Sharpen: proven hurdlers lightly raced on the Flat in Irish summer staying handicaps

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
   • H228-dual-purpose-horse-in-an-irish-summer-flat-handicap (H228: dual purpose horse in an irish summer flat handicap) was GREEN in the earlier research: A/E 1.05 on 1890 bets (periods 1.03 / 1.08 / 0.99); now PROMISING. O-a317bdab
2. AskedIs the lift strongest where the Flat mark has the least information, i.e. horses lightly raced on the Flat but proven over hurdles? answered
3. MechanismA Flat handicap mark is set from a few Flat runs, often run before the horse matured over hurdles. A horse that has won over hurdles but has few Flat runs carries a mark that has not caught up with what it can do now, which is exactly the lag the thread claims.
4. Why the market misses itThe Flat form book shows modest Flat figures. Hurdle wins sit in a different code, and Flat punters and compilers weight them lightly.
5. Predictedpositive, A/E 1.08 to 1.15 on 250+ bets, across all Irish courses
6. Falsified ifA/E at or below 1.00 on 200+ bets, or a lift only at 16/1 or bigger
7. Rule sketchregion = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs <= 6

Mechanism and rationale

A Flat handicap mark is set from a few Flat runs, often run before the horse matured over hurdles. A horse that has won over hurdles but has few Flat runs carries a mark that has not caught up with what it can do now, which is exactly the lag the thread claims.

The rule

region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs <= 6

Reads: region, race_type, is_handicap, month, distance_yards, hurdle_wins, flat_runs

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

37f404a77ce822b285334e3ccee3504ccb9a3d499f26e30832cb253ad6d85d49

Frozen

17 Oct, 03:29

Nothing in the definition can change after this.

Confounders

• Long-priced winner variance: the parent's BSP ROI (+40.7%) and SP ROI (-23.7%) far apart suggests profit comes from a few big winners. The 14/1-or-shorter refinement checks this.
• Big yards (Mullins, Elliott and others) aiming dual-purpose horses at summer festivals: a trainer-targeting effect rather than mark lag
• Course concentration at Galway, Killarney, Tralee and Listowel festivals; the live system's four-course slice was chosen after seeing the data. The no-Galway refinement checks this.
• Left-censoring: flat_runs and hurdle_wins only count from July/October 2020, so older horses early in the window may look lightly raced on the Flat or have uncounted hurdle wins. The low-Flat-runs group may then include exposed horses (diluting the effect) or be enriched for younger horses.
• Age and progression: horses lightly raced on the Flat with hurdle wins are often young, improving horses, so a general improver effect could pass for code-specific mark lag
• Parent rule failed on the lab window (0.986 on 744), so any lift here may be a slice of noise from a failed parent

Discovery result

pass broad rule (niche) · 17 Oct, 03:21

Bets

50

6 winners

A/E

1.26

95%: 0.46 to 2.75

SP return

-29.0%

level stakes

Betfair SP return

-8.4%

50 bets with a BSP

Strike rate 12.0%. p = 0.34. Same-race control: the other runners in these races had A/E 0.94, 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. 1 of 5 survived.

• perturbation: Flat-runs cut-off loosened to 8 or fewerpass

  60 bets · A/E 1.34 (0.58 to 2.63) · control 0.91 · 17 Oct, 03:29

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-270–

  after 2026-04-27601.34

• perturbation: Shorter staying trip (1m3f+, 2400 yards) and Flat runs 5 or fewerinfo

  44 bets · A/E 1.20 (0.39 to 2.79) · control 0.96 · 17 Oct, 03:29

  Breakdown

Sealed holdout (vault)

Bets

340

17 Oct, 03:29

A/E

0.99

p = 0.55

BH threshold then

0.0019

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: >= 2640 -> 2112 (-20%)pass

  54 bets · A/E 1.24 (0.46 to 2.71) · control 0.95 · 17 Oct, 03:30

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-270–

  after 2026-04-27541.24

• neighbour: >= 2640 -> 3168 (+20%)pass

  31 bets · A/E 1.30 (0.35 to 3.33) · control 0.92 · 17 Oct, 03:30

  Breakdown

Lineage

Agent runs

• designer v1ok · 38s · 17 Oct, 03:16

  Output (schema-checked)

  {
    "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs <= 6",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs <= 6 AND sp_decimal <= 15.0"
      },
      {
        "label": "very lightly raced on the Flat (dose-response: least mark information)",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs <= 3"
      },
      {
        "label": "excluding Galway (course-concentration check)",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs <= 6 AND course <> 'GALWAY'"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) The standard same-race control: other runners in the same Irish summer staying Flat handicaps should sit at or a little below A/E 1.00 (about 0.95–1.00), so any lift is not a property of the races. (2) A mechanism control: proven hurdle winners in the same race set who have MORE Flat runs (7+), whose Flat marks are well informed. The mechanism predicts these show clearly less lift than the main rule (around 1.00). If they lift just as much, the effect is a generic dual-purpose/hurdler effect, not mark lag from little Flat information.",
      "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_wins >= 1 AND flat_runs > 6"
    },
    "pass_criteria": {
      "min_bets": 200,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Long-priced winner variance: the parent's BSP ROI (+40.7%) and SP ROI (-23.7%) far apart suggests profit comes from a few big winners. The 14/1-or-shorter refinement checks this.",
      "Big yards (Mullins, Elliott and others) aiming dual-purpose horses at summer festivals: a trainer-targeting effect rather than mark lag",
      "Course concentration at Galway, Killarney, Tralee and Listowel festivals; the live system's four-course slice was chosen after seeing the data. The no-Galway refinement checks this.",
      "Left-censoring: flat_runs and hurdle_wins only count from July/October 2020, so older horses early in the window may look lightly raced on the Flat or have uncounted hurdle wins. The low-Flat-runs group may then include exposed horses (diluting the effect) or be enriched for younger horses.",
      "Age and progression: horses lightly raced on the Flat with hurdle wins are often young, improving horses, so a general improver effect could pass for code-specific mark lag",
      "Parent rule failed on the lab window (0.986 on 744), so any lift here may be a slice of noise from a failed parent"
    ],
    "mechanism": "An Irish Flat handicap mark comes from a few Flat runs, often run before the horse matured over hurdles. A horse that has since won over hurdles but has had 6 or fewer Flat runs carries a Flat mark that has not caught up with its current ability. Over 1m4f+ in summer, where stamina from hurdling counts, it should outrun its mark. The effect should be strongest where the Flat mark has the least information (fewest Flat runs) and fade as Flat runs pile up.",
    "market_miss": "The Flat form book shows modest Flat figures. Hurdle wins sit in a different code, and Flat punters and odds compilers weight them lightly. The price is anchored to the stale Flat mark and Flat form, not to the horse's jumps improvement.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.08 to 1.15 on 250+ bets, spread across Irish courses rather than confined to the festival tracks",
      "falsified_if": "A/E at or below 1.00 on 200+ bets; or the lift exists only at 16/1 or bigger (14/1-or-shorter refinement at or below 1.00); or the 7+ Flat runs control lifts as much as the main rule"
    },
    "notes": "I checked the columns: hurdle_wins (from 2020-10), flat_runs (from 2020-07), month and distance_yards are all pre_off with IRE coverage. hurdle_wins cannot leak today's result because today's race is Flat. I kept the rule sketch unchanged. The live system uses prior_hurdle_runs and the parent uses hurdle_runs; this rule uses neither. The forward bar is set a little above the floor at A/E 1.03, because the parent family has already failed once on the lab window."
  }

Timeline

1. Job 2660 finished

   17 Oct, 03:30 · engine:compute-1

2. hypothesis note H-cf4b198a

   17 Oct, 03:30 · engine:perturb

3. perturbation result for H-cf4b198a (neighbour: <= 6 -> 7 (+20%)): pass

   17 Oct, 03:30 · engine:perturb

4. perturbation result for H-cf4b198a (neighbour: <= 6 -> 5 (-20%)): pass

   17 Oct, 03:30 · engine:perturb

5. perturbation result for H-cf4b198a (neighbour: >= 1 -> 2 (+20%)): pass

   17 Oct, 03:30 · engine:perturb

6. perturbation result for H-cf4b198a (neighbour: >= 1 -> 0 (-20%)): pass

   17 Oct, 03:30 · engine:perturb

7. perturbation result for H-cf4b198a (neighbour: >= 2640 -> 3168 (+20%)): pass

   17 Oct, 03:30 · engine:perturb

8. perturbation result for H-cf4b198a (neighbour: >= 2640 -> 2112 (-20%)): pass

   17 Oct, 03:30 · engine:perturb

9. compute-1 took job 2660 (mutate, attempt 1)

   17 Oct, 03:30 · compute-1

10. Job 2678 finished

    17 Oct, 03:29 · engine:compute-1

11. H-cf4b198a Vault → KILLEDfailed the vault vault CONTRADICTED: A/E 0.99 on 340 bets, p 0.5454 against BH threshold 0.001923 over 52 vault tests

    17 Oct, 03:29 · engine:orchestrator

12. vault result for H-cf4b198a (vault query (contradicted)): fail

    17 Oct, 03:29 · engine:orchestrator

13. H-cf4b198a Frozen → FROZENvault vault query started

    17 Oct, 03:29 · engine:orchestrator

14. compute-1 took job 2678 (vault, attempt 1)

    17 Oct, 03:29 · compute-1

15. Job 2677 finished

    17 Oct, 03:29 · engine:compute-1

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

    17 Oct, 03:29 · engine:scheduler

17. H-cf4b198a Falsification → FROZEN survived falsification

    17 Oct, 03:29 · engine:falsification