Rival: short-priced spring handicap returners pay at any age, so H126's short band is not about four-year-olds

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-5998bc20 (Lightly raced spring handicap returners that won last time, at age thr) passed discovery as a niche: A/E 1.27 on 185 bets (p 0.11), same-race control 0.95. O-46d14312
   • H-5998bc20 (Lightly raced spring handicap returners that won last time, at age thr) passed discovery but was falsified: driven by one price band (up to 3/1): the rest score 0.90. O-c2bb350a
2. AskedDo short-priced (3/1 or shorter) Flat handicap returners after 150+ days in spring beat their prices at any age and experience level, which would make the H126 short-band lift a general returner-confidence effect? answered
3. MechanismWhen a horse returns from a long break, the trainer decides whether it runs and the money shows whether the yard thinks it is fit. A short price on a returner after 150+ days reflects stable confidence, but the public and the compilers still shade it for rust. That discount would apply at any age.
4. Why the market misses itThe form book shows a long absence. Odds compilers apply a standard 'needs the run' discount even when market support says the yard has the horse ready, so fancied returners go off slightly too long.
5. Predictedpositive, A/E 1.05 to 1.15 on 300+ bets across all ages
6. Falsified ifA/E below 1.02 on 300+ bets, or ages 5 and over below 0.97 while ages 3 and 4 carry the lift. Either would point the short-band effect back to young, lightly raced horses.

Mechanism and rationale

When a horse returns from a long break, the trainer decides whether it runs and the money shows whether the yard thinks it is fit. A short price on a returner after 150+ days reflects stable confidence, but the public and the compilers still shade it for rust. That discount would apply at any age.

The rule

race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run >= 150 AND sp_decimal <= 4

Reads: race_type, is_handicap, month, days_since_last_run, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

586eabc2ac54a0e0666c1811e6ac76389e80b362b877c47ecb2e0991450d7364

Frozen

not frozen

The definition can still change.

Confounders

• Favourite-longshot bias: short prices score near 1.0 everywhere, so the rule must beat the short-priced non-returner control and not just the same-race runners
• Overlap with the H126 four-year-olds: their up-to-3/1 band scored 1.97 and could carry the whole rule, so the age <> 4 and age >= 5 refinements must be read separately
• Only one or two springs in the discovery window: one hot season, yard or meeting can dominate, so check the periods split
• Small Irish samples. Earlier rules in this thread died on Ireland, so read the GB and IRE splits
• Few bets at 3/1 or shorter means few losers at high strike rates, so a handful of results can swing A/E
• Prior findings show a general mild returner lift (1.07 on 971), so a positive result may reflect returners in general and not stable-confidence pricing

Control

{
  "description": "Same-race other runners (lab default) should score at or below 1.0, since they absorb the overround and the longshot tail. That comparison is confounded by favourite-longshot bias, because short prices score near 1.0 everywhere. The key control is short-priced spring Flat handicappers that are NOT returning from a break, at the same price band and in the same months. If the mechanism is a rust discount on fancied returners, the control should score about 1.0 or below (expect 0.95 to 1.02), and the returner rule should beat it by at least 0.05 A/E. If the control scores as high as the rule, the lift is a property of short prices in spring handicaps, not of returners.",
  "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run < 150 AND sp_decimal <= 4"
}

Discovery result

pass broad rule (niche) · 16 Oct, 13:28

Bets

141

48 winners

A/E

1.10

95%: 0.81 to 1.46

SP return

+2.0%

level stakes

Betfair SP return

+11.7%

141 bets with a BSP

Strike rate 34.0%. p = 0.27. 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.

• partition: Rival test: ages other than fourpass

  70 bets · A/E 1.08 (0.68 to 1.63) · control 0.96 · 16 Oct, 13:34

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27511.14

  after 2026-04-2719–

• partition: Exposed returners only (more than 10 prior runs)info

  25 bets · A/E 1.15 (0.49 to 2.26) · control 0.94 · 16 Oct, 13:34

  Breakdown

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

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

  160 bets · A/E 1.06 (0.79 to 1.39) · control 0.97 · 16 Oct, 13:28

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271051.11

  after 2026-04-27550.94

• neighbour: >= 150 -> 180 (+20%)pass

  97 bets · A/E 1.05 (0.72 to 1.48) · control 0.98 · 16 Oct, 13:28

  Breakdown

Lineage

Agent runs

• designer v1ok · 30s · 16 Oct, 13:28

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run >= 150 AND sp_decimal <= 4",
    "refinements": [
      {
        "label": "Ages other than four (the direct rival test: lift without H126's four-year-olds)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run >= 150 AND sp_decimal <= 4 AND age <> 4"
      },
      {
        "label": "Ages five and over (the falsification arm named in the prediction)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run >= 150 AND sp_decimal <= 4 AND age >= 5"
      },
      {
        "label": "Exposed returners (more than 10 prior runs): is the lift free of light racing?",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run >= 150 AND sp_decimal <= 4 AND prior_runs > 10"
      }
    ],
    "control": {
      "description": "Same-race other runners (lab default) should score at or below 1.0, since they absorb the overround and the longshot tail. That comparison is confounded by favourite-longshot bias, because short prices score near 1.0 everywhere. The key control is short-priced spring Flat handicappers that are NOT returning from a break, at the same price band and in the same months. If the mechanism is a rust discount on fancied returners, the control should score about 1.0 or below (expect 0.95 to 1.02), and the returner rule should beat it by at least 0.05 A/E. If the control scores as high as the rule, the lift is a property of short prices in spring handicaps, not of returners.",
      "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND days_since_last_run < 150 AND sp_decimal <= 4"
    },
    "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
    },
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.05 to 1.15 on 300+ bets across all ages, with ages other than four also at or above about 1.05",
      "falsified_if": "A/E below 1.02 on 300+ bets; or ages 5+ below 0.97 while ages 3 and 4 carry the lift; or the non-returner short-price control scores within 0.03 of the rule. Each would put the short-band effect back on young, lightly raced horses or on price alone."
    },
    "confounders": [
      "Favourite-longshot bias: short prices score near 1.0 everywhere, so the rule must beat the short-priced non-returner control and not just the same-race runners",
      "Overlap with the H126 four-year-olds: their up-to-3/1 band scored 1.97 and could carry the whole rule, so the age <> 4 and age >= 5 refinements must be read separately",
      "Only one or two springs in the discovery window: one hot season, yard or meeting can dominate, so check the periods split",
      "Small Irish samples. Earlier rules in this thread died on Ireland, so read the GB and IRE splits",
      "Few bets at 3/1 or shorter means few losers at high strike rates, so a handful of results can swing A/E",
      "Prior findings show a general mild returner lift (1.07 on 971), so a positive result may reflect returners in general and not stable-confidence pricing"
    ],
    "mechanism": "After a break of 150+ days, the trainer chooses whether the horse runs, and the money shows whether the yard thinks it is fit. A returner sent off at 3/1 or shorter has been backed on stable confidence. The public and the compilers still shade it for rust. That discount comes from the absence, not the horse's age, so it should appear across ages 3, 4 and 5+ and in exposed horses as well as lightly raced ones.",
    "market_miss": "The form book shows a long absence, so compilers and the public apply a standard 'needs the run' discount. They apply it even when market support shows the yard has the horse ready. The price shortens but not far enough, so fancied returners go off slightly too long compared with fancied horses that ran recently.",
    "notes": "This is a rival explanation for H126's short-band lift (four-year-old returners up to 3/1 at 1.97). The broad rule removes the age and prior_runs conditions on purpose. The refinements isolate ages other than four, ages five and over, and exposed horses. If the lift exists only where age = 4 overlaps, the rival fails even if the broad rule passes. The month column is assumed available, as in the earlier rules in this family. A minimum of 300 bets was chosen to match the prediction's sample size."
  }

Timeline

1. Job 1635 finished

   16 Oct, 13:34 · engine:compute-1

2. H-71a94a07 Falsification → KILLED falsified: driven by one price band (up to 3/1): the rest score 0.71; IRE: too few bets to judge

   16 Oct, 13:34 · engine:falsification

3. falsification result for H-71a94a07 (control: Short-break control: fancied runners 60 to 149 days off): pass

   16 Oct, 13:34 · engine:evaluator

4. compute-1 took job 1635 (attack, attempt 1)

   16 Oct, 13:34 · compute-1

5. Job 1634 finished

   16 Oct, 13:34 · engine:compute-1

6. falsification result for H-71a94a07 (perturbation: Tighter thresholds: 180+ days off, up to 5/2): pass

   16 Oct, 13:34 · engine:evaluator

7. compute-1 took job 1634 (attack, attempt 1)

   16 Oct, 13:34 · compute-1

8. Job 1633 finished

   16 Oct, 13:34 · engine:compute-1

9. falsification result for H-71a94a07 (perturbation: Looser thresholds: 120+ days off, up to 9/2): pass

   16 Oct, 13:34 · engine:evaluator

10. compute-1 took job 1633 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

11. Job 1632 finished

    16 Oct, 13:34 · engine:compute-1

12. falsification result for H-71a94a07 (partition: Exposed returners only (more than 10 prior runs)): info

    16 Oct, 13:34 · engine:evaluator

13. compute-1 took job 1632 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

14. Job 1631 finished

    16 Oct, 13:34 · engine:compute-1

15. falsification result for H-71a94a07 (partition: Rival test: ages other than four): pass

    16 Oct, 13:34 · engine:evaluator

16. compute-1 took job 1631 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

17. Job 1612 finished

    16 Oct, 13:34 · engine:ai-1

18. Queued attack on the compute lane: attack: Short-break control: fancied runners 60 to 149 days off

    16 Oct, 13:34 · engine:scheduler

19. Queued attack on the compute lane: attack: Tighter thresholds: 180+ days off, up to 5/2

    16 Oct, 13:34 · engine:scheduler