feat(scoring): model (or explicitly scope out) the issue-discovery / issue-spam dimension

[JSONbored]

Found in the 2026-06-17 scoring-sync audit. Largest unmodeled gap.

Problem

10 of 27 upstream constants are parsed/stored but never read by any scoring logic (model.ts:18-24): MIN_VALID_SOLVED_ISSUES, MIN_ISSUE_CREDIBILITY, MIN_TOKEN_SCORE_FOR_VALID_ISSUE, OPEN_ISSUE_SPAM_BASE_THRESHOLD, OPEN_ISSUE_SPAM_TOKEN_SCORE_PER_SLOT, MAX_OPEN_ISSUE_THRESHOLD, ISSUE_TREASURY_EMISSION_SHARE, PR_LOOKBACK_DAYS, TEST_FILE_CONTRIBUTION_WEIGHT, MIN_VALID_MERGED_PRS. These cover the issue-discovery / issue-spam / mirror-validity lane — gittensory ingests them but does not model their effect, so issue-lane planning/outcomes are incomplete.

Deliverable

• Decide per constant: model its scoring effect (issue-discovery lane, issue-spam thresholds, merged-PR validity floor, test-file weighting, lookback window) OR annotate as intentionally out-of-scope so it stops reading as 'synced'
• Wire the modeled ones into score preview + decision pack + reward-risk

Outcome

The issue-discovery lane (a real gittensor reward channel) is correctly scored — affects miner planning #544 and the conversion hook accuracy. Milestone M1.