Wide trip at a galloping track last time should pay less if lost ground on bends is the cause

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-17cdbcfa (Wide trip at a sharp track last time costs more ground, so it pays mor) passed discovery as a niche: A/E 1.18 on 612 bets (p 0.054), same-race control 0.97. O-0c2f7d11
2. AskedWithin the full parent rule (tilt exclusion and figures kept), does the wide-trip A/E differ between sharp and galloping last-time tracks, and between low and high draws today? answered
3. MechanismA horse covers extra ground mostly on bends. On a galloping track with long, sweeping turns, a wide trip should cost less, so a close finish there undersells the horse less.
4. Why the market misses itThe bare beaten distance is read the same whatever the track shape. If the edge is the same at galloping and sharp tracks, it is not about bends at all; it could be a remark-tag effect.
5. Predictedpositive, A/E 1.00 to 1.08 on 300+ bets, clearly below the 1.18 at sharp tracks
6. Falsified ifA/E 1.15 or higher, the same as sharp tracks. Track shape would then not matter and the ground-on-bends story would be wrong.
7. Rule sketchlast_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness = 0

Mechanism and rationale

A horse covers extra ground mostly on bends. On a galloping track with long, sweeping turns, a wide trip should cost less, so a close finish there undersells the horse less.

The rule

last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness = 0

Reads: last_raced_wide, last_position, last_beaten_lengths, last_slow_start, last_track_sharpness

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

163127531d76bf648581769b8dac61a4430f6af9c3b779e31252883af2045908

Frozen

16 Oct, 12:12

Nothing in the definition can change after this.

Confounders

• Galloping tracks host better-class races with deeper, more efficient markets, which could lower A/E for reasons that have nothing to do with bends
• Field size: galloping tracks often have bigger fields, so the yardstick must be corrected for field size
• Irish share of galloping tracks differs, and Irish markets and race-comment tagging may differ (GB-only refinement)
• The base rule lacks the parent's tilt exclusion (last_run_style midfield/held_up with last_race_tilt >= 0.2). It is kept off on purpose so the rule matches the sharp-track sibling, which means the result cannot be compared with the parent's 1.27
• last_raced_wide exists only from Nov 2025, giving about one year in two halves, so the sample is thin
• Longshot lift (16/1+) common to horses beaten last time could inflate A/E without any track-shape effect
• The code for last_track_sharpness is assumed to be 0 = galloping and >= 2 = sharp, following the sibling rule

Control

{
  "description": "The main comparison is the sibling rule with the same base at sharp last-time tracks (H-17cdbcfa, A/E 1.18 on 612). It uses the same base, also without the tilt exclusion, so track shape is the only thing that changes. The mechanism needs galloping-track A/E to be clearly below the sharp-track figure, with the gap the same in both halves of the Nov 2025+ window. The other runners in the same races should show about 1.00. If they are also high, the lift belongs to the races (class or field size at galloping tracks), not to the wide-trip runners.",
  "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness >= 2"
}

Discovery result

pass broad rule (niche) · 16 Oct, 12:11

Bets

270

57 winners

A/E

1.23

95%: 0.93 to 1.60

SP return

-10.2%

level stakes

Betfair SP return

+7.3%

270 bets with a BSP

Strike rate 21.1%. p = 0.069. Same-race control: the other runners in these races had A/E 0.95, 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.

• price_band: Drop the up-to-3/1 band that carries the liftfail

  190 bets · A/E 0.99 (0.58 to 1.56) · control 1.00 · 16 Oct, 12:12

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27580.84

  after 2026-04-271321.06

• control: Control: same base at galloping tracks but NOT raced widepass

  7,100 bets · A/E 1.01 (0.95 to 1.07) · control 1.00 · 16 Oct, 12:12

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 · · 16 Oct, 12:12

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: > 1 -> 0 (-20%)pass

  386 bets · A/E 1.11 (0.88 to 1.40) · control 0.97 · 16 Oct, 12:12

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271410.96

  after 2026-04-272451.20

• neighbour: > 1 -> 2 (+20%)pass

  180 bets · A/E 1.14 (0.78 to 1.61) · control 0.97 · 16 Oct, 12:12

  Breakdown

Lineage

Agent runs

• designer v1ok · 36s · 16 Oct, 12:06

  Output (schema-checked)

  {
    "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness = 0",
    "refinements": [
      {
        "label": "GB only (removes the Irish share of galloping tracks)",
        "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness = 0 AND region = 'GB'"
      },
      {
        "label": "14/1 or shorter (the wide-trip family's longshot lift is a known artefact of horses beaten last time)",
        "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness = 0 AND sp_decimal <= 15"
      }
    ],
    "control": {
      "description": "The main comparison is the sibling rule with the same base at sharp last-time tracks (H-17cdbcfa, A/E 1.18 on 612). It uses the same base, also without the tilt exclusion, so track shape is the only thing that changes. The mechanism needs galloping-track A/E to be clearly below the sharp-track figure, with the gap the same in both halves of the Nov 2025+ window. The other runners in the same races should show about 1.00. If they are also high, the lift belongs to the races (class or field size at galloping tracks), not to the wide-trip runners.",
      "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND last_track_sharpness >= 2"
    },
    "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": [
      "Galloping tracks host better-class races with deeper, more efficient markets, which could lower A/E for reasons that have nothing to do with bends",
      "Field size: galloping tracks often have bigger fields, so the yardstick must be corrected for field size",
      "Irish share of galloping tracks differs, and Irish markets and race-comment tagging may differ (GB-only refinement)",
      "The base rule lacks the parent's tilt exclusion (last_run_style midfield/held_up with last_race_tilt >= 0.2). It is kept off on purpose so the rule matches the sharp-track sibling, which means the result cannot be compared with the parent's 1.27",
      "last_raced_wide exists only from Nov 2025, giving about one year in two halves, so the sample is thin",
      "Longshot lift (16/1+) common to horses beaten last time could inflate A/E without any track-shape effect",
      "The code for last_track_sharpness is assumed to be 0 = galloping and >= 2 = sharp, following the sibling rule"
    ],
    "mechanism": "Extra ground on a wide trip is lost mostly on bends. Long, sweeping turns at galloping tracks cost a wide runner less ground than the tight turns at sharp tracks. So a close finish (beaten 5 lengths or less, not first, no slow start) after racing wide at a galloping track undersells the horse less than the same run at a sharp track, and today's overlay should be smaller.",
    "market_miss": "Punters read the bare beaten distance the same whatever the track shape, so a wide trip at a sharp track is under-credited more than one at a galloping track. If the edge is equally large at galloping tracks, the market is missing something other than ground lost on bends. That could be a remark-tag effect, where the 'raced wide' comment marks horses the market under-rates for other reasons.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.00 to 1.08 on 300+ bets, clearly below the 1.18 the same base shows at sharp last-time tracks",
      "falsified_if": "A/E 1.15 or higher at galloping tracks, about equal to or above the sharp-track 1.18. Track shape would then not matter and the ground-on-bends mechanism would be wrong. A/E clearly below 1.00 would also contradict the mechanism, which predicts a smaller but still non-negative wide-trip credit."
    },
    "notes": "This is a mechanism test, not an edge hunt. The pre-registered prediction expects this rule to fall short of the discovery floor (A/E 1.05 or more at p 0.05). A FAIL at about 1.00 to 1.08 together with a sharp-track control at about 1.18 supports the bend mechanism. A pass at 1.15 or above falsifies it and points to a remark-tag effect. The tilt exclusion is left out on purpose so that only track shape differs from the sharp-track sibling. The sharpness coding (0 = galloping) is assumed from the sibling's >= 2 = sharp; the catalogue gives no value labels. Verify the code before measuring. Refinements are only looked at if the broad rule shows evidence. Judge the 16/1+ split as the known longshot artefact, not as support."
  }

Timeline

1. Job 1286 finished

   16 Oct, 12:12 · engine:compute-1

2. hypothesis note H-463f02e9

   16 Oct, 12:12 · engine:perturb

3. perturbation result for H-463f02e9 (neighbour: drop: last_position > 1): pass

   16 Oct, 12:12 · engine:perturb

4. perturbation result for H-463f02e9 (neighbour: drop: last_raced_wide = 1): pass

   16 Oct, 12:12 · engine:perturb

5. perturbation result for H-463f02e9 (neighbour: <= 5 -> 6 (+20%)): pass

   16 Oct, 12:12 · engine:perturb

6. perturbation result for H-463f02e9 (neighbour: <= 5 -> 4 (-20%)): pass

   16 Oct, 12:12 · engine:perturb

7. perturbation result for H-463f02e9 (neighbour: > 1 -> 2 (+20%)): pass

   16 Oct, 12:12 · engine:perturb

8. perturbation result for H-463f02e9 (neighbour: > 1 -> 0 (-20%)): pass

   16 Oct, 12:12 · engine:perturb

9. compute-1 took job 1286 (mutate, attempt 1)

   16 Oct, 12:12 · compute-1

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

    16 Oct, 12:12 · engine:scheduler

11. Job 1281 finished

    16 Oct, 12:12 · engine:compute-1

12. forward result for H-463f02e9 (forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}): info

    16 Oct, 12:12 · engine:evaluator

13. compute-1 took job 1281 (forward_check, attempt 1)

    16 Oct, 12:12 · compute-1

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

    16 Oct, 12:12 · engine:scheduler

15. Job 1276 finished

    16 Oct, 12:12 · engine:compute-1

16. H-463f02e9 Frozen → SHADOW vault skipped: last_raced_wide, last_beaten_lengths, last_slow_start do not exist before the discovery zone; track it forward instead

    16 Oct, 12:12 · engine:orchestrator