Lightly raced four-year-olds back from the winter on turf only

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-3cdb61f5 (Second run after the winter break for lightly raced four-year-olds) pointed the predicted way but failed the bar: A/E 1.06 on 459 bets (p 0.353), control 0.99. O-2c8b0b64
   • H-3cdb61f5 (Second run after the winter break for lightly raced four-year-olds): the period 'after 2026-04-27' carries the effect (A/E 1.16) while the rest score 2025-10-01 to 2026-04-27 0.61. O-c392fc29
   • H-625f1f2d (Fewest prior runs: the mark has the least information) passed discovery as a niche: A/E 1.18 on 218 bets (p 0.236), same-race control 0.98. O-e2338df0
2. AskedIs the H126 lift and its second-run follow-up a late-spring turf effect, weak or absent on the all-weather in March and early April? answered
3. MechanismFour-year-olds returning on turf in spring meet handicappers who are still rated on last season's form. Horses that spent the winter on the all-weather have already been re-assessed. So the mark lag should show on turf and fade on the all-weather.
4. Why the market misses itThe form book shows a three-year-old turf mark and a long absence. Punters cannot see how much the horse has strengthened, and the all-weather form some rivals bring makes those rivals look more current.
5. Predictedpositive, A/E 1.12 to 1.20 on 400+ bets on turf, with the all-weather part near 1.00
6. Falsified if

Mechanism and rationale

Four-year-olds returning on turf in spring meet handicappers who are still rated on last season's form. Horses that spent the winter on the all-weather have already been re-assessed. So the mark lag should show on turf and fade on the all-weather.

The rule

race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND surface = 'Turf'

Reads: race_type, is_handicap, month, age, prior_runs, days_since_last_run, surface

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

54728436aa2c96301b37f2833ca1f7654130a8697a09606b7d6c294749c1c233

Frozen

16 Oct, 12:22

Nothing in the definition can change after this.

Confounders

• Surface and calendar are confounded: GB turf Flat starts late March, so turf runners are mostly April-May and AW runners mostly March; with one post-2025 spring, a turf lift may just be a late-spring lift (the April-May refinement and month split test this).
• Field sizes are larger on turf, which lowers win probability and can distort A/E at longer prices.
• Irish turf share differs from GB (Irish AW is one course), so region mix differs between the turf and AW groups; split by region.
• Longshot artefact: 16/1+ lifts are common across lab rules; the 14/1-or-shorter refinement checks the edge is not carried by a few big-priced winners.
• Selection edge only at better-than-SP prices (H126 BSP ROI +1.6% vs SP ROI -19.6%): A/E against SP may overstate what is bettable.
• The turf rule is a subset of H126 (A/E 1.12 on 2958), so a positive result is partly inherited, not independent evidence.

Control

{
  "description": "Two comparisons. (1) Same-race other runners on turf: expected near A/E 1.00 (H126's control was 0.98), showing the lift belongs to the returning four-year-olds, not to spring turf handicaps. (2) The identical selection run on the all-weather (only the surface flipped): the mechanism predicts it near 1.00 and clearly below turf. If the AW group scores as high as turf, the surface story is falsified and the lift is a general freshness/age effect. Because turf Flat in GB starts late March, the AW group is March-heavy; read the surface split alongside month to avoid confusing surface with calendar.",
  "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND surface = 'AW'"
}

Discovery result

pass broad rule (niche) · 16 Oct, 12:20

Bets

379

44 winners

A/E

1.20

95%: 0.87 to 1.61

SP return

-13.1%

level stakes

Betfair SP return

+9.2%

379 bets with a BSP

Strike rate 11.6%. p = 0.13. Same-race control: the other runners in these races had A/E 0.97, 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. 5 of 6 survived.

• control: Surface flip: identical selection on the all-weatherpass

  140 bets · A/E 1.18 (0.71 to 1.84) · control 0.97 · 16 Oct, 12:22

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271121.30

  after 2026-04-2728–

• control: Not fresh: lightly raced 4yo spring turf handicappers who ran within 60 daysfail

  647 bets · A/E 0.99 (0.77 to 1.25) · control 1.01 · 16 Oct, 12:22

  Breakdown

Sealed holdout (vault)

Bets

1,741

16 Oct, 12:22

A/E

1.10

p = 0.10

BH threshold then

0.0033

over every vault test so far

Outcome

Failed

Forward record

• forward vs bar {'min_bets': 150, 'min_ae': 1.0, 'max_p': 0.2}info

  0 bets · · 16 Oct, 12:22

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: <= 10 -> 8 (-20%)pass

  296 bets · A/E 1.24 (0.87 to 1.73) · control 0.96 · 16 Oct, 12:29

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271471.26

  after 2026-04-271491.23

• neighbour: <= 10 -> 12 (+20%)pass

  445 bets · A/E 1.17 (0.86 to 1.55) · control 0.97 · 16 Oct, 12:29

  Breakdown

Lineage

Agent runs

• designer v1ok · 1m 15s · 16 Oct, 12:13

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND surface = 'Turf'",
    "refinements": [
      {
        "label": "Late spring only (April-May)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND surface = 'Turf'"
      },
      {
        "label": "14/1 or shorter (edges die at big prices; known 16/1+ longshot artefact)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND surface = 'Turf' AND sp_decimal <= 15"
      },
      {
        "label": "Most lightly raced (5 or fewer runs: mark carries least information)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 5 AND days_since_last_run >= 150 AND surface = 'Turf'"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) Same-race other runners on turf: expected near A/E 1.00 (H126's control was 0.98), showing the lift belongs to the returning four-year-olds, not to spring turf handicaps. (2) The identical selection run on the all-weather (only the surface flipped): the mechanism predicts it near 1.00 and clearly below turf. If the AW group scores as high as turf, the surface story is falsified and the lift is a general freshness/age effect. Because turf Flat in GB starts late March, the AW group is March-heavy; read the surface split alongside month to avoid confusing surface with calendar.",
      "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND surface = 'AW'"
    },
    "pass_criteria": {
      "min_bets": 400,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Surface and calendar are confounded: GB turf Flat starts late March, so turf runners are mostly April-May and AW runners mostly March; with one post-2025 spring, a turf lift may just be a late-spring lift (the April-May refinement and month split test this).",
      "Field sizes are larger on turf, which lowers win probability and can distort A/E at longer prices.",
      "Irish turf share differs from GB (Irish AW is one course), so region mix differs between the turf and AW groups; split by region.",
      "Longshot artefact: 16/1+ lifts are common across lab rules; the 14/1-or-shorter refinement checks the edge is not carried by a few big-priced winners.",
      "Selection edge only at better-than-SP prices (H126 BSP ROI +1.6% vs SP ROI -19.6%): A/E against SP may overstate what is bettable.",
      "The turf rule is a subset of H126 (A/E 1.12 on 2958), so a positive result is partly inherited, not independent evidence."
    ],
    "mechanism": "Lightly raced four-year-olds returning after 150+ days carry a mark set on three-year-old form and have physically strengthened over the winter. On turf in spring they meet fields whose handicappers are still working from last season's turf form, so the mark lags most there. Runners who spent the winter on the all-weather have been re-assessed recently, and AW handicaps in spring are full of such current-form rivals, so the returning four-year-old's relative advantage over its mark should be smaller on the all-weather.",
    "market_miss": "The form book shows a stale three-year-old turf rating and a long absence, which punters discount; the improvement from maturing over the winter is invisible. Rivals with recent all-weather form look more current and attract money, so on turf the returning four-year-old is under-bet relative to its true chance.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.12 to 1.20 on 400+ turf bets, with the identical all-weather selection near 1.00",
      "falsified_if": "Turf scores A/E below 1.05 on 400+ bets, or the all-weather control scores as high as or higher than turf (within 0.03), or the turf lift is confined to one month/period and vanishes within April-May matched comparisons."
    },
    "notes": "The sketch used surface = 'turf'; catalogue meaning is 'turf or all-weather' and measured lab rules use surface = 'AW' and last_surface = 'Turf', so the value was corrected to 'Turf' (verify value casing before measuring; a zero-bet result signals a casing mismatch, not a null). Answers Q-b27afe7d: is H126's lift a late-spring turf effect? The April-May refinement directly addresses the calendar side; the AW control addresses the surface side. Per the method finding, period splits in this family coincide with AW-vs-turf, so instability across periods must be read against surface."
  }

Timeline

1. Job 1323 finished

   16 Oct, 12:29 · engine:compute-1

2. hypothesis note H-29eb96fe

   16 Oct, 12:29 · engine:perturb

3. perturbation result for H-29eb96fe (neighbour: drop: is_handicap = 1): pass

   16 Oct, 12:29 · engine:perturb

4. perturbation result for H-29eb96fe (neighbour: drop: race_type = 'Flat'): pass

   16 Oct, 12:29 · engine:perturb

5. perturbation result for H-29eb96fe (neighbour: >= 150 -> 180 (+20%)): pass

   16 Oct, 12:29 · engine:perturb

6. perturbation result for H-29eb96fe (neighbour: >= 150 -> 120 (-20%)): pass

   16 Oct, 12:29 · engine:perturb

7. perturbation result for H-29eb96fe (neighbour: <= 10 -> 12 (+20%)): pass

   16 Oct, 12:29 · engine:perturb

8. perturbation result for H-29eb96fe (neighbour: <= 10 -> 8 (-20%)): pass

   16 Oct, 12:29 · engine:perturb

9. compute-1 took job 1323 (mutate, attempt 1)

   16 Oct, 12:29 · compute-1

10. Job 1320 finished

    16 Oct, 12:22 · engine:compute-1

11. forward result for H-29eb96fe (forward vs bar {'min_bets': 150, 'min_ae': 1.0, 'max_p': 0.2}): info

    16 Oct, 12:22 · engine:evaluator

12. compute-1 took job 1320 (forward_check, attempt 1)

    16 Oct, 12:22 · compute-1

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

    16 Oct, 12:22 · engine:scheduler

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

    16 Oct, 12:22 · engine:scheduler

15. Job 1318 finished

    16 Oct, 12:22 · engine:compute-1

16. H-29eb96fe Vault → SHADOW vault SUPPORTIVE: A/E 1.101 on 1741 bets, p 0.102 against BH threshold 0.003333 over 30 vault tests

    16 Oct, 12:22 · engine:orchestrator