You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A miner whose own recent PRs are trending toward closed/rejected outcomes on a given repo should slow down or pause on that repo rather than continuing to fire submissions at the same rate — a self-correcting throttle distinct from the global/per-repo rate limit, driven by the miner's own historical merge/close ratio rather than a fixed cadence. This is bounded to the miner's own local history and degrades/recovers cadence rather than imposing a permanent block. Open, scored contributor work, provided the fail-open-only-on-insufficient-sample-size behavior is thoroughly tested.
Deliverables
A selfReputationThrottle(recentOwnOutcomes, thresholds) function the Governor chokepoint consults before an open_pr/file_issue action, reading the miner's own local outcome history (not any shared/cross-fleet data — per the architecture skeleton, this stays 100% local)
A degrading cadence: as the miner's own recent closed/rejected ratio on a repo rises past a configured threshold, allowed submission frequency to that specific repo drops (not a hard permanent ban — a repo with an improving ratio recovers cadence)
Configurable thresholds sourced from .gittensory-miner.yml, conservative built-in defaults
Tests: a clean track record allows normal cadence, a degrading ratio measurably throttles cadence, an empty/insufficient history (new miner, new repo) does not falsely throttle (fails open only on genuinely insufficient sample size, documented explicitly), and a recovering ratio restores cadence
Throttle decisions recorded to the governor ledger with the outcome ratio that triggered them
References
src/signals/reward-risk.ts (buildContributorRewardRiskStrategy, buildMaintainerNoiseReport) — the existing outcome-history-driven risk/strategy vocabulary this throttle's shape should stay consistent with, adapted to the miner's own local-only history
The Governor chokepoint (companion maintainer issue) — this check gates submission write paths alongside the self-plagiarism throttle
The append-only governor ledger (companion foundation issue) — the local source of the miner's own outcome history this throttle reads
A miner whose own recent PRs are trending toward closed/rejected outcomes on a given repo should slow down or pause on that repo rather than continuing to fire submissions at the same rate — a self-correcting throttle distinct from the global/per-repo rate limit, driven by the miner's own historical merge/close ratio rather than a fixed cadence. This is bounded to the miner's own local history and degrades/recovers cadence rather than imposing a permanent block. Open, scored contributor work, provided the fail-open-only-on-insufficient-sample-size behavior is thoroughly tested.
Deliverables
selfReputationThrottle(recentOwnOutcomes, thresholds)function the Governor chokepoint consults before anopen_pr/file_issueaction, reading the miner's own local outcome history (not any shared/cross-fleet data — per the architecture skeleton, this stays 100% local).gittensory-miner.yml, conservative built-in defaultsReferences
src/signals/reward-risk.ts(buildContributorRewardRiskStrategy,buildMaintainerNoiseReport) — the existing outcome-history-driven risk/strategy vocabulary this throttle's shape should stay consistent with, adapted to the miner's own local-only history