Lightly raced four-year-old returners in Flat non-handicaps: progress the market doubts, with no mark involved

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-b05a33c3 (Lightly raced four-year-old returners on the all-weather: is the edge ) met the sealed vault supportive: A/E 1.1 on 698 bets (p 0.2014, threshold 0.0032). O-3b4db8ee
   • H-568b7703 (Lightly raced four-year-old returners whose mark was raised: improvers) survived falsification: 1 of 1 attacks left it standing; it is frozen for its one vault query. O-6bb82233
   • H-568b7703 (Lightly raced four-year-old returners whose mark was raised: improvers) passed discovery as a niche: A/E 1.38 on 41 bets (p 0.231), same-race control 0.94. O-f62c7b90
2. AskedIs the lift a property of the handicap mark, or does it also show in Flat non-handicaps, where there is no mark to lag? testing
3. MechanismA lightly raced four-year-old can keep strengthening over the winter. The market anchors on its last visible three-year-old form whatever the race conditions, so the progress should be underpriced in conditions races too.
4. Why the market misses itThe form book shows only three-year-old runs that are months old, and there is no fresh evidence of the winter's development. In non-handicaps, the weight-for-age and penalty structure hides this further.
5. Predictedpositive, A/E 1.05 to 1.12 on 300+ bets
6. Falsified ifA/E below 1.00 on 300+ bets. That would tie the effect to handicaps and the mark, not to underrated progress.
7. Rule sketchrace_type = 'Flat' AND is_handicap = 0 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150

Mechanism and rationale

A lightly raced four-year-old can keep strengthening over the winter. The market anchors on its last visible three-year-old form whatever the race conditions, so the progress should be underpriced in conditions races too.

The rule

race_type = 'Flat' AND is_handicap = 0 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150

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

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

6e6b202ab0e373b710e6390f1fd99dad85e5b19218ec0eabfead43d273ad61a0

Frozen

not frozen

The definition can still change.

Confounders

• Group and Listed races are dominated by top yards and short-priced, well-known horses, where progress is already priced in. race_class cannot separate them in Ireland before 2026, so check the result by price band.
• Maidens hold a weaker, different population of four-year-olds (horses that have not yet won), so check the maiden/non-maiden split
• The field size, class and price mix of non-handicaps differs from handicaps, so compare against same-race runners, not against the handicap result
• Long-priced runners in conditions races can drag A/E down or inflate it on a few wins. In earlier discovery samples 16/1+ returners went 0 from 48.
• Region mix: Irish spring conditions races are dominated by big yards. Check GB and IRE separately.
• Top yards may give winter-developed horses targeted spring returns, so a trainer effect could be mistaken for horse progress

Control

{
  "description": "Main comparison: the other runners in the same races. If the mechanism holds they should be near A/E 1.00 (or slightly under) while the four-year-old returners are above it. If the other runners show a similar lift, the effect belongs to the races (spring conditions races, field mix), not to the horses. Second comparison: the same profile in spring Flat handicaps (the original H126 rule). That is already measured at 1.12 on 2958 and acts as the reference the non-handicap result is read against, so no new control rule is needed.",
  "rule": null
}

Discovery result

pass broad rule (niche) · 16 Oct, 12:35

Bets

182

26 winners

A/E

1.07

95%: 0.70 to 1.57

SP return

-1.7%

level stakes

Betfair SP return

+54.5%

182 bets with a BSP

Strike rate 14.3%. p = 0.38. Same-race control: the other runners in these races had A/E 0.96, 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: 14/1 or shorter only: strip the 2 long-price winnerspass

  118 bets · A/E 1.04 (0.67 to 1.54) · control 0.96 · 16 Oct, 12:43

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27610.97

  after 2026-04-27571.13

• partition: Non-maidens only: horses with winning formpass

  134 bets · A/E 1.14 (0.72 to 1.71) · control 0.93 · 16 Oct, 12:43

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: <= 10 -> 8 (-20%)pass

  171 bets · A/E 1.12 (0.72 to 1.65) · control 0.95 · 16 Oct, 12:43

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27941.06

  after 2026-04-27771.19

• neighbour: <= 10 -> 12 (+20%)pass

  189 bets · A/E 1.04 (0.68 to 1.52) · control 0.97 · 16 Oct, 12:43

  Breakdown

Lineage

Agent runs

• designer v1ok · 30s · 16 Oct, 12:34

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 0 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "race_type = 'Flat' AND is_handicap = 0 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_decimal <= 15"
      },
      {
        "label": "Non-maidens only",
        "rule": "race_type = 'Flat' AND is_handicap = 0 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND is_maiden = 0"
      },
      {
        "label": "Non-maidens at 14/1 or shorter",
        "rule": "race_type = 'Flat' AND is_handicap = 0 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND is_maiden = 0 AND sp_decimal <= 15"
      }
    ],
    "control": {
      "description": "Main comparison: the other runners in the same races. If the mechanism holds they should be near A/E 1.00 (or slightly under) while the four-year-old returners are above it. If the other runners show a similar lift, the effect belongs to the races (spring conditions races, field mix), not to the horses. Second comparison: the same profile in spring Flat handicaps (the original H126 rule). That is already measured at 1.12 on 2958 and acts as the reference the non-handicap result is read against, so no new control rule is needed.",
      "rule": null
    },
    "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": [
      "Group and Listed races are dominated by top yards and short-priced, well-known horses, where progress is already priced in. race_class cannot separate them in Ireland before 2026, so check the result by price band.",
      "Maidens hold a weaker, different population of four-year-olds (horses that have not yet won), so check the maiden/non-maiden split",
      "The field size, class and price mix of non-handicaps differs from handicaps, so compare against same-race runners, not against the handicap result",
      "Long-priced runners in conditions races can drag A/E down or inflate it on a few wins. In earlier discovery samples 16/1+ returners went 0 from 48.",
      "Region mix: Irish spring conditions races are dominated by big yards. Check GB and IRE separately.",
      "Top yards may give winter-developed horses targeted spring returns, so a trainer effect could be mistaken for horse progress"
    ],
    "mechanism": "A lightly raced four-year-old (10 or fewer career runs) can keep strengthening physically over the winter. On its first run back after 150+ days in spring, the market anchors on its months-old three-year-old form. If the underpricing comes from doubt about continued progress rather than from a lagging handicap mark, it should also appear in non-handicaps, where there is no mark. This matters more because the rating_change split went against the mark-lag account: the not-raised version scored 0.95.",
    "market_miss": "The only visible form is stale three-year-old form, and there is no public evidence of the winter's development. In conditions races, weight-for-age allowances and penalties are fixed by the conditions and do not reflect improvement, so bettors read the four-year-old against older rivals through old form and underrate it.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.05 to 1.12 on 300+ bets, with the lift concentrated at 14/1 or shorter",
      "falsified_if": "A/E below 1.00 on 300+ bets, or same-race other runners showing an equal or larger lift. Either would tie the returner effect to handicaps and the mark (or to the races), not to underrated progress."
    },
    "notes": "I kept the rule sketch unchanged: every column is pre-off, and it differs from H126 only in is_handicap = 0, so it answers question Q-39f14d17 directly. I set min_bets at 300 (stricter than the floor) to match the predicted sample. The refinements follow from the earlier findings (no edge at 16/1+) and from the maiden confounder. I did not add a race_class refinement because Irish race_class only starts in 2026-07."
  }

Timeline

1. Job 1390 finished

   16 Oct, 12:43 · engine:compute-1

2. hypothesis note H-1a3e129f

   16 Oct, 12:43 · engine:perturb

3. perturbation result for H-1a3e129f (neighbour: drop: is_handicap = 0): pass

   16 Oct, 12:43 · engine:perturb

4. perturbation result for H-1a3e129f (neighbour: drop: race_type = 'Flat'): fail

   16 Oct, 12:43 · engine:perturb

5. perturbation result for H-1a3e129f (neighbour: >= 150 -> 180 (+20%)): pass

   16 Oct, 12:43 · engine:perturb

6. perturbation result for H-1a3e129f (neighbour: >= 150 -> 120 (-20%)): pass

   16 Oct, 12:43 · engine:perturb

7. perturbation result for H-1a3e129f (neighbour: <= 10 -> 12 (+20%)): pass

   16 Oct, 12:43 · engine:perturb

8. perturbation result for H-1a3e129f (neighbour: <= 10 -> 8 (-20%)): pass

   16 Oct, 12:43 · engine:perturb

9. compute-1 took job 1390 (mutate, attempt 1)

   16 Oct, 12:43 · compute-1

10. Job 1416 finished

    16 Oct, 12:43 · engine:compute-1

11. H-1a3e129f Falsification → KILLED falsified: IRE A/E 0.861; driven by one price band (up to 3/1): the rest score 1.00

    16 Oct, 12:43 · engine:falsification

12. falsification result for H-1a3e129f (control: Control: same four-year-olds that are not returners (14 to 60 days)): fail

    16 Oct, 12:43 · engine:evaluator

13. compute-1 took job 1416 (attack, attempt 1)

    16 Oct, 12:43 · compute-1

14. Job 1415 finished

    16 Oct, 12:43 · engine:compute-1

15. falsification result for H-1a3e129f (perturbation: Looser thresholds: 12 or fewer runs, 120+ days off): pass

    16 Oct, 12:43 · engine:evaluator

16. compute-1 took job 1415 (attack, attempt 1)

    16 Oct, 12:43 · compute-1