Skip to content

Releases: m1r4g3-code/kairos

Kairos v2.0.0 - Sharp-Line Edge + Data-Fed Engine

06 Jun 20:39
@m1r4g3-code m1r4g3-code
v2.0.0
28d9f5e

Choose a tag to compare

View all tags
Kairos v2.0.0 - Sharp-Line Edge + Data-Fed Engine Latest
Latest

The sophistication upgrade: stop guessing, start measuring.

The shift

v1 estimated goals by judgment then priced from scratch — the inputs were guesses. v2.0 adds the proven retail edge: don't out-predict the market — beat the soft book (SportyBet) against the sharpest one (Pinnacle). Where SportyBet pays more than the sharp de-vigged "true" price, that gap is real value.

New (all pure-stdlib urllib, still zero-dependency)

  • engine/edge.py — sharp-line comparison core (reuses market.py de-vig).
  • engine/sources/ — data adapters: The Odds API (Pinnacle + many books), Club Elo, Understat xG, Football-Data.co.uk.
  • engine/backtest.py — soft-vs-sharp replay on real history → ROI / hit-rate / CLV (prove the edge before paying).
  • engine/report.py — new SHARP% column (ODDS% vs SHARP% vs MY%).
  • engine/config.py + .env.example — one free key (The Odds API) via gitignored .env.

Data-fed xG/Elo flow through the existing engine (run.py strengths/elo specs) — no math changed. Keyless except The Odds API (free, 500/mo).

Quality

  • 64 → 104 tests across 4 suites; all run fully offline on committed sample fixtures.
  • CI runs all four on Python 3.10–3.12.

Honest ceiling

Soft-vs-sharp value is real but fragile (limits, line moves, thin margins). v2.0 makes the inputs data-grounded and provable via backtest — a big upgrade over guesses — but profit is never guaranteed.

🤖 Generated with Claude Code

Assets 2
Loading

Kairos v1.0.0 - Disciplined-Decision Aid

06 Jun 20:23
@m1r4g3-code m1r4g3-code
v1.0.0
3d75afc

Choose a tag to compare

View all tags
Kairos v1.0.0 - Disciplined-Decision Aid

First stable release of Kairos as a disciplined football betting decision-aid.

What's in v1.0.0

  • Engine (pure stdlib): Poisson + Dixon-Coles, Elo, Monte Carlo, de-vig, EV + fractional Kelly, lambda fragility test.
  • Report renderer: consistent plain-English card (team names, HOME/AWAY, ODDS% vs MY%, BET/SKIP/TRAP).
  • Ledger: prediction log + calibration (Brier / ROI / CLV), score-based settling, duplicate-ID safety.
  • Tests: 64 assertions across engine + ledger; GitHub Actions CI.

Posture

Recommend-only, keyless. Value-first; willing to say "bet nothing". Honest about its ceiling — the inputs are hand-estimated, which Phase 0 (v2.0.0) addresses with sharp-line comparison + data feeds.

🤖 Generated with Claude Code

Loading