[tdub29]

What changed

Bug fix: every CLI command failed on a fresh clone

• The move from src/portal to src/ncaa/portal left every module-level ROOT one directory short. Config, the SQLite path, board output, geo, TrackMan, the bio sources and the CLI's scripts/ paths all resolved under src/. On main, python -m ncaa.cli init --force crashes looking for src/config.example.yaml.
• Fixed in 10 files. New tests/test_paths.py pins each ROOT to the repo root; it fails 10/10 against the old code and passes after the fix.
• README quickstart corrected: the seed workbook is not committed, so its path is passed explicitly.

Bug fix: the pbp scripts could not find their data

• The pbp scripts read and wrote next to themselves, but the CSVs live in data/pbp/ and the text reports in data/pbp/validation/. ncaa.pbp now defines DATA_DIR and VALIDATION_DIR, and 15 scripts use them.
• Checked by re-running merge_season_pdf_inputs: its outputs are byte-identical to the committed files.

Bug fix: SDSU half-innings relabelled as USD (game 6530004)

• _align_to_game_team matched "San Diego St. Aztecs" to "San Diego Toreros" on the shared first word. The batter vote then rewrote 35 of SDSU's 44 batting rows as USD, zeroing USD's defensive battle metrics for that game on every re-run. main reproduces this.
• Exact matches now win, and a looser match only counts when it picks exactly one team. New tests: 3 fail on the old code and all pass with the fix.

Bug fix: the battle calc exited without Playwright

• ncaa_pbp_playwright raised SystemExit at import, which except ImportError could not catch. The battle calc also imported it by bare module name, so package imports silently skipped the outs recompute.
• Playwright is now imported inside the one function that opens a browser, and the parsers are imported by package path. Run without Playwright, the calc reproduces the committed battle report exactly. New regression test fails on the old code.

Third-party data no longer published

• The 2026-09-12 clearance in .gitignore covers USD's own TrackMan and branding, and data/643_exports/README.md already said the 6-4-3 exports were gitignored.
• Removed and ignored: 6-4-3 Charts exports, the PerfectGame, Sidearm and D1Baseball-derived bios, the D1Baseball reference PDF, and the generated recruiting boards (they name athletes and carry fit and likelihood scores).
• Kept: USD TrackMan, the public NCAA play-by-play, and the small warehouse CI fixtures. The board is still built and published locally by scripts/build_html.py; nothing reads the removed files.

Removed from the tree

• The Cursor folder and chat export, debug and scratch files, copy.txt, the X-likes notes, the input-unlock packet, and an empty notebook.

Machine-specific paths

• 17 notebooks, 3 scripts and 5 docs no longer point at one Windows user's Downloads and OneDrive folders.

Write-ups

• README: an author line, a Highlights section drawn only from results already in the repo, an accurate data note, and corrected counts.
• REPORT.md: a plain-English summary, plus a Who reviewed note.
• MODEL-CARD.md: says the reviews were separate AI agent sessions, as the plan logs record, not human peer review.
• Derived pbp columns and description lists regenerated from the current 51-game CSV; the battle report is unchanged.

Verification

• python -m pytest -q: 164 passed
• ruff check src scripts tests: clean
• Quickstart run on a fresh clone; battle calc run with and without Playwright

Deliberately not done

• No git history rewrite. The MLB pre-registration cites commit SHAs as proof of when each test was frozen and scored (5992fdf, ba559be, 6d2503c). Rewriting history would change every SHA and break that evidence. The removed files remain in history.
• stealth.py kept. The D1Baseball live feed depends on it, and the cookie is the program's own subscription. Paste mode remains the no-scrape fallback.