Demonstration data

Every figure on this site is deterministic demonstration data. It is not a live source, not a forecast, and not a recommendation. No wagering market is connected, and nothing here should be acted on.

Model limitations

What this system has not demonstrated, stated plainly. Read this before reading anything else on the site as a claim.

Limitations of the current state

These are true because the work has not been done yet. Each will change when its task lands, and none of them changes by itself.

Nothing has been measured

Task 000 builds the website and design foundation. The data foundation, formula engine, probability council, calibration and evaluation are Tasks 001 through 009 and none has begun.

No canonical archive exists, so nothing can be measured against anything.
No formula has had its contribution measured, so no formula is known to contribute anything.
No model produces an opinion, so there is no disagreement to report and no distribution to publish.
No settled outcome has been scored, so calibration is unmeasured — not poor, not good, unmeasured.
No holdout partition exists, so no unbiased evaluation has been possible.

Limitations that will remain

These are properties of the problem rather than of the current state. Building the remaining tasks will not remove them.

Racing is close to efficiently priced, and the market is a strong competitor.Any formula that appears to contribute must be measured after controlling for the market, because most apparent signal is the market’s signal arriving second-hand.
Sample sizes are small where they matter most. A specific runner at a specific course over a specific distance on specific going has run a handful of times, if ever. That is why a rate below 8 runs is withheld throughout this system rather than published with a caveat.
Evidence is unevenly available across jurisdictions. Sectional times do not exist in several regions; barrier trials are not published in others; going scales are not comparable between any two. A model that needs an evidence family cannot operate at full capability where that family is absent, and that is reported as reduced capability rather than absorbed silently.
Conditions change in ways nothing records. Rail position, track preparation, weather during a meeting and in-running interference all affect a result and are variously unpublished, published late, or published inconsistently.
A probability is not a prediction. A runner given 30% is expected to lose roughly seven times in ten. Neither its winning nor its losing tells you the estimate was good; only the rate across many such estimates does.
Calibration needs volume that accrues slowly. Meaningful calibration requires hundreds of settled races per segment. Until then, a calibration figure is a number with too little behind it — which is why nothing is summarised below 20 scorable entries.

Limitations of the interface itself

What the presentation cannot do, however good the underlying data becomes.

Evidence coverage is a statement about availability, not quality. Twelve of twelve families available says nothing about whether any of them holds enough observations to matter.
A demonstration-data label cannot survive a screenshot. The banner is on every page and is not dismissible for exactly this reason, but a cropped image can still remove it.
Automated accessibility testing catches perhaps a third of WCAG failures. Conformance is not established by the tests this project runs, and none is claimed. See accessibility.
Cross-browser behaviour is not verified. Only Chromium is available in the development environment; Firefox and WebKit have not been tested and no claim is made about them.

What this system will never do

Not currently absent — architecturally absent.

Handle deposits, withdrawals or payments of any kind.
Place a bet, or connect to anything that does.
Issue a staking instruction or calculate a personal return.
Offer a selection, a tip, a “best bet” or a “banker”.
Claim certainty, a guaranteed outcome or a sure thing.
Use casino-style engagement mechanics — artificial urgency, celebratory animation, flashing prices or gambling calls to action.

These are not features behind a flag. There is no payment code, no wagering integration and no staking module in this repository, and the prohibited vocabulary is asserted absent from the interface copy by tests/unit/display-contract.test.ts.

Where the method and the record are stated.

Methodology — how outputs would be produced and what constrains them.
Accuracy ledger — every published forecast against its outcome, including the misses.
Coverage — which regions are reachable and which are not.

Demonstration data

Every figure on this site is deterministic demonstration data. It is not a live source, not a forecast, and not a recommendation. No wagering market is connected, and nothing here should be acted on.

Model limitations

What this system has not demonstrated, stated plainly. Read this before reading anything else on the site as a claim.

Limitations of the current state

These are true because the work has not been done yet. Each will change when its task lands, and none of them changes by itself.

Nothing has been measured

Task 000 builds the website and design foundation. The data foundation, formula engine, probability council, calibration and evaluation are Tasks 001 through 009 and none has begun.

No canonical archive exists, so nothing can be measured against anything.
No formula has had its contribution measured, so no formula is known to contribute anything.
No model produces an opinion, so there is no disagreement to report and no distribution to publish.
No settled outcome has been scored, so calibration is unmeasured — not poor, not good, unmeasured.
No holdout partition exists, so no unbiased evaluation has been possible.

Limitations that will remain

These are properties of the problem rather than of the current state. Building the remaining tasks will not remove them.

Racing is close to efficiently priced, and the market is a strong competitor.Any formula that appears to contribute must be measured after controlling for the market, because most apparent signal is the market’s signal arriving second-hand.
Sample sizes are small where they matter most. A specific runner at a specific course over a specific distance on specific going has run a handful of times, if ever. That is why a rate below 8 runs is withheld throughout this system rather than published with a caveat.
Evidence is unevenly available across jurisdictions. Sectional times do not exist in several regions; barrier trials are not published in others; going scales are not comparable between any two. A model that needs an evidence family cannot operate at full capability where that family is absent, and that is reported as reduced capability rather than absorbed silently.
Conditions change in ways nothing records. Rail position, track preparation, weather during a meeting and in-running interference all affect a result and are variously unpublished, published late, or published inconsistently.
A probability is not a prediction. A runner given 30% is expected to lose roughly seven times in ten. Neither its winning nor its losing tells you the estimate was good; only the rate across many such estimates does.
Calibration needs volume that accrues slowly. Meaningful calibration requires hundreds of settled races per segment. Until then, a calibration figure is a number with too little behind it — which is why nothing is summarised below 20 scorable entries.

Limitations of the interface itself

What the presentation cannot do, however good the underlying data becomes.

Evidence coverage is a statement about availability, not quality. Twelve of twelve families available says nothing about whether any of them holds enough observations to matter.
A demonstration-data label cannot survive a screenshot. The banner is on every page and is not dismissible for exactly this reason, but a cropped image can still remove it.
Automated accessibility testing catches perhaps a third of WCAG failures. Conformance is not established by the tests this project runs, and none is claimed. See accessibility.
Cross-browser behaviour is not verified. Only Chromium is available in the development environment; Firefox and WebKit have not been tested and no claim is made about them.

What this system will never do

Not currently absent — architecturally absent.

Handle deposits, withdrawals or payments of any kind.
Place a bet, or connect to anything that does.
Issue a staking instruction or calculate a personal return.
Offer a selection, a tip, a “best bet” or a “banker”.
Claim certainty, a guaranteed outcome or a sure thing.
Use casino-style engagement mechanics — artificial urgency, celebratory animation, flashing prices or gambling calls to action.

Where the method and the record are stated.

Methodology — how outputs would be produced and what constrains them.
Accuracy ledger — every published forecast against its outcome, including the misses.
Coverage — which regions are reachable and which are not.

Racing

Evidence

About

Model limitations

Limitations of the current state

Nothing has been measured

Limitations that will remain

Limitations of the interface itself

What this system will never do

Related

Racing

Evidence

About

Model limitations

Limitations of the current state

Nothing has been measured

Limitations that will remain

Limitations of the interface itself

What this system will never do

Related