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.
Accuracy ledger
The immutable public record of every published forecast against its outcome — including the ones that were wrong.
24 entries, 21 scorable. Scorable excludes races that were not run, outputs withheld before the race, and quarantined records. Those are retained in the ledger and listed below.5 correct, 12 outside the first three. Shown together in the page header so neither figure can be read without the other.
No model has produced a real forecast
Every entry below is deterministic demonstration data illustrating what the ledger records and how it must present it. No probability engine exists — it arrives with Task 005 — so nothing here measures the accuracy of anything. The page is built now so that the disclosure requirements are fixed before there is any result to be tempted by.
Summary
Every rate states the denominator it was computed over. Below the reporting threshold no rate is published at all.
Top selection won
5of 21 scorable
A count, not a rate. The rate is only meaningful against the probabilities that were published alongside these forecasts.
Finished outside the first three
12of 21 scorable
Shown at the same size and in the same place as the successes, because that is what an accuracy record is for.
Top-selection rate (%)
23.8
Published only at or above 20 scorable entries. 21 available.
Mean Brier score
0.1583
Lower is better. A proper scoring rule: it rewards a probability that matches the observed frequency, not one that is confidently right.
Mean log loss
0.4914
Lower is better. Punishes confident errors far more heavily than Brier does.
Not scored
3
1 race not run, 1 withheld before the race, 1 quarantined. All retained; none scored, because scoring them would fabricate a measurement.
Every entry (24)
In date order, newest first. Not sorted by outcome, not filtered, and not separable into a successes view — the ledger read has no outcome parameter.
Every ledger entry with its original probability, prediction timestamp, outcome, Brier score, log loss and evidence coverage
Race date
Race
Region
Top selection
Published probability
Outcome
Finished
Brier
Log loss
Publication
Evidence
2026-07-31
AUS demonstration race 1
AUS
Meridian Drift
34.0%
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.
1
0.4356
1.0788
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
11 of 12 evidence families. Unavailable: Track bias measurements.
2026-07-30
GB demonstration race 2
GB
Quiet Arithmetic
29.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
7
0.0841
0.3425
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
11 of 12 evidence families. Unavailable: Track bias measurements.
2026-07-29
IRE demonstration race 3
IRE
O'Callaghan's Reckoning
31.0%
Placed top three. The runner the model rated highest finished in the first three but did not win.
3
0.0961
0.3711
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
12 of 12 evidence families. No evidence families are missing for this record.
2026-07-28
FR demonstration race 4
FR
Ná Fionnuisce
22.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
9
0.0484
0.2485
Standard research. Published as standard research. A single evidence family supports the output.
7 of 12 evidence families. Unavailable: Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
2026-07-27
NZ demonstration race 5
NZ
Vale
41.0%
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.
1
0.3481
0.8916
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
12 of 12 evidence families. No evidence families are missing for this record.
2026-07-26
RSA demonstration race 6
RSA
Considerable Uncertainty
18.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
11
0.0324
0.1985
Standard research. Published as standard research. A single evidence family supports the output.
Placed top three. The runner the model rated highest finished in the first three but did not win.
2
0.0729
0.3147
Standard research. Published as standard research. A single evidence family supports the output.
9 of 12 evidence families. Unavailable: Trainer record, Weight and handicap, Track bias measurements.
2026-07-24
HK demonstration race 8
HK
Bâtiment Rouge
24.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
6
0.0576
0.2744
Standard research. Published as standard research. A single evidence family supports the output.
8 of 12 evidence families. Unavailable: Jockey record, Trainer record, Weight and handicap, Track bias measurements.
2026-07-23
JPN demonstration race 9
JPN
The Longest Possible Name In This Field
35.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
8
0.1225
0.4308
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
11 of 12 evidence families. Unavailable: Track bias measurements.
2026-07-22
UAE demonstration race 10
UAE
Sable Interval
26.0%
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.
1
0.5476
1.3471
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
10 of 12 evidence families. Unavailable: Weight and handicap, Track bias measurements.
2026-07-21
AUS demonstration race 11
AUS
Kōwhai Ascendant
19.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
12
0.0361
0.2107
Standard research. Published as standard research. A single evidence family supports the output.
Placed top three. The runner the model rated highest finished in the first three but did not win.
3
0.1089
0.4005
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
11 of 12 evidence families. Unavailable: Track bias measurements.
2026-07-19
IRE demonstration race 13
IRE
Marginal Utility
21.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
5
0.0441
0.2357
Standard research. Published as standard research. A single evidence family supports the output.
7 of 12 evidence families. Unavailable: Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
2026-07-18
FR demonstration race 14
FR
Sixpenny Ledger
38.0%
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.
1
0.3844
0.9676
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
12 of 12 evidence families. No evidence families are missing for this record.
2026-07-17
NZ demonstration race 15
NZ
Anemone Bright
23.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
10
0.0529
0.2614
Standard research. Published as standard research. A single evidence family supports the output.
6 of 12 evidence families. Unavailable: Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
2026-07-16
RSA demonstration race 16
RSA
Tōkyō Reckoner
30.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
4
0.0900
0.3567
Standard research. Published as standard research. A single evidence family supports the output.
9 of 12 evidence families. Unavailable: Trainer record, Weight and handicap, Track bias measurements.
2026-07-15
USA demonstration race 17
USA
Under Review
28.0%
Placed top three. The runner the model rated highest finished in the first three but did not win.
2
0.0784
0.3285
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
10 of 12 evidence families. Unavailable: Weight and handicap, Track bias measurements.
2026-07-14
HK demonstration race 18
HK
Cartographer's Error
17.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
14
0.0289
0.1863
Standard research. Published as standard research. A single evidence family supports the output.
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
6
0.1296
0.4463
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
11 of 12 evidence families. Unavailable: Track bias measurements.
2026-07-12
UAE demonstration race 20
UAE
Wintering Ground
32.0%
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.
1
0.4624
1.1394
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
11 of 12 evidence families. Unavailable: Track bias measurements.
2026-07-11
AUS demonstration race 21
AUS
Almost Certainly Not
25.0%
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.
8
0.0625
0.2877
Standard research. Published as standard research. A single evidence family supports the output.
8 of 12 evidence families. Unavailable: Jockey record, Trainer record, Weight and handicap, Track bias measurements.
2026-07-10
GB demonstration race 22
GB
Ferrous Dawn
29.0%
Race not run. The race was abandoned, postponed or cancelled. The forecast was neither right nor wrong, and is scored as neither — assigning it an outcome would fabricate a measurement.
Not run
Not scorable
Not scorable
High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.
10 of 12 evidence families. Unavailable: Weight and handicap, Track bias measurements.
2026-07-09
IRE demonstration race 23
IRE
Rúnda Solas
44.0%
Withheld. An output was produced but deliberately not published, because it did not meet the publication threshold. It is retained here so the withholding is visible rather than looking like a race that was never assessed.
Not run
Not scorable
Not scorable
Withheld. An output exists but was deliberately not published, because it did not meet the publication threshold. The reason is recorded.
Quarantined. The underlying records failed validation, so nothing derived from them was published. Retained for the same reason.
Not run
Not scorable
Not scorable
Quarantined. The underlying records failed validation. Nothing derived from them is published.
6 of 12 evidence families. Unavailable: Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
AUS demonstration race 1
Race date
2026-07-31
Region
AUS
Race type
Handicap
Top selection
Meridian Drift
Published probability
34.0%
Forecast made
2026-07-31T02:10:00.000Z
Outcome
Won
Finished
1
Brier score
0.4356
Log loss
1.0788
Model version
0.0.0-none
Formula version
0.0.0-none
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.11 of 12 evidence families. Unavailable: Track bias measurements.
GB demonstration race 2
Race date
2026-07-30
Region
GB
Race type
Set weights
Top selection
Quiet Arithmetic
Published probability
29.0%
Forecast made
2026-07-30T02:22:00.000Z
Outcome
Outside top three
Finished
7
Brier score
0.0841
Log loss
0.3425
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.11 of 12 evidence families. Unavailable: Track bias measurements.
IRE demonstration race 3
Race date
2026-07-29
Region
IRE
Race type
Maiden
Top selection
O'Callaghan's Reckoning
Published probability
31.0%
Forecast made
2026-07-29T03:01:00.000Z
Outcome
Placed top three
Finished
3
Brier score
0.0961
Log loss
0.3711
Model version
0.0.0-none
Formula version
0.0.0-none
Placed top three. The runner the model rated highest finished in the first three but did not win.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.12 of 12 evidence families. No evidence families are missing for this record.
FR demonstration race 4
Race date
2026-07-28
Region
FR
Race type
Group 3
Top selection
Ná Fionnuisce
Published probability
22.0%
Forecast made
2026-07-28T02:14:00.000Z
Outcome
Outside top three
Finished
9
Brier score
0.0484
Log loss
0.2485
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.7 of 12 evidence families. Unavailable: Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
NZ demonstration race 5
Race date
2026-07-27
Region
NZ
Race type
Listed
Top selection
Vale
Published probability
41.0%
Forecast made
2026-07-27T01:49:00.000Z
Outcome
Won
Finished
1
Brier score
0.3481
Log loss
0.8916
Model version
0.0.0-none
Formula version
0.0.0-none
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.12 of 12 evidence families. No evidence families are missing for this record.
RSA demonstration race 6
Race date
2026-07-26
Region
RSA
Race type
Handicap
Top selection
Considerable Uncertainty
Published probability
18.0%
Forecast made
2026-07-26T02:34:00.000Z
Outcome
Outside top three
Finished
11
Brier score
0.0324
Log loss
0.1985
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.5 of 12 evidence families. Unavailable: Course configuration, Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
USA demonstration race 7
Race date
2026-07-25
Region
USA
Race type
Set weights
Top selection
Half-Light Amendment
Published probability
27.0%
Forecast made
2026-07-25T03:38:00.000Z
Outcome
Placed top three
Finished
2
Brier score
0.0729
Log loss
0.3147
Model version
0.0.0-none
Formula version
0.0.0-none
Placed top three. The runner the model rated highest finished in the first three but did not win.Standard research. Published as standard research. A single evidence family supports the output.9 of 12 evidence families. Unavailable: Trainer record, Weight and handicap, Track bias measurements.
HK demonstration race 8
Race date
2026-07-24
Region
HK
Race type
Maiden
Top selection
Bâtiment Rouge
Published probability
24.0%
Forecast made
2026-07-24T01:32:00.000Z
Outcome
Outside top three
Finished
6
Brier score
0.0576
Log loss
0.2744
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.8 of 12 evidence families. Unavailable: Jockey record, Trainer record, Weight and handicap, Track bias measurements.
JPN demonstration race 9
Race date
2026-07-23
Region
JPN
Race type
Group 3
Top selection
The Longest Possible Name In This Field
Published probability
35.0%
Forecast made
2026-07-23T01:57:00.000Z
Outcome
Outside top three
Finished
8
Brier score
0.1225
Log loss
0.4308
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.11 of 12 evidence families. Unavailable: Track bias measurements.
UAE demonstration race 10
Race date
2026-07-22
Region
UAE
Race type
Listed
Top selection
Sable Interval
Published probability
26.0%
Forecast made
2026-07-22T02:44:00.000Z
Outcome
Won
Finished
1
Brier score
0.5476
Log loss
1.3471
Model version
0.0.0-none
Formula version
0.0.0-none
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.10 of 12 evidence families. Unavailable: Weight and handicap, Track bias measurements.
AUS demonstration race 11
Race date
2026-07-21
Region
AUS
Race type
Handicap
Top selection
Kōwhai Ascendant
Published probability
19.0%
Forecast made
2026-07-21T02:02:00.000Z
Outcome
Outside top three
Finished
12
Brier score
0.0361
Log loss
0.2107
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.4 of 12 evidence families. Unavailable: Market observations, Course configuration, Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
GB demonstration race 12
Race date
2026-07-20
Region
GB
Race type
Set weights
Top selection
Provisional Finding
Published probability
33.0%
Forecast made
2026-07-20T02:12:00.000Z
Outcome
Placed top three
Finished
3
Brier score
0.1089
Log loss
0.4005
Model version
0.0.0-none
Formula version
0.0.0-none
Placed top three. The runner the model rated highest finished in the first three but did not win.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.11 of 12 evidence families. Unavailable: Track bias measurements.
IRE demonstration race 13
Race date
2026-07-19
Region
IRE
Race type
Maiden
Top selection
Marginal Utility
Published probability
21.0%
Forecast made
2026-07-19T02:28:00.000Z
Outcome
Outside top three
Finished
5
Brier score
0.0441
Log loss
0.2357
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.7 of 12 evidence families. Unavailable: Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
FR demonstration race 14
Race date
2026-07-18
Region
FR
Race type
Group 3
Top selection
Sixpenny Ledger
Published probability
38.0%
Forecast made
2026-07-18T01:38:00.000Z
Outcome
Won
Finished
1
Brier score
0.3844
Log loss
0.9676
Model version
0.0.0-none
Formula version
0.0.0-none
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.12 of 12 evidence families. No evidence families are missing for this record.
NZ demonstration race 15
Race date
2026-07-17
Region
NZ
Race type
Listed
Top selection
Anemone Bright
Published probability
23.0%
Forecast made
2026-07-17T02:03:00.000Z
Outcome
Outside top three
Finished
10
Brier score
0.0529
Log loss
0.2614
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.6 of 12 evidence families. Unavailable: Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
RSA demonstration race 16
Race date
2026-07-16
Region
RSA
Race type
Handicap
Top selection
Tōkyō Reckoner
Published probability
30.0%
Forecast made
2026-07-16T03:38:00.000Z
Outcome
Outside top three
Finished
4
Brier score
0.0900
Log loss
0.3567
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.9 of 12 evidence families. Unavailable: Trainer record, Weight and handicap, Track bias measurements.
USA demonstration race 17
Race date
2026-07-15
Region
USA
Race type
Set weights
Top selection
Under Review
Published probability
28.0%
Forecast made
2026-07-15T01:36:00.000Z
Outcome
Placed top three
Finished
2
Brier score
0.0784
Log loss
0.3285
Model version
0.0.0-none
Formula version
0.0.0-none
Placed top three. The runner the model rated highest finished in the first three but did not win.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.10 of 12 evidence families. Unavailable: Weight and handicap, Track bias measurements.
HK demonstration race 18
Race date
2026-07-14
Region
HK
Race type
Maiden
Top selection
Cartographer's Error
Published probability
17.0%
Forecast made
2026-07-14T03:11:00.000Z
Outcome
Outside top three
Finished
14
Brier score
0.0289
Log loss
0.1863
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.3 of 12 evidence families. Unavailable: Stewards reports, Market observations, Course configuration, Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
JPN demonstration race 19
Race date
2026-07-13
Region
JPN
Race type
Group 3
Top selection
Silent Denominator
Published probability
36.0%
Forecast made
2026-07-13T02:24:00.000Z
Outcome
Outside top three
Finished
6
Brier score
0.1296
Log loss
0.4463
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.11 of 12 evidence families. Unavailable: Track bias measurements.
UAE demonstration race 20
Race date
2026-07-12
Region
UAE
Race type
Listed
Top selection
Wintering Ground
Published probability
32.0%
Forecast made
2026-07-12T02:42:00.000Z
Outcome
Won
Finished
1
Brier score
0.4624
Log loss
1.1394
Model version
0.0.0-none
Formula version
0.0.0-none
Won. The runner the model rated highest won. A single such result says almost nothing on its own; it is the rate across many races, against the stated probability, that carries information.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.11 of 12 evidence families. Unavailable: Track bias measurements.
AUS demonstration race 21
Race date
2026-07-11
Region
AUS
Race type
Handicap
Top selection
Almost Certainly Not
Published probability
25.0%
Forecast made
2026-07-11T02:31:00.000Z
Outcome
Outside top three
Finished
8
Brier score
0.0625
Log loss
0.2877
Model version
0.0.0-none
Formula version
0.0.0-none
Outside top three. The runner the model rated highest finished outside the first three. Recorded and displayed identically to a correct forecast.Standard research. Published as standard research. A single evidence family supports the output.8 of 12 evidence families. Unavailable: Jockey record, Trainer record, Weight and handicap, Track bias measurements.
GB demonstration race 22
Race date
2026-07-10
Region
GB
Race type
Set weights
Top selection
Ferrous Dawn
Published probability
29.0%
Forecast made
2026-07-10T02:46:00.000Z
Outcome
Race not run
Finished
The race was abandoned after a track inspection. The forecast was neither right nor wrong and is excluded from every scoring rule — assigning it an outcome would fabricate a measurement.
Brier score
Not scorable — scoring this entry would fabricate a measurement.
Log loss
Not scorable.
Model version
0.0.0-none
Formula version
0.0.0-none
Race not run. The race was abandoned, postponed or cancelled. The forecast was neither right nor wrong, and is scored as neither — assigning it an outcome would fabricate a measurement.High evidence. Multiple independent evidence families agree. This indicates agreement between methods — it does not indicate certainty, it is not a guaranteed outcome, and it is not a prediction of the result.10 of 12 evidence families. Unavailable: Weight and handicap, Track bias measurements.
The race was abandoned after a track inspection. The forecast was neither right nor wrong and is excluded from every scoring rule — assigning it an outcome would fabricate a measurement.
IRE demonstration race 23
Race date
2026-07-09
Region
IRE
Race type
Maiden
Top selection
Rúnda Solas
Published probability
44.0%
Forecast made
2026-07-09T01:38:00.000Z
Outcome
Withheld
Finished
An output was produced but did not meet the publication threshold: only four evidence families were available and no independent family corroborated the estimate. It was withheld before the race and is retained here so the withholding is visible rather than looking like a race nobody assessed.
Brier score
Not scorable — scoring this entry would fabricate a measurement.
Log loss
Not scorable.
Model version
0.0.0-none
Formula version
0.0.0-none
Withheld. An output was produced but deliberately not published, because it did not meet the publication threshold. It is retained here so the withholding is visible rather than looking like a race that was never assessed.Withheld. An output exists but was deliberately not published, because it did not meet the publication threshold. The reason is recorded.4 of 12 evidence families. Unavailable: Market observations, Course configuration, Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
An output was produced but did not meet the publication threshold: only four evidence families were available and no independent family corroborated the estimate. It was withheld before the race and is retained here so the withholding is visible rather than looking like a race nobody assessed.
FR demonstration race 24
Race date
2026-07-08
Region
FR
Race type
Group 3
Top selection
Pale Constant
Published probability
31.0%
Forecast made
2026-07-08T02:04:00.000Z
Outcome
Quarantined
Finished
The declarations for this race failed validation — two runners shared a saddle-cloth number in the source payload. Nothing derived from the record was published.
Brier score
Not scorable — scoring this entry would fabricate a measurement.
Log loss
Not scorable.
Model version
0.0.0-none
Formula version
0.0.0-none
Quarantined. The underlying records failed validation, so nothing derived from them was published. Retained for the same reason.Quarantined. The underlying records failed validation. Nothing derived from them is published.6 of 12 evidence families. Unavailable: Going history, Pedigree, Jockey record, Trainer record, Weight and handicap, Track bias measurements.
The declarations for this race failed validation — two runners shared a saddle-cloth number in the source payload. Nothing derived from the record was published.
Weak and strong periods
Both, with the weakest first. A page that surfaced only its strong periods would be the flattering-subset failure this ledger exists to prevent.
Weakest recorded period
Entries
98 scorable
Top-selection rate (%)
Below sample threshold
8 of 20 required to publish a rate.
Mean Brier
Below sample threshold
Coverage averaged five of twelve evidence families across this period. The rate is shown with its sample and is BELOW the reporting threshold, so it should not be read as a measurement of capability — it is shown because hiding the weakest period is exactly what an accuracy ledger must not do.
Strongest recorded period
Entries
99 scorable
Top-selection rate (%)
Below sample threshold
9 of 20 required to publish a rate.
Mean Brier
Below sample threshold
Coverage averaged eleven of twelve evidence families. Also below the reporting threshold, and shown with the same prominence and the same caveat as the weakest period. Neither figure is evidence of skill on this sample size.
Corrections and withheld outputs
Corrections are appended, never applied by rewriting an entry. A withheld or quarantined output stays in the ledger with the reason.
Appended corrections (1)
JPN demonstration race 19 — 2026-07-13
2026-07-13T07:30:00.000ZThe original result placed this runner second. A stewards inquiry amended the placings after publication, moving it to sixth. The original entry is retained above; this correction is appended rather than applied to it.
Withheld and quarantined (2)
IRE demonstration race 23 — Withheld An output was produced but did not meet the publication threshold: only four evidence families were available and no independent family corroborated the estimate. It was withheld before the race and is retained here so the withholding is visible rather than looking like a race nobody assessed.
FR demonstration race 24 — Quarantined The declarations for this race failed validation — two runners shared a saddle-cloth number in the source payload. Nothing derived from the record was published.
Calibration
Whether the stated probabilities match the observed frequencies. This is the measurement that matters, and it needs far more data than the ledger holds.
Calibration and model evaluation
Design ready. The interface, contracts and extension points exist. No implementation is connected. Requires settled outcomes to measure against. None exist.
Not available yet
Calibration and Learning requires the research role. This capability arrives with Task 009.
Owning task: 009 — Calibration and Learning.
Module version 0.1.0, feature state "design-ready".
The interface, contracts and extension point exist. Nothing is computed, so nothing is shown — an empty version of this area would be indistinguishable from one whose values happened to be missing.
How to read this ledger
Stated plainly, because an accuracy page is the easiest page in a research product to misread.
A correct forecast is not evidence of skill, and an incorrect one is not evidence of its absence. A probability of 30% is supposed to be wrong roughly seven times in ten.
The only figure here that speaks to quality is calibration — whether the things given a 30% probability happen about 30% of the time. That requires hundreds of settled races, and none exist.
Nothing on this page is a selection, a recommendation, or a statement about any future race. See model limitations for what this system has and has not demonstrated, and methodology for how outputs are produced.
No accuracy claim is made
This ledger currently demonstrates a disclosure format. It measures no model, because no model exists. Any figure read from it as a performance claim would be a misreading of demonstration data.
Provenance
Where this ledger came from and how old it is.
Provenance
When this answer was assembled, and how old the underlying data was when it was. These are separate facts, and showing only the first would let a cached response read as current.
Source
Deterministic demonstration data. Not a live source.
Answer assembled
2026-08-01T04:30:00.000Z
Underlying data last updated
2026-08-01T04:24:00.000Z — 6 minutes old
Data state
Quarantined. Held back from the canonical archive by a validation rule. Excluded from every output.
Regions excluded from this answer, each with the recorded reason. An excluded region is a fact about coverage, not an empty space:
Singapore — The upstream source returned repeated 504 responses. Three retries were exhausted; the region is recorded as failed with its reason rather than silently omitted. Nothing is shown for it, because nothing was retrieved.
Korea — No fixtures scheduled for this date. The region was attempted and returned an empty set — this is an absence of racing, not a failure to look, and the two are recorded differently.
India — No entitled source family serves this region. The endpoints are registered and are never called, so no attempt was made — which is different again from an attempt that returned nothing.