Replicate: one-off chase error, well beaten, in Irish chases

Reasoning

Why this hypothesis exists: what the lab saw, the question it asked, the mechanism it proposed, what it predicted, and the result that would prove it wrong. Written before the test was run.

1. Observed
   • H-db7f42b8 (Rival: GB no-drop chase lift is a GB-region effect, not class choice) passed discovery as a niche: A/E 1.24 on 347 bets (p 0.073), same-race control 0.97. O-19c9c833
   • H-c2928b4e (Generalise: one-off chase error, well beaten, not dropping in class (G) survived falsification: 5 of 5 attacks left it standing; it is frozen for its one vault query. O-2f35030a
2. AskedIs the class 'no-drop' lift really a GB-only effect, because race_class is missing for IRE before July 2026 and the class condition removes Irish runners? answered
3. MechanismA single jumping error in a chase costs ground and momentum. A horse beaten well because of that error has better ability than the result shows, whatever the jurisdiction.
4. Why the market misses itThe form figure shows a heavy defeat or a non-completion. The two clean runs before it are discounted, and Irish punters read the same misleading result.
5. Predictedpositive, A/E 1.10 to 1.30 on about 180 bets
6. Falsified ifA/E of 1.02 or lower on the IRE pool, which would make the chase edge a GB effect
7. Rule sketchregion = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)

Mechanism and rationale

A single jumping error in a chase costs ground and momentum. A horse beaten well because of that error has better ability than the result shows, whatever the jurisdiction.

The rule

region = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)

Reads: region, race_type, remarked_last3, jumping_errors_last3, last_jumping_error, last_position, last_beaten_lengths

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

781066242b896497685fbe76dda6d8730f2a0ef15c3889e31dc663a1843d2cad

Frozen

17 Oct, 01:28

Nothing in the definition can change after this.

Confounders

• Small IRE sample: the last-3 remark columns only start in November 2025, so the pool is about one season and may fall short of 150 bets
• IRE festival concentration (Leopardstown at Christmas and the Dublin Racing Festival, Punchestown) inside a short window
• Period effect: GB chase results were stronger before April 2026, and the IRE window covers the same period
• Field size in IRE chases: large handicap fields and high fall rates raise the base rate of errors and non-completions
• No class condition is possible (IRE race_class starts in July 2026), so class relief cannot be separated from the error effect
• Overlap with the GB one-off chase-error findings: a pass here is a replication in a separate population, but the mechanism and the column definitions are shared

Control

{
  "description": "Same-race other runners should score about 1.0 (GB pools showed 0.96-0.97). If they are well above 1.0, the lift belongs to the races (e.g. festival fields with a weak market), not the horses. Second control: Irish chasers well beaten last time with three remarked runs and no jumping error in any of them. If they match the selection's A/E, the edge is a general bounce after a heavy defeat and not caused by the error.",
  "rule": "region = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_jumping_error = 0 AND (last_position IS NULL OR last_beaten_lengths > 10)"
}

Discovery result

pass broad rule (niche) · 17 Oct, 01:26

Bets

189

20 winners

A/E

1.19

95%: 0.73 to 1.84

SP return

-31.4%

level stakes

Betfair SP return

-14.3%

189 bets with a BSP

Strike rate 10.6%. p = 0.25. Same-race control: the other runners in these races had A/E 0.98, so the selection beat its own races. Zone: discovery. Evaluator eval-1.0, data 6a4e98be18e83933

Falsification

Each attack is a deliberate attempt to kill the idea. 4 of 5 survived.

• perturbation: Beaten-lengths cut loosened to >8pass

  209 bets · A/E 1.13 (0.71 to 1.72) · control 0.98 · 17 Oct, 01:28

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27661.22

  after 2026-04-271431.10

• perturbation: Beaten-lengths cut tightened to >13pass

  173 bets · A/E 1.21 (0.72 to 1.91) · control 0.98 · 17 Oct, 01:28

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

Forward record

• forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}info

  0 bets · · 17 Oct, 01:28

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: > 10 -> 8 (-20%)pass

  209 bets · A/E 1.13 (0.71 to 1.72) · control 0.98 · 17 Oct, 01:26

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27661.22

  after 2026-04-271431.10

• neighbour: > 10 -> 12 (+20%)pass

  178 bets · A/E 1.27 (0.78 to 1.96) · control 0.97 · 17 Oct, 01:26

  Breakdown

Lineage

Agent runs

• designer v1ok · 41s · 17 Oct, 01:24

  Output (schema-checked)

  {
    "rule": "region = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)",
    "refinements": [
      {
        "label": "under 16/1 (chase error pools pay under 16/1)",
        "rule": "region = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND sp_decimal IS NOT NULL AND sp_decimal < 17"
      },
      {
        "label": "close earlier run (the error hid real form)",
        "rule": "region = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 5"
      }
    ],
    "control": {
      "description": "Same-race other runners should score about 1.0 (GB pools showed 0.96-0.97). If they are well above 1.0, the lift belongs to the races (e.g. festival fields with a weak market), not the horses. Second control: Irish chasers well beaten last time with three remarked runs and no jumping error in any of them. If they match the selection's A/E, the edge is a general bounce after a heavy defeat and not caused by the error.",
      "rule": "region = 'IRE' AND race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_jumping_error = 0 AND (last_position IS NULL OR last_beaten_lengths > 10)"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Small IRE sample: the last-3 remark columns only start in November 2025, so the pool is about one season and may fall short of 150 bets",
      "IRE festival concentration (Leopardstown at Christmas and the Dublin Racing Festival, Punchestown) inside a short window",
      "Period effect: GB chase results were stronger before April 2026, and the IRE window covers the same period",
      "Field size in IRE chases: large handicap fields and high fall rates raise the base rate of errors and non-completions",
      "No class condition is possible (IRE race_class starts in July 2026), so class relief cannot be separated from the error effect",
      "Overlap with the GB one-off chase-error findings: a pass here is a replication in a separate population, but the mechanism and the column definitions are shared"
    ],
    "mechanism": "A single jumping error in a chase costs ground and momentum, and often ends the horse's chance or its race. When the horse's other two recent runs carried no error, the heavy defeat or non-completion reflects the error and not its ability. Its true level is better than the last form figure shows, whatever the jurisdiction.",
    "market_miss": "The form figure shows a heavy defeat (beaten more than 10 lengths) or a non-completion. Punters and the prices weight that last run heavily and discount the two clean runs before it. Irish punters read the same misleading figure, and the in-running comment explaining the error is not priced in.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.30 on about 150-180 bets, with same-race runners near 1.0",
      "falsified_if": "A/E of 1.02 or lower on the IRE pool, which would make the chase edge a GB effect. Also falsified if the no-error well-beaten IRE control scores the same A/E, which would show the lift is not caused by the error."
    },
    "notes": "This is a replication in a separate population of the GB one-off chase-error pool, which scored 1.24 on 347. It is the population that pool's subsets never touched, so it is the first independent test of that thread. The rule sketch is kept unchanged: all its columns are pre-off and available in IRE from 2025-11-01. The sample may come in below 150 bets. If it points the right way it goes to falsification and forward tracking, and it should not be counted as confirmation. The refinements follow the observed price structure (under 16/1) and the close-earlier-run version, and are looked at only if the broad rule shows evidence."
  }

Timeline

1. Job 2255 finished

   17 Oct, 01:28 · engine:compute-1

2. forward result for H-64933537 (forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}): info

   17 Oct, 01:28 · engine:evaluator

3. compute-1 took job 2255 (forward_check, attempt 1)

   17 Oct, 01:28 · compute-1

4. Queued forward_check on the compute lane: daily forward check of a shadow system

   17 Oct, 01:28 · engine:scheduler

5. Job 2252 finished

   17 Oct, 01:28 · engine:compute-1

6. H-64933537 Frozen → SHADOW vault skipped: remarked_last3, jumping_errors_last3, last_jumping_error, last_beaten_lengths do not exist before the discovery zone; track it forward instead

   17 Oct, 01:28 · engine:orchestrator

7. compute-1 took job 2252 (vault, attempt 1)

   17 Oct, 01:28 · compute-1

8. Job 2251 finished

   17 Oct, 01:28 · engine:compute-1

9. Queued vault on the compute lane: frozen: one sealed vault query

   17 Oct, 01:28 · engine:scheduler

10. H-64933537 Falsification → FROZEN survived falsification

    17 Oct, 01:28 · engine:falsification

11. falsification result for H-64933537 (control: Control: repeat jumping errors, not a one-off): pass

    17 Oct, 01:28 · engine:evaluator

12. compute-1 took job 2251 (attack, attempt 1)

    17 Oct, 01:28 · compute-1

13. Job 2250 finished

    17 Oct, 01:28 · engine:compute-1

14. falsification result for H-64933537 (control: Control: the one error came earlier, not last time): fail

    17 Oct, 01:28 · engine:evaluator

15. compute-1 took job 2250 (attack, attempt 1)

    17 Oct, 01:28 · compute-1

16. Job 2249 finished

    17 Oct, 01:28 · engine:compute-1

17. falsification result for H-64933537 (partition: Completed runs only (beaten >10, excluding non-finishers)): pass

    17 Oct, 01:28 · engine:evaluator

18. compute-1 took job 2249 (attack, attempt 1)

    17 Oct, 01:28 · compute-1

19. Job 2248 finished

    17 Oct, 01:28 · engine:compute-1

20. falsification result for H-64933537 (perturbation: Beaten-lengths cut tightened to >13): pass

    17 Oct, 01:28 · engine:evaluator