# Cursor Optimization Decisions

**Last Updated:** 2026-04-03

Decision log for locked defaults, threshold changes, and governance updates.

## Decision Template

- Decision ID:
- Date (UTC):
- Owner:
- Change:
- Reason/Evidence:
- Expected KPI impact:
- Rollback criteria:

## Locked Defaults

- Decision ID: `DEC-0001`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Execution model = manual-only (no default auto-run)
- Reason/Evidence: reduces automation risk and enforces deliberate review
- Expected KPI impact: stable quality, slower but safer improvements
- Rollback criteria: if backlog latency becomes unacceptable for 2+ cycles

- Decision ID: `DEC-0002`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Scope model = alternating slices (A/B)
- Reason/Evidence: balances AI-core and docs/process upkeep
- Expected KPI impact: sustained coverage without overloading single runs
- Rollback criteria: if either slice accumulates unresolved backlog >2 runs

- Decision ID: `DEC-0003`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Run budget default = 90-120 minutes
- Reason/Evidence: enough depth for meaningful deltas with manageable effort
- Expected KPI impact: better throughput than quick-pass mode
- Rollback criteria: if average completion rate drops below 60%

- Decision ID: `DEC-0004`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Retention policy = archive + pointer (no destructive deletion)
- Reason/Evidence: preserves history/discoverability and lowers accidental loss risk
- Expected KPI impact: safer consolidation
- Rollback criteria: if archive discoverability noise materially worsens

- Decision ID: `DEC-0005`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Risk policy = stop + carry-forward backlog on risk signals
- Reason/Evidence: prevents risky partial cleanups
- Expected KPI impact: fewer regressions
- Rollback criteria: if stop frequency blocks all progress for 2+ runs

- Decision ID: `DEC-0006`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: CI progression = warn-then-gate staged rollout
- Reason/Evidence: lowers short-term friction while hardening over time
- Expected KPI impact: durable compliance after stabilization
- Rollback criteria: if gates cause persistent false-positive blocking

- Decision ID: `DEC-0007`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Threshold baseline set in `CURSOR_OPTIMIZATION_THRESHOLDS.json`
- Reason/Evidence: explicit numeric control for overlap/freshness regression checks
- Expected KPI impact: predictable regression guardrails
- Rollback criteria: if thresholds prove unstable across 2 stable runs

- Decision ID: `DEC-0008`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: CI stage initialized to Stage 1 (warn-only for new overlap/freshness checks)
- Reason/Evidence: follows warn-then-gate rollout to avoid abrupt workflow friction
- Expected KPI impact: smoother adoption while capturing regression signals
- Rollback criteria: if warnings remain noisy after two manual runs, tune checks before stage promotion

- Decision ID: `DEC-0009`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Add warning-first rule-size/no-glob policy with thresholds from `CURSOR_OPTIMIZATION_THRESHOLDS.json`
- Reason/Evidence: long rule bodies and long no-glob instructions increase routing/token overhead and reduce maintainability
- Expected KPI impact: gradual reduction of large-rule context load without unsafe hard gating
- Rollback criteria: if warning noise prevents prioritization, tighten classifier logic before promotion

- Decision ID: `DEC-0010`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Add warning-first skill-trigger quality policy (must contain clear use boundaries, validation, and fallback)
- Reason/Evidence: thin-router skills reduce accidental over-invocation and preserve predictable routing
- Expected KPI impact: improved speed/reliability from cleaner skill activation and lower duplicate guidance
- Rollback criteria: if policy causes false positives on valid minimal skills, refine checker criteria and rerun

- Decision ID: `DEC-0011`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Pointerize long no-glob rules first; preserve full guidance in archive docs with backlinks
- Reason/Evidence: safety-first way to reduce active rule payload while keeping historical/reference coverage
- Expected KPI impact: fewer active long no-glob rules and lower context volume with no overlap regression
- Rollback criteria: if golden probes show routing degradation, restore active rule bodies from archive

- Decision ID: `DEC-0012`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Promote CI stage default fallback from `1` to `2` in validation workflow
- Reason/Evidence: multiple stable manual runs completed; stage-2 gating can be enabled safely while still allowing repo variable overrides
- Expected KPI impact: stronger integrity enforcement on metadata/links/skill refs/index drift
- Rollback criteria: if stage-2 gating introduces sustained false-positive blocks, revert fallback to `1` and patch validators

- Decision ID: `DEC-0013`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Tighten overlap thresholds to current improved baseline (`exact_glob_overlap_count <= 46`, `blog_json_overlap_count <= 6`)
- Reason/Evidence: regression floor should track latest stable improvements
- Expected KPI impact: prevents silent backsliding after overlap reductions
- Rollback criteria: if intentional safety overlap additions are required, log exception and adjust thresholds explicitly

- Decision ID: `DEC-0014`
- Date (UTC): 2026-04-03
- Owner: Team
- Change: Pointerize `mcp-usage`, `og-images`, and `partner-resources-assets` with archive companions
- Reason/Evidence: reduced long/no-glob warning footprint and active context size while preserving deep references
- Expected KPI impact: improved token efficiency via smaller active rules and higher pointer ratio
- Rollback criteria: restore full in-rule guidance if routing quality degrades on affected workflows