The problem
A number on a screen is not an audit trail. Prices move, views change, feeds go stale and some results never arrive. If a tracker silently repeats a stake or mistakes a missing result for a loss, its performance report becomes misleading.
The system
The extension reads supported visible markets from owned capture sessions. It checks freshness and market identity, saves meaningful revisions, then creates a separate simulated portfolio. The record includes price, score, phase, source, model and all outcome probabilities.
The portfolio reserves integer-penny stakes for each ticket. Every overlap counts. A result can settle all matching revisions and accumulator legs without double-crediting a return. Voids, pending outcomes and withheld forecasts stay distinct.
A real parsing bug
A live football page displayed a global Full Time filter, while an individual row actually showed Next Goal. Treating the filter as authoritative would create a forecast for the wrong market. A captured public DOM fixture now checks that row-local market labels take precedence and ambiguous outcomes are withheld.
What was built
- Manifest V3 extension with a movable research Guide and persistent local history.
- Live/upcoming sports observation, market identity checks and stale-data handling.
- £1,000 virtual portfolio with singles and 2–6-leg accumulator experiments.
- Probability calibration evaluated on later data, with explicit unproven-performance status.
- Separate game research desks, records and Excel-compatible exports.
- This Cloudflare-hosted demo, reviewed download package and public release evidence.
What the evidence supports
| Evidence | What it establishes |
|---|---|
| 922 passing regression checks | Tested software behaviour for the 17.18.0 release; not accuracy or profit. |
| Captured public DOM fixtures | Specific observed layouts and parsing regressions; not every game or provider. |
| Synthetic browser trials | Accounting, UI, export and persistence under fabricated scenarios. |
| 320 / 390 / 840 px checks | Paper portfolio wrapping at the tested widths. |
| Explicit final-result settlement | Fixture-tested exact football full-time handling; live end-of-match acceptance remains separate. |
Read the machine-readable release evidence ↗
What remains open
Exact-provider coverage, independent security review, narrower extension permissions and a general-public distribution review remain unfinished. A well-tested ledger can expose a weak strategy; it cannot turn that strategy into a profitable one. No measured independent prediction edge or guaranteed arbitrage is claimed.
Why this project matters
The useful engineering work is the discipline around the prediction: ownership, chronology, persistence, deduplication, numerical precision and transparent failure states. Those are the same habits needed for reliable automation, monitoring and decision-support products.