistryRepoConfig.

Worth double-checking

• If config is undefined in some code paths, config?.timeDecay will be undefined, which is intended, but ensure no runtime errors arise.
• Overrides with non‑numeric values should correctly fall back to defaults; verify edge cases with NaN or Infinity.

Reviewer B · nemotron-3-120b-a12b — recommends ✅ merge
The PR implements per-repo time-decay hyperparameters for scoring, allowing repository-specific overrides of grace period, sigmoid midpoint, steepness, and minimum multiplier. It updates scoring logic, registry normalization, types, and adds comprehensive tests. The feature is gated by SCORING_TIME_DECAY_ENABLED (set to true) and appears correct, well-tested, and consistent with the project's deterministic scoring goals.

Suggestions

• Add a comment in resolveTimeDecay clarifying that null overrides are treated as 'no overrides' (already covered by pickOverride logic).
• Consider adding a test for extreme override values (e.g., negative gracePeriodHours) to ensure they fall back to defaults.
• In wrangler.jsonc, the comment mentions 'Owner-reviewed; this is the go-live switch'—ensure this matches the project's governance process for feature flags.

Worth double-checking

• Verify that the time decay multiplier calculation remains deterministic and safe for all possible override combinations (e.g., zero or negative values).
• Ensure the feature flag SCORING_TIME_DECAY_ENABLED is respected in all code paths (it is gated in computeScoreCore via input.applyTimeDecay, which likely derives from this flag).
• Check that the registry normalization handles malformed time_decyn objects gracefully (already covered by tests).

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.