Wide-trip edge outside the longshot band

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
   • H-baf747c3 (Wide last time, drawn low today on the Flat) passed discovery as a niche: A/E 1.12 on 456 bets (p 0.187), same-race control 0.98. O-b284ee76
2. AskedDoes the wide-trip edge survive outside the 16/1+ band, or is it mostly the longshot lift seen in every beaten-last-time rule? answered
3. MechanismIf a wide trip truly hid ability, the effect should show among fancied runners too, not only among a handful of big-priced winners.
4. Why the market misses itPunters read the finishing position and beaten distance, not the extra ground covered. That applies at any price, so a real edge should not depend on 16/1+ shots.
5. Predictedpositive, A/E 1.08 to 1.15 on 450+ bets at SP up to 14/1
6. Falsified ifA/E 1.02 or lower at SP up to 14/1. The edge would then be mostly the known longshot artefact.
7. Rule sketchlast_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15

Mechanism and rationale

If a wide trip truly hid ability, the effect should show among fancied runners too, not only among a handful of big-priced winners.

The rule

last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15

Reads: last_raced_wide, last_position, last_beaten_lengths, last_slow_start, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

7c2a975a322f58ce96e4118dbfb0b695932d036a6d26832d34ebd6cd7e9512e0

Frozen

16 Oct, 12:12

Nothing in the definition can change after this.

Confounders

• Longshot artefact: beaten-last-time horses at 16/1+ show A/E above 1.10 across many triggers, and the price cap is there to remove it
• Few wins at 16/1+ make the excluded band noisy, so it should be read only as a control and never as evidence
• SP versus BSP gap at short prices: the edge may be smaller at the exchange, or vanish there, for fancied runners
• Country mix: GB and IRE may differ in how wide trips are recorded and priced
• Data window: last_raced_wide exists only from November 2025, so there is about one year in two halves and the result may depend on the period (the low-draw child was carried by post-April 2026)
• The yardstick must be corrected for field size as well as price
• The parent IQ140a used a tilt exclusion (held-up/midfield in tilted races) that this simpler base leaves out, so compare against the same simple base with no price cap, not against IQ140a's 1.27

Control

{
  "description": "Two comparisons. (1) Other runners in the same races priced 14/1 or shorter should come out near A/E 1.00, so any lift belongs to the wide-trip horses and not to the races. (2) The same wide-trip base at bigger than 14/1 (control rule) shows how much of the parent's 1.18-1.27 comes from the longshot band. Known longshot artefact: beaten-last-time horses at 16/1+ usually show A/E above 1.10 whatever the trigger. If the edge is real, the <=14/1 band should hold A/E >= 1.05 by itself. If it is the artefact, the >14/1 control carries the lift and the <=14/1 band sits near 1.00.",
  "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal > 15"
}

Discovery result

pass broad rule · 16 Oct, 12:11

Bets

974

211 winners

A/E

1.21

95%: 1.05 to 1.39

SP return

+6.3%

level stakes

Betfair SP return

+25.1%

973 bets with a BSP

Strike rate 21.7%. p = 0.0037. 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.

• control: Control: same base without the wide tripfail

  21,842 bets · A/E 0.99 (0.96 to 1.02) · control 1.00 · 16 Oct, 12:12

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-278,5460.97

  after 2026-04-2713,2961.00

• price_band: Fancied core: 5/1 or shorter onlypass

  530 bets · A/E 1.16 (0.99 to 1.36) · control 0.94 · 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

  1,391 bets · A/E 1.13 (1.00 to 1.26) · control 0.97 · 16 Oct, 12:22

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-276541.10

  after 2026-04-277371.15

• neighbour: > 1 -> 2 (+20%)pass

  645 bets · A/E 1.25 (1.04 to 1.48) · control 0.95 · 16 Oct, 12:22

  Breakdown

Lineage

Agent runs

• designer v1ok · 38s · 16 Oct, 12:07

  Output (schema-checked)

  {
    "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15",
    "refinements": [
      {
        "label": "fancied: 7/1 or shorter",
        "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND sp_decimal <= 8"
      },
      {
        "label": "GB only (country mix)",
        "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND region = 'GB'"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) Other runners in the same races priced 14/1 or shorter should come out near A/E 1.00, so any lift belongs to the wide-trip horses and not to the races. (2) The same wide-trip base at bigger than 14/1 (control rule) shows how much of the parent's 1.18-1.27 comes from the longshot band. Known longshot artefact: beaten-last-time horses at 16/1+ usually show A/E above 1.10 whatever the trigger. If the edge is real, the <=14/1 band should hold A/E >= 1.05 by itself. If it is the artefact, the >14/1 control carries the lift and the <=14/1 band sits near 1.00.",
      "rule": "last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal > 15"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Longshot artefact: beaten-last-time horses at 16/1+ show A/E above 1.10 across many triggers, and the price cap is there to remove it",
      "Few wins at 16/1+ make the excluded band noisy, so it should be read only as a control and never as evidence",
      "SP versus BSP gap at short prices: the edge may be smaller at the exchange, or vanish there, for fancied runners",
      "Country mix: GB and IRE may differ in how wide trips are recorded and priced",
      "Data window: last_raced_wide exists only from November 2025, so there is about one year in two halves and the result may depend on the period (the low-draw child was carried by post-April 2026)",
      "The yardstick must be corrected for field size as well as price",
      "The parent IQ140a used a tilt exclusion (held-up/midfield in tilted races) that this simpler base leaves out, so compare against the same simple base with no price cap, not against IQ140a's 1.27"
    ],
    "mechanism": "A horse beaten 5 lengths or less after racing wide (and not after a slow start) covered extra ground, so its true ability is better than its form figures show. If that hidden ability is real, it should raise the horse's win rate above its price among fancied runners (14/1 or shorter) as well. It should not appear only among a handful of big-priced winners.",
    "market_miss": "Punters and the form book read the finishing position and beaten distance, not the extra ground covered. That under-rating applies at any price, so a real edge should show in the 14/1-and-shorter band, where prices are formed most efficiently, and should not depend on 16/1+ shots.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.08 to 1.15 on 450+ bets at SP up to 14/1",
      "falsified_if": "A/E 1.02 or lower at SP up to 14/1, or a lift that appears in only one of the two time halves. The wide-trip edge would then be mostly the known longshot artefact of beaten-last-time horses."
    },
    "notes": "I kept the rule sketch: sp_decimal <= 15 is the same as 14/1 or shorter. Both refinements keep the full base, so only one thing changes. The reasoning suggests a 'fancied' cut (7/1 or shorter) because the mechanism says the effect should show among fancied runners too. A GB-only cut addresses the country-mix confounder. Pass min_bets is raised to 300 because the prediction expects 450+ bets. Report the A/E for each time half and the BSP result as well as SP."
  }

Timeline

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

   17 Oct, 12:12 · engine:scheduler

2. Job 1287 finished

   16 Oct, 12:22 · engine:compute-1

3. hypothesis note H-c8ec6922

   16 Oct, 12:22 · engine:perturb

4. perturbation result for H-c8ec6922 (neighbour: <= 15 -> 18 (+20%)): pass

   16 Oct, 12:22 · engine:perturb

5. perturbation result for H-c8ec6922 (neighbour: <= 15 -> 12 (-20%)): pass

   16 Oct, 12:22 · engine:perturb

6. perturbation result for H-c8ec6922 (neighbour: <= 5 -> 6 (+20%)): pass

   16 Oct, 12:22 · engine:perturb

7. perturbation result for H-c8ec6922 (neighbour: <= 5 -> 4 (-20%)): pass

   16 Oct, 12:22 · engine:perturb

8. perturbation result for H-c8ec6922 (neighbour: > 1 -> 2 (+20%)): pass

   16 Oct, 12:22 · engine:perturb

9. perturbation result for H-c8ec6922 (neighbour: > 1 -> 0 (-20%)): pass

   16 Oct, 12:22 · engine:perturb

10. compute-1 took job 1287 (mutate, attempt 1)

    16 Oct, 12:22 · compute-1

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

    16 Oct, 12:12 · engine:scheduler

12. Job 1282 finished

    16 Oct, 12:12 · engine:compute-1

13. forward result for H-c8ec6922 (forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}): info

    16 Oct, 12:12 · engine:evaluator

14. compute-1 took job 1282 (forward_check, attempt 1)

    16 Oct, 12:12 · compute-1

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

    16 Oct, 12:12 · engine:scheduler

16. Job 1277 finished

    16 Oct, 12:12 · engine:compute-1

17. H-c8ec6922 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