Turf-to-all-weather handicap returners after 150+ days beat their prices in any month

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-ce91b64d (All-weather spring returners carry the calendar lift, not turf returne) passed discovery as a niche: A/E 1.2 on 480 bets (p 0.11), same-race control 0.96. O-0ab1a07a
   • H-ce91b64d (All-weather spring returners carry the calendar lift, not turf returne): the country 'GB' carries the effect (A/E 1.23) while the rest score IRE 0.45. O-389ee4f8
2. AskedIs the returner lift a year-round all-weather effect, strongest when a horse switches from a turf campaign to the AW after 150+ days, rather than anything seasonal? answered
3. MechanismA horse that last ran on turf and returns to an AW handicap after 150+ days does so by its trainer's choice. AW programmes run all year, so the trainer can pick the day the horse is ready and the race suits. It meets exposed AW regulars on a mark set on turf.
4. Why the market misses itPunters price AW handicaps on recent AW figures. A turf horse back from a long break has none, only old turf form, so the market discounts it twice: once for rust and once for the surface.
5. Predictedpositive, A/E 1.10 to 1.25 on 600+ bets
6. Falsified ifA/E below 1.03 on 600+ bets, or the months outside March-April scoring below 1.0, which would put the lift back in the calendar

Mechanism and rationale

A horse that last ran on turf and returns to an AW handicap after 150+ days does so by its trainer's choice. AW programmes run all year, so the trainer can pick the day the horse is ready and the race suits. It meets exposed AW regulars on a mark set on turf.

The rule

race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND days_since_last_run >= 150

Reads: race_type, is_handicap, surface, last_surface, days_since_last_run

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

4ba4ce9f7cd7320977b69083175f0fb6575360941954f1f523fa53cca4264ea5

Frozen

16 Oct, 14:13

Nothing in the definition can change after this.

Confounders

• Not independent: built from discovery-zone neighbours of H-ce91b64d (AW March-April returners at 1.195 on 480), so the broad rule's result will be partly inherited from that spring subset; only the outside-March-April refinement tests anything new
• Late-autumn/winter AW returners may be turf horses dropping in class after a turf season; race_class is not available for IRE in the window, so the class drop cannot be cleanly separated
• GB-heavy: Dundalk is the only Irish AW track, and Ireland has underperformed across this thread
• The discovery window covers only about one year, so one season's crop or one set of yards may drive the result
• Trainer selection: the yards that switch fresh turf horses to AW may be strong yards whose edge would show anyway
• Price band: earlier returner lifts sat in the short bands, but AW 16/1+ returners scored 1.28 on 626, so the result should be checked by price band for concentration

Control

{
  "description": "Same-race other runners should score about 0.95-1.0 (overround-neutral). A second control separates the surface switch from plain freshness: AW handicap returners after 150+ days whose last run was also on AW. If the turf-switch mechanism is real, these should score clearly below the main rule (about 1.0 or lower). If they match it, the lift belongs to long-break AW returners in general, not to the turf-to-AW switch.",
  "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'AW' AND days_since_last_run >= 150"
}

Discovery result

pass broad rule (niche) · 16 Oct, 14:05

Bets

654

64 winners

A/E

1.23

95%: 0.95 to 1.57

SP return

-14.7%

level stakes

Betfair SP return

+2.8%

642 bets with a BSP

Strike rate 9.8%. p = 0.060. 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 5 survived.

• perturbation: Break threshold loosened to 120 dayspass

  899 bets · A/E 1.29 (1.04 to 1.59) · control 0.96 · 16 Oct, 14:12

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-277391.31

  after 2026-04-271601.22

• perturbation: Break threshold tightened to 180 dayspass

  446 bets · A/E 1.09 (0.77 to 1.50) · control 0.99 · 16 Oct, 14:13

Sealed holdout (vault)

Bets

3,155

16 Oct, 14:13

A/E

0.98

p = 0.64

BH threshold then

0.0024

over every vault test so far

Outcome

Failed

Forward record

No forward record. Shadow tracking starts after the vault and counts only races run after the freeze.

Other experiments

• neighbour: >= 150 -> 120 (-20%)pass

  899 bets · A/E 1.29 (1.04 to 1.59) · control 0.96 · 16 Oct, 14:05

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-277391.31

  after 2026-04-271601.22

• neighbour: >= 150 -> 180 (+20%)pass

  446 bets · A/E 1.09 (0.77 to 1.50) · control 0.99 · 16 Oct, 14:05

  Breakdown

Lineage

Agent runs

• designer v1ok · 1m 6s · 16 Oct, 14:03

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND days_since_last_run >= 150",
    "refinements": [
      {
        "label": "outside March-April (the calendar falsification test)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND days_since_last_run >= 150 AND month NOT IN (3, 4)"
      },
      {
        "label": "GB only (the thread's Ireland drag)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND days_since_last_run >= 150 AND region = 'GB'"
      }
    ],
    "control": {
      "description": "Same-race other runners should score about 0.95-1.0 (overround-neutral). A second control separates the surface switch from plain freshness: AW handicap returners after 150+ days whose last run was also on AW. If the turf-switch mechanism is real, these should score clearly below the main rule (about 1.0 or lower). If they match it, the lift belongs to long-break AW returners in general, not to the turf-to-AW switch.",
      "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'AW' AND days_since_last_run >= 150"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.08,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Not independent: built from discovery-zone neighbours of H-ce91b64d (AW March-April returners at 1.195 on 480), so the broad rule's result will be partly inherited from that spring subset; only the outside-March-April refinement tests anything new",
      "Late-autumn/winter AW returners may be turf horses dropping in class after a turf season; race_class is not available for IRE in the window, so the class drop cannot be cleanly separated",
      "GB-heavy: Dundalk is the only Irish AW track, and Ireland has underperformed across this thread",
      "The discovery window covers only about one year, so one season's crop or one set of yards may drive the result",
      "Trainer selection: the yards that switch fresh turf horses to AW may be strong yards whose edge would show anyway",
      "Price band: earlier returner lifts sat in the short bands, but AW 16/1+ returners scored 1.28 on 626, so the result should be checked by price band for concentration"
    ],
    "mechanism": "A horse whose last run was on turf and that comes back in an AW Flat handicap after 150+ days is back because its trainer chose to bring it back then. AW racing runs all year, so the trainer can pick the day the horse is fit and the race suits it. It then meets exposed AW regulars off a mark set on turf, often a mark that has eased during the turf campaign or is soft by AW standards. If this is right, the lift should show in every month, not only in March-April, and it should beat AW-to-AW returners after the same break.",
    "market_miss": "Punters price AW handicaps on recent AW form. A turf horse back from a long break has neither recent form nor AW figures, only old turf form, so the market marks it down twice: once for rust and once for the surface change. Neither mark-down allows for the trainer having picked the moment.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.25 on 600+ bets",
      "falsified_if": "Broad-rule A/E below 1.03 on 600+ bets; or the outside-March-April refinement scoring below 1.0 (the lift is calendar, not surface switch); or the AW-to-AW returner control scoring as high as the main rule (the lift is freshness, not the switch)"
    },
    "notes": "I checked the catalogue: surface, last_surface and month are pre-off and available from 2020-07 in both regions. The rule sketch is kept as it was. I added no price cap because the findings show AW 16/1+ returners did well (1.28 on 626), so the reasoning does not call for one. A class-drop refinement was left out because race_class only starts for IRE in 2026-07. The pass criteria are stricter than the floors (300 bets, A/E 1.08), because the rule overlaps H-ce91b64d and should clear more than the minimum to count as new evidence."
  }

Timeline

1. Job 1770 finished

   16 Oct, 14:13 · engine:compute-1

2. H-98ae1db9 Vault → KILLEDfailed the vault vault CONTRADICTED: A/E 0.979 on 3155 bets, p 0.6362 against BH threshold 0.002381 over 42 vault tests

   16 Oct, 14:13 · engine:orchestrator

3. vault result for H-98ae1db9 (vault query (contradicted)): fail

   16 Oct, 14:13 · engine:orchestrator

4. H-98ae1db9 Frozen → FROZENvault vault query started

   16 Oct, 14:13 · engine:orchestrator

5. compute-1 took job 1770 (vault, attempt 1)

   16 Oct, 14:13 · compute-1

6. Job 1768 finished

   16 Oct, 14:13 · engine:compute-1

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

   16 Oct, 14:13 · engine:scheduler

8. H-98ae1db9 Falsification → FROZEN survived falsification

   16 Oct, 14:13 · engine:falsification

9. falsification result for H-98ae1db9 (control: Control: turf-to-AW switchers after a medium 60-149 day gap): pass

   16 Oct, 14:13 · engine:evaluator

10. compute-1 took job 1768 (attack, attempt 1)

    16 Oct, 14:13 · compute-1

11. Job 1767 finished

    16 Oct, 14:13 · engine:compute-1

12. falsification result for H-98ae1db9 (price_band: 14/1 or shorter only: strip the longshot tail): pass

    16 Oct, 14:13 · engine:evaluator

13. compute-1 took job 1767 (attack, attempt 1)

    16 Oct, 14:13 · compute-1

14. Job 1766 finished

    16 Oct, 14:13 · engine:compute-1

15. falsification result for H-98ae1db9 (partition: Outside March-April: the year-round claim): pass

    16 Oct, 14:13 · engine:evaluator

16. compute-1 took job 1766 (attack, attempt 1)

    16 Oct, 14:13 · compute-1

17. Job 1765 finished

    16 Oct, 14:13 · engine:compute-1

18. falsification result for H-98ae1db9 (perturbation: Break threshold tightened to 180 days): pass

    16 Oct, 14:13 · engine:evaluator

19. compute-1 took job 1765 (attack, attempt 1)

    16 Oct, 14:12 · compute-1