Deterministic sports analysis
Hollersports
A deterministic, provenance-tracked sports analytics engine that prevents slate leakage and produces reproducible outputs.
On hold · v0.5.0 advisory beta
What it is
Hollersports is a deterministic sports analysis engine that isolates every slate and fingerprints its inputs.
Same games and inputs produce the same outputs — no silent bleed between runs.
In short
State isolation. Every slate starts from init_new_slate_state with no artifacts from previous runs.
Provenance. Inputs and intermediate artifacts are fingerprinted—source games and datasets, category, game context, and calibration—so analysis can be audited end to end.
Input validation. Before processing, state is asserted against the current games, datasets, and provider so stale state cannot be reused by accident.
Controlled calibration. Cross-slate calibration is opt-in and explicit; the default is discard, not silent merge.
A companion feedback loop can ingest expected outcomes and game results, compute residuals, and update model priors with append-only ledgers. Engine architecture and validation live in the product repository.
Built with
- Pythonthe analysis engine and command-line tools
- FastAPIthe local web service behind the workbench
- Pydanticchecks that every data record is well-formed
- JSON Schemapublished formats for every data record
- Next.js and Reactthe local operator workbench in the browser
- TypeScriptthe workbench interface code
- ESPN public scoreboardsfree game schedules and final scores
- pytestthe automated test suite
Changelog
Milestones in plain English, newest first, to Nov 7, 2026
- Added a first-run age and paper-only acknowledgment, plus privacy, terms and responsible-gambling pages.
- Redesigned the local workbench around Today, Paper book and Health views.
- Froze v0.5.0 advisory beta for field testing, with a repeatable field-test check and receipt.
- Released the v0.3.0 advisory beta: replayable sample game days, free public scores and paper-only simulation.
- Added a feedback loop that compares expected outcomes with game results and keeps append-only records.
- Every game day now starts from a clean state, so nothing carries over between runs.
Roadmap
What comes next. Plans, not promises, and no dates.
- PlannedPublish the analysis engine as an installable Python package.