OutlierBaseline

Use the web draft board

Open https://outlierbaseline.com/ in any modern browser. The website supports every team-count, QB, PPR, and TE-premium ranking combination, persistent drafted-player markers, column filters, sorting, CSV export, rookie and current-injury labels, and source-linked player intel. Click a player name to see current roster status, depth position, material arrivals/departures, injury data, and matched ESPN headlines. Default market values use only ADP sources that match the selected format; unsupported formats display No market until a matching private file is imported. Sources and ADP SD show evidence count and standard deviation. Imported data stays in that browser, is bound to the selected format, and is never uploaded. A separate Kicker & D/ST page compares the current Sleeper, ESPN, and MyFantasyLeague markets. Desktop downloads remain available under GitHub Releases.

Automatic site updates (no API key)

The factual timeline uses public nflverse roster, depth-chart, and weekly injury releases plus ESPN's current injury designations and matched NFL headlines. Players are joined by stable NFL ID when available, with name plus position as the fallback; ambiguous same-name headlines are skipped instead of guessed. All current primary and secondary injuries appear in the rankings. Questionable and probable designations remain visible without changing the market-based draft tag, while Out, Doubtful, injured-reserve, suspension, and exempt-list situations become RISK. It does not copy Rotoworld blurbs or require an AI key.

The website refreshes direct Sleeper half-PPR, ESPN PPR, and MyFantasyLeague recent-redraft ADP, the separate Kicker & D/ST market, all 60 rankings, and the factual player-news timeline every day at midnight and noon Eastern time. Yahoo's official developer access requires approval and OAuth, so the updater keeps the last authorized Yahoo snapshot instead of scraping a protected page or erasing that column. The freshness line above the board shows the date for each source.

Run the complete update locally:

python refresh_draft_board.py --keep-stats

You can also open Actions → Update site data → Run workflow on GitHub at any time. The scheduled workflow handles daylight-saving changes automatically and publishes changed ADP, rankings, and news files to the website.

Update AI player intel

The public website never receives an OpenAI API key. A private GitHub Action researches the latest role changes, arrivals, departures, injuries, and value-changing news, then publishes date-stamped reports to docs/data/player_intel.json.

One-time setup:

1. Create an OpenAI API key with API billing enabled.
2. In this GitHub repository, open Settings → Secrets and variables → Actions.
3. Choose New repository secret, name it OPENAI_API_KEY, and paste the key there.
4. Open Actions → Update player intel → Run workflow.
5. Start with 50 players. Run it again for the next stale group, or enter one exact player name for a focused update.

The action commits successful reports back to the website automatically. OpenAI API usage is billed separately from a ChatGPT subscription, and web searches can add tool-call costs. Reports include clickable sources and should be checked before making a draft decision.

To update locally after setting OPENAI_API_KEY as an environment variable:

python update_player_intel.py --limit 50
python update_player_intel.py --player "Josh Allen"

A football-intelligence application that combines NFL season stats, rookie betting-line projections, and five-source market ADP. It produces interactive rankings for 8–16 teams, 1QB/2QB, Standard/Half/Full PPR, and optional TE premium scoring. Market +/- shows projected fantasy points above or below the same-position regression expectation at a player's consensus ADP; VORP remains the separate comparison with the replacement player. Sources counts the available ADP feeds and Spread measures their disagreement. Draft tags use Market +/-: TARGET is +50 points, VALUE is +25 to +49.9, FAIR is -19.9 to +24.9, and REACH is -20 or worse. A current risk signal overrides the market tier with RISK; a team change with no other material update becomes NEW TEAM.

Run the desktop application

python -m app.desktop

The application displays the rankings directly. It supports drafted-player tracking, reversible column filtering, header sorting, live league-format changes, data refreshes, and Excel export. Type a filter, then press Enter or click Apply. Show All always restores the complete player pool.

Refresh from the command line

Reuse the saved stats while refreshing the four direct ADP feeds:

python refresh_draft_board.py --keep-stats

When the refresh runs from the source repository, it also updates docs/data/rankings.json for the website. Pass --saved-adp only when you are offline and want to reuse every saved ADP value. Pass --skip-workbook for a faster website-only refresh.

The legacy manual-table importer remains available for a page or saved HTML file you are authorized to use:

python refresh_draft_board.py --adp-source "URL-OR-PATH"

To promote a user-downloaded Yahoo snapshot into the site-wide default, use a CSV containing Player or Name, Position or Pos, and Yahoo or Y!. The updater replaces only Yahoo, preserves the direct public feeds, rebuilds every board, and writes the new source date:

python refresh_draft_board.py --yahoo-snapshot "PATH-TO-FILE.csv" --keep-stats --skip-workbook

The workbook is written to output/<projection season>_preseason/ProjectFootMoneyball_Draft_Board.xlsx.

Build the standalone Windows release

.\scripts\build_windows.ps1

The script creates releases/current/ProjectFootMoneyball-Windows.zip. Recipients extract the complete folder and run Project Foot Moneyball.exe; Python is not required on their computer.

Build the standalone macOS release

A real macOS application must be built on macOS. The Build desktop apps workflow under GitHub Actions creates separate ZIPs for Apple Silicon and Intel Macs, plus the Windows ZIP. Run it manually for test builds. Pushing a version tag such as v0.2.0 also creates a pre-release and attaches all four downloads automatically.

The workflow also creates ProjectFootMoneyball-All-Platforms.zip. This is the easiest download to share: it contains all three applications and a short START HERE guide so the recipient can choose after downloading.

On a Mac, the same build can be run locally:

bash scripts/build_macos.sh

Generated spreadsheets and saved draft state are stored under Documents/Project Foot Moneyball on macOS.

Active project layout

config.py                         Season, league, and folder settings
main.py                           Compatibility command-line entry point
refresh_draft_board.py            Stats/ADP refresh command
update_player_intel.py             Source-linked AI player news updater
update_player_news.py              No-key roster, depth-chart, injury, and headline updater
app/                              Desktop UI, board service, workbook export
pipeline/                         Pipeline orchestration
data_fetcher/                     Active ADP and rookie projection importers
resources/                        Column-cleaning configuration
scripts/                          Windows and macOS build tooling
.github/workflows/                Automated Windows/macOS release builds
releases/current/                 Current packaged desktop releases
stat_utils/pipeline_cleaning.py    Final dataset cleaning policy
stat_utils/data_analytics/         Fantasy scoring, regression, and rankings
tests/                            Active automated tests
archive/                          Superseded, recoverable versions

Tests

python -m pip install -r requirements-dev.txt
python -m pytest -q

Pytest is configured to ignore archived copies and generated build folders.

Data sources

• NFL season totals: nflverse player-stat releases
• ADP: direct Sleeper, ESPN, and MyFantasyLeague feeds, plus the last authorized Yahoo snapshot; market calculations use only format-matched sources
• Rookie projections: betting-line inputs blended with historical position profiles

Review each provider's usage and redistribution terms before distributing refreshed source data.