Changelog

Technical release notes for the OGAAT system. The product moved from broad autonomous market hunting to a constrained daily NBA ATS protocol.

v4.0 — Daily NBA ATS protocol Current

Status: live MVP. Scope: one official ATS pick per day, selected from a generated board and tracked publicly.

CHANGED
  - Replaced multi-market opportunity hunting with one daily NBA ATS decision.
  - Added a deterministic official-pick path from the qualified edge board.
  - Added public board output so the final pick can be evaluated against alternatives.
  - Added account-truth bankroll reporting: cash, open position, account value, P&L.
  - Added settlement-aware ledger states: pending, win, loss, push/void where applicable.
  - Added press-mode sizing when system state qualifies.

SYSTEM BEHAVIOR
  - Generate the daily ATS board before tip-off.
  - Select one official pick.
  - Size the position from bankroll/state rules.
  - Execute through Kalshi when a matching market is available.
  - Publish pick, execution state, and settlement record.

KNOWN LIMITATIONS
  - Public source links are hidden while the repo is cleaned up.
  - Model artifacts exist locally but are not yet exposed as a polished public proof trail.
  - Settlement and ledger correctness matter more than feature breadth during launch.

v3.x — Live momentum trading Mar–Apr 2026

Status: deprecated. Scope: live market monitoring and opportunistic entries.

CHANGED
  - Added live-market monitoring for sports prediction markets.
  - Tested momentum/dislocation logic when prices appeared to lag new information.
  - Expanded execution attempts beyond pregame-only decisions.

WHAT FAILED
  - Opportunity volume created noise instead of discipline.
  - Execution timing became a bigger driver than model quality.
  - Too many trades made results hard to audit cleanly.
  - A machine crash interrupted the live cycle and exposed operational fragility.

REMOVED / REPLACED BY v4
  - Broad live-trading loop.
  - Multi-entry same-day behavior.
  - Any workflow where activity could masquerade as edge.

v2.x — NCAA basketball edge engine Mar 2026

Status: superseded. Scope: sports-specific modeling using stronger basketball data.

CHANGED
  - Moved from general market guessing into basketball-specific modeling.
  - Added external team-quality inputs including KenPom and BartTorvik.
  - Compared model-derived views against sportsbook / market prices.
  - Tested tournament-style edge discovery against available lines.

DATA SOURCES
  - KenPom
  - BartTorvik
  - Odds API
  - Public books / Kalshi where available

LESSON CARRIED FORWARD
  - Better data helped, but the product still needed a narrower operating contract:
    one sport, one cadence, one ledger, one official decision.

v1.x — Autonomous market hunter Early 2026

Status: retired. Scope: search for stale prices across many market types.

INITIAL SCOPE
  - CPI / inflation threshold markets.
  - Oil and gas markets.
  - Index / S&P-style markets.
  - One-off sports opportunities.

CORE IDEA
  - Read external signals.
  - Identify where public markets looked slow to reprice.
  - Bet when available prices appeared stale.

WHY IT DID NOT SCALE
  - Too many market types made evaluation messy.
  - Each market required different data, timing, and settlement logic.
  - The system needed a repeatable game, not a pile of unrelated guesses.

CARRIED FORWARD
  - The stale-price thesis.
  - The requirement that every decision should leave a public audit trail.

Roadmap

Near-term work is boring on purpose: make the daily receipt loop reliable before adding scope.

NEXT
  - Harden pregame artifact save and proof-capture path.
  - Improve public ledger clarity around settlement, fees, and account value.
  - Add weekly recap: record, P&L, biggest miss, and model change if any.
  - Add recursive learning notes only when a change actually affects the model.

LATER
  - Public source release after repo cleanup.
  - Better calibration and feature testing.
  - MLB path.
  - NFL path.

PRODUCT RULE
  - One official pick per day remains the constraint.