Rival: the IRE wide-trip lift is regional, not geometric; test IRE sprint last runs

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-cf793824 (Test-mechanism: the 7f+ wide-trip lift is larger after a big-field las) passed discovery: A/E 1.35 on 227 bets (p 0.021), same-race control 0.92. O-107fcec5
   • H-2c6644e4 (Rival: the 7f+ wide-trip lift is a turf effect; all-weather last runs ) pointed the predicted way but failed the bar: A/E 1.19 on 294 bets (p 0.092), control 0.95. O-20d5482d
2. AskedIs the IRE wide-trip lift a property of IRE racing (comment-writing or the IRE market) rather than ground lost on bends? If so, IRE sprint last runs should score as high as IRE 7f+ ones. testing
3. MechanismRival cause: IRE comment-writers may only call a run 'wide' when it clearly cost the horse. Or the IRE market, which is thinner and centred on a few big yards, may underrate exposed handicappers' close defeats whatever the trip. Either would make the IRE lift a regional effect, not a ground-loss effect.
4. Why the market misses itIf the effect is regional, the IRE form comment carries information the IRE market does not use. Sprint wide trips would then be missed as much as round-course ones, because the narrow defeat looks ordinary in the result.
5. Predictedpositive, A/E 1.25 or more on about 100 to 150 IRE sprint bets if the rival holds
6. Falsified ifA/E at or below 1.05 on 80+ bets. IRE sprint wide trips at par would mean the IRE lift needs bends, and the ground-loss mechanism stands for IRE too.

Mechanism and rationale

Rival cause: IRE comment-writers may only call a run 'wide' when it clearly cost the horse. Or the IRE market, which is thinner and centred on a few big yards, may underrate exposed handicappers' close defeats whatever the trip. Either would make the IRE lift a regional effect, not a ground-loss effect.

The rule

race_type = 'Flat' AND region = 'IRE' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND sp_decimal <= 15 AND last_distance_yards < 1540

Reads: race_type, region, last_raced_wide, last_position, last_beaten_lengths, sp_decimal, last_distance_yards

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

57d3fd8de20c69182b729bf2132f7c4628d633e1c267e5396dab65e155f9fb83

Frozen

17 Oct, 05:00

Nothing in the definition can change after this.

Confounders

• Sample is small: IRE 7f+ cuts had only 35 to 98 bets, and the sprint cut may fall short of the 150-bet floor. A thin result that points the right way goes to falsification and forward tracking, not straight to a pass.
• IRE sprint fields (e.g. the Curragh, Dundalk) are often big, so field size and region stay mixed. A wide trip in a big field may reflect traffic rather than ground lost on bends.
• IRE race_class only starts in 2026-07, so class cannot be controlled for most of the sample.
• The finding that sprint last runs score at par (O-6497b473) was GB-dominated, so it is not a clean prior for IRE.
• Dundalk all-weather sprints have a bend even at 5f to 6f. A sprint wide trip there is not geometry-free, which blurs the regional-versus-geometric contrast.
• This rule overlaps the frozen ≤14/1 wide-trip rules only on IRE sub-7f runners. It should not be read as an independent confirmation of the round-course lift, nor that lift as support for this rule.

Control

{
  "description": "The other runners in the same races should be at about par (A/E 0.92 to 1.0), so a lift is not just a property of IRE sprint races. The key comparison is IRE sprint narrow defeats with no wide trip. This control separates the two arms of the rival. If those runners also beat their price (A/E ≥ 1.15), the IRE market underrates close defeats in general (the market arm), and 'wide' adds nothing. If they are at par while the wide group is lifted, the wide comment itself carries IRE-specific information (the comment-writer arm). For context: GB sprint wide trips were near par (O-6497b473).",
  "rule": "race_type = 'Flat' AND region = 'IRE' AND last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND sp_decimal <= 15 AND last_distance_yards < 1540"
}

Discovery result

pass broad rule (niche) · 17 Oct, 04:59

Bets

34

7 winners

A/E

1.19

95%: 0.48 to 2.46

SP return

-33.7%

level stakes

Betfair SP return

-21.2%

34 bets with a BSP

Strike rate 20.6%. p = 0.37. Same-race control: the other runners in these races had A/E 0.96, 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. 2 of 6 survived.

• perturbation: Beaten-lengths cut tightened to 4info

  31 bets · A/E 1.23 (0.49 to 2.53) · control 0.95 · 17 Oct, 05:00

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-278–

  after 2026-04-2723–

• perturbation: Beaten-lengths cut loosened to 6 and price cap to 18pass

  50 bets · A/E 1.21 (0.55 to 2.29) · control 0.95 · 17 Oct, 05:00

  Breakdown

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, 05:01

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: > 1 -> 0 (-20%)pass

  41 bets · A/E 1.17 (0.51 to 2.31) · control 0.96 · 17 Oct, 05:01

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-2711–

  after 2026-04-2730–

• neighbour: > 1 -> 2 (+20%)fail

  23 bets · A/E 0.76 (0.09 to 2.74) · control 1.04 · 17 Oct, 05:01

  Breakdown

  bets

Lineage

Agent runs

• designer v1ok · 31s · 17 Oct, 04:50

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND region = 'IRE' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND sp_decimal <= 15 AND last_distance_yards < 1540",
    "refinements": [
      {
        "label": "IRE handicaps only (exposed handicappers, per the market-arm mechanism)",
        "rule": "race_type = 'Flat' AND region = 'IRE' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND sp_decimal <= 15 AND last_distance_yards < 1540 AND is_handicap = 1"
      },
      {
        "label": "Straight sprints only (6f and under, few or no bends)",
        "rule": "race_type = 'Flat' AND region = 'IRE' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND sp_decimal <= 15 AND last_distance_yards <= 1320"
      }
    ],
    "control": {
      "description": "The other runners in the same races should be at about par (A/E 0.92 to 1.0), so a lift is not just a property of IRE sprint races. The key comparison is IRE sprint narrow defeats with no wide trip. This control separates the two arms of the rival. If those runners also beat their price (A/E ≥ 1.15), the IRE market underrates close defeats in general (the market arm), and 'wide' adds nothing. If they are at par while the wide group is lifted, the wide comment itself carries IRE-specific information (the comment-writer arm). For context: GB sprint wide trips were near par (O-6497b473).",
      "rule": "race_type = 'Flat' AND region = 'IRE' AND last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND sp_decimal <= 15 AND last_distance_yards < 1540"
    },
    "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": [
      "Sample is small: IRE 7f+ cuts had only 35 to 98 bets, and the sprint cut may fall short of the 150-bet floor. A thin result that points the right way goes to falsification and forward tracking, not straight to a pass.",
      "IRE sprint fields (e.g. the Curragh, Dundalk) are often big, so field size and region stay mixed. A wide trip in a big field may reflect traffic rather than ground lost on bends.",
      "IRE race_class only starts in 2026-07, so class cannot be controlled for most of the sample.",
      "The finding that sprint last runs score at par (O-6497b473) was GB-dominated, so it is not a clean prior for IRE.",
      "Dundalk all-weather sprints have a bend even at 5f to 6f. A sprint wide trip there is not geometry-free, which blurs the regional-versus-geometric contrast.",
      "This rule overlaps the frozen ≤14/1 wide-trip rules only on IRE sub-7f runners. It should not be read as an independent confirmation of the round-course lift, nor that lift as support for this rule."
    ],
    "mechanism": "Rival to the ground-loss explanation, with two arms. (a) IRE form comment-writers use 'raced wide' only when it clearly cost the horse, so the flag marks real hard-luck runs whatever the trip. (b) The thinner IRE market, centred on a few big yards, underrates exposed horses' close defeats in general. Either arm makes the IRE lift a regional effect that should appear over sprint trips, where there are few or no bends to lose ground on. The no-wide control separates the two arms.",
    "market_miss": "The IRE market prices off the result line, and a defeat by 5 lengths or less looks ordinary there. The selective IRE 'wide' comment, or the general closeness of the defeat, carries information the thin IRE market does not price in, and this would hold over sprints as much as over round courses.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.25 or more on about 100 to 150 IRE sprint bets if the regional rival holds",
      "falsified_if": "A/E at or below 1.05 on 80 or more bets. That would mean IRE sprint wide trips are at par, so the IRE lift needs bends and the ground-loss mechanism stands for IRE too."
    },
    "notes": "The rule is kept exactly as sketched. Every column is pre-off: last_raced_wide, last_position, last_beaten_lengths and last_distance_yards describe the previous run. last_distance_yards < 1540 means last run under 7f (1 furlong = 220 yards), the complement of the 7f+ cuts. sp_decimal <= 15 matches the ≤14/1 convention of the frozen wide-trip rules. last_position > 1 drops winners and non-finishers (NULL). The control rule is the decisive reading. A lifted wide group with a par no-wide control supports the comment-writer arm. Both groups lifted supports the market arm. A wide group at par falsifies the rival."
  }

Timeline

1. Job 3130 finished

   17 Oct, 05:01 · engine:compute-1

2. hypothesis note H-19f53be9

   17 Oct, 05:01 · engine:perturb

3. perturbation result for H-19f53be9 (neighbour: <= 15 -> 18 (+20%)): pass

   17 Oct, 05:01 · engine:perturb

4. perturbation result for H-19f53be9 (neighbour: <= 15 -> 12 (-20%)): pass

   17 Oct, 05:01 · engine:perturb

5. perturbation result for H-19f53be9 (neighbour: <= 5 -> 6 (+20%)): pass

   17 Oct, 05:01 · engine:perturb

6. perturbation result for H-19f53be9 (neighbour: <= 5 -> 4 (-20%)): pass

   17 Oct, 05:01 · engine:perturb

7. perturbation result for H-19f53be9 (neighbour: > 1 -> 2 (+20%)): fail

   17 Oct, 05:01 · engine:perturb

8. perturbation result for H-19f53be9 (neighbour: > 1 -> 0 (-20%)): pass

   17 Oct, 05:01 · engine:perturb

9. compute-1 took job 3130 (mutate, attempt 1)

   17 Oct, 05:01 · compute-1

10. Queued mutate on the compute lane: parameter stability: do neighbouring thresholds hold?

    17 Oct, 05:01 · engine:scheduler

11. Job 3124 finished

    17 Oct, 05:01 · engine:compute-1

12. forward result for H-19f53be9 (forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}): info

    17 Oct, 05:01 · engine:evaluator

13. compute-1 took job 3124 (forward_check, attempt 1)

    17 Oct, 05:01 · compute-1

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

    17 Oct, 05:01 · engine:scheduler

15. Job 3100 finished

    17 Oct, 05:00 · engine:compute-1

16. H-19f53be9 Frozen → SHADOW vault skipped: last_raced_wide, last_beaten_lengths do not exist before the discovery zone; track it forward instead

    17 Oct, 05:00 · engine:orchestrator