# Cursor Optimization Scorecard

**Last Updated:** 2026-04-03  
**Primary KPI:** Token/cost reduction  
**Secondary KPI:** Speed/latency  
**Mode:** Phased safe cleanup

## Baseline (Wave 0)

- Rules count: `128`
- Always-apply rules: `1` (`global.mdc`)
- Rules without globs: `17`
- High exact-glob overlap hotspots: API, compare pages, tools/blog JSON, broad page globs
- Docs inventory total: `6888`
- Redundant filename patterns: `NEXT_STEPS (100)`, `FINAL_STATUS (16)`, `README (159)`
- Known reliability issue: local validator dependency gap (`PyYAML` missing outside prepared env)
- Known quality issue: stale/non-existent rule references in skills/docs

## Latest Snapshot

- `rules_count`: `128`
- `always_apply_count`: `1`
- `rules_without_globs`: `14`
- `exact_glob_overlap_count`: `46`
- `blog_json_overlap_count`: `6`
- `pointer_rules_count`: `38`
- `active_full_rules_count`: `90`
- `skill_broken_rule_refs`: `0`
- `docs_hot_path_count`: `5`
- `archive_noise_count`: `155`
- `docs.total_docs`: `6909`

## Regression Thresholds

- `exact_glob_overlap_count <= 46`
- `blog_json_overlap_count <= 6`
- `skill_broken_rule_refs == 0`
- `always_apply_count == 1`
- `volatile_docs_freshness`: all pass under configured max age (`CURSOR_VOLATILE_DOCS_CONFIG.json`)

## Wave Tracking

| Wave | Focus | Status | KPI Delta | Notes |
|---|---|---|---|---|
| 1 | Dead refs + validator reliability | Completed | `skill_broken_rule_refs: >0 → 0` | Added skill validator, CI wiring, print-only validator mode |
| 2 | Rule overlap + pointerization | Completed | `exact_glob_overlap_count: 56 → 50` | Reduced broad globs, pointerized long no-glob rules, and merged compare/API overlap safely |
| 5 | Next-step overlap reduction (blog + dashboard) | Completed | `exact_glob_overlap_count: 50 → 48` | Narrowed advisory blog JSON rules and consolidated dashboard runtime routing into `dashboard.mdc` |
| 3 | SSOT docs + archive cleanup | Completed | N/A (governance quality) | Added archive index, changelog pointers, hot-path reading order |
| 4 | `.cursorignore` tuning + final benchmark | Completed | ~9.34 MB known noisy artifacts excluded | Added targeted high-noise exclusions + workflow safety verification |
| 6 | Continuous manual optimization program | Completed | Governance + CI staging + new KPI instrumentation | Added runbook/log/backlog/risk/decisions, threshold checks, snapshot metrics, CODEOWNERS |
| 7 | Rule/skill quality hardening | Completed | `rules_without_globs: 17 → 14`, `pointer_rules_count: 32 → 35` | Added rule/skill census, warning-first rule-size and skill-trigger checks, normalized 7 skills, and pointerized long no-glob rules |
| 8 | Pending execution pass (stage + warning triage) | Completed | `pointer_rules_count: 35 → 38`, no-glob warning set reduced | Pointerized `mcp-usage`, `og-images`, `partner-resources-assets`; promoted CI default stage fallback to 2 after stable runs |

## Run Trend (Last 8)

| Run ID | Date | Slice | exact_glob_overlap_count | blog_json_overlap_count | skill_broken_rule_refs | Status |
|---|---|---|---:|---:|---:|---|
| OPT-RUN-2026-04-03-01 | 2026-04-03 | A | 48 | 7 | 0 | improved |
| OPT-RUN-2026-04-03-02 | 2026-04-03 | B | 48 | 7 | 0 | flat |
| OPT-RUN-2026-04-03-03 | 2026-04-03 | A | 48 | 7 | 0 | improved |
| OPT-RUN-2026-04-03-04 | 2026-04-03 | A | 46 | 6 | 0 | improved |
| OPT-RUN-2026-04-03-05 | 2026-04-03 | A | 46 | 6 | 0 | improved |

## KPI Status Notes

- `exact_glob_overlap_count`: improved (from prior 48 to 46).
- `blog_json_overlap_count`: improved (from 7 to 6) by removing broad cross-surface FAQ routing from blog JSON and keeping blog-specific FAQ routing.
- `skill_broken_rule_refs`: flat at 0 (healthy).
- `always_apply_count`: flat at 1 (target preserved).
- `rules_without_globs`: improved (from 17 to 14) via safe pointerization of long no-glob guidance.
- `pointer_rules_count`: improved (from 32 to 35) with no overlap regression.
- `docs.total_docs`: degraded vs historical baseline due planned governance artifacts added; acceptable and documented.

### Degraded KPI Explanations

- `docs.total_docs` increased (`6888 -> 6909`) because this implementation intentionally introduced permanent governance docs and snapshot artifacts required by the continuous program.

## Required Checks Per Wave

1. `make cursor-integrity`
2. `make cursor-benchmark`
3. Golden-task regression pass (see `CURSOR_GOLDEN_TASK_SET.md`)
4. Verify no unintended behavior regressions in active routing docs and skills