feat: WS5 — service bundles (config builder → widget packages → Overview) (#138)

test(e2e): Playwright deep-dive suite (local-deep + staging-smoke)

docs: per-folder CLAUDE.md files + fix README for the SaaS

chore(dev): browser-test harness for the dashboard (seed + Playwright smoke)

feat(audit): step-by-step audit + ablation study tooling

perf(exp4): trainer memory + visibility — freeze layers, batch 1, step logging, mid-epoch ckpt

fix(exp4): trainer crashed at loss — build mask/class labels explicitly

feat(exp4): verdicts_to_exclude.py — review verdicts -> trainer exclude-file

fix(exp4): review-page export works in the artifact sandbox (mobile)

revert(exp4): labeler back to 648016a baseline — drop patio class + shadow repairs

feat(exp4): patio/deck class + shared taxonomy + frozen review legend

fix(exp4): per-round verdict namespace on review pages

feat(exp4): lighting-invariant label repairs from owner round-1 review

fix(exp4): review-page export works on mobile — verdicts shown in a copyable box

feat(exp4): tune SAM-3 fusion on smoke evidence — dormant turf is LAWN

feat(exp4): --purge-lidar flag — cap exporter disk to ~1 cluster of tiles

feat(exp4): Phase 2/3 engine — crop exporter, SAM-3 auto-labeler, review page, trainer

feat(exp4): DOGIS export probe — GSD + season verified against the live service

chore(exp4): MPS smoke test v2 — self-contained copy-back report for the owner's Mac run

feat(exp4): Phase 1 bake-off — LAWN_MODEL override, EoMT-DINOv3 candidate, findings

feat(eval): labeled before/after eval harness + exp5 green-reclaim results

Revert "Merge pull request #29 from sergioreyesmle/feat/demo-asset-generator"

feat(demo): sanitized public-showcase asset generator

refactor: ports + adapters + estimation-strategy chain (Douglas only, byte-identical)

feat: county 2025 display imagery + parallel-offset curb extension

refactor: restructure into src/lawn_estimator package