roadmap: upstream gittensor scoring-sync hardening + parity

[JSONbored]

Parent: #525 (Wave 1) · relates pillar #529 (foundations) and #544 (miner intelligence).

Why this matters

gittensory's scoring/outcomes/planning are only as accurate as its sync with upstream entrius/gittensor. The 2026-06-17 audit found the live saturation-model constants do refresh hourly and win over local defaults — but there are 2 real correctness bugs, an entire unmodeled dimension (issue-discovery/issue-spam), missing constants, and drift that's reported but never acted on. This underpins everything downstream — especially the PMF conversion hook (#800) that shows a contributor 'you'd score ~X'.

Correctness (→ M0, fix now)

• [Bug]: treasury emission-share constant name mismatch never syncs from upstream #806 treasury emission-share name mismatch never syncs (frozen at local default)
• [Bug]: MAX_CONTRIBUTION_BONUS fallback is 5x upstream (25 vs 5) #807 MAX_CONTRIBUTION_BONUS fallback 5x upstream (25 vs 5)

Parity + hardening (→ M1)

• feat(scoring): model (or explicitly scope out) the issue-discovery / issue-spam dimension #808 model (or scope out) the issue-discovery / issue-spam dimension — 10 stored-but-unused constants
• feat(scoring): pull + model upstream constants not currently fetched #809 pull + model constants not fetched (open-PR review collateral, non-code line cap, default issue-discovery share)
• feat(scoring): sync hardening — staleness gate + robust constant parser #810 sync hardening — staleness gate + robust constant parser
• feat(scoring): act on unmodeled-constant drift (not just report it) #811 act on unmodeled-constant drift (not just report it)
• chore(scoring): single-source fallbacks + drop dead density branch #812 single-source fallbacks + drop dead density branch

Upstream watch

• track: upstream entrius/gittensor planning/scoring tooling (#1460 gitt miner advisor, #630 scoring change) #813 track upstream tooling/scoring changes (feat(mcp): expose maintainer-lane triage via gittensory_get_maintainer_lane #1460 gitt miner advisor/scan, test: verify reviewbot unified review (throwaway) #630 scoring-weight change)

How sync works today (for context)

Two pipelines: A src/scoring/model.ts (refreshScoringModelSnapshot → constants.py + language weights, hourly, feeds the live preview) and B src/upstream/ruleset.ts (semantic-hash drift diff that auto-files GitHub issues, gated by GITTENSORY_AUTO_FILE_DRIFT_ISSUES). Pipeline A reports drift but doesn't act; Pipeline B acts but on a different signal — #811 bridges them.

Outcome

gittensory's scoring is provably in parity with upstream, drift is auto-surfaced, and a deliberate build-vs-consume decision is made on gitt-native planning (#1460).