Test-mechanism: straight from hurdles into an Irish summer staying Flat handicap

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-6234a360 (Rival: lightly raced Flat staying handicappers beat their prices wheth) pointed the predicted way but failed the bar: A/E 1.05 on 215 bets (p 0.449), control 0.98. O-09786346
   • H-cf4b198a (Sharpen: proven hurdlers lightly raced on the Flat in Irish summer sta) survived falsification: 1 of 1 attacks left it standing; it is frozen for its one vault query. O-5547c5b3
   • H-cf4b198a (Sharpen: proven hurdlers lightly raced on the Flat in Irish summer sta) failed the vault (contradicted): A/E 0.99 on 340 bets. O-af584569
2. AskedIn Irish summer staying Flat handicaps, does a runner whose last run was over hurdles beat its price, because the Flat mark has not yet caught up with that most recent hurdle form? answered
3. MechanismIf the Flat mark lags behind hurdle form, the lag is largest just after a hurdle run, because the Flat handicapper has not yet reacted to that form. The horse is also race-fit from jumping. Career counts of Flat and hurdle runs measure this lag only loosely, which may be why the count-based cut failed.
4. Why the market misses itThe Flat form lines on show are old and often moderate. The recent hurdle run is in a different code, and punters reading Flat form can give it little weight.
5. Predictedpositive, A/E 1.06 to 1.12 on 300+ bets, with the lift not confined to 16/1 or bigger
6. Falsified if

Mechanism and rationale

If the Flat mark lags behind hurdle form, the lag is largest just after a hurdle run, because the Flat handicapper has not yet reacted to that form. The horse is also race-fit from jumping. Career counts of Flat and hurdle runs measure this lag only loosely, which may be why the count-based cut failed.

The rule

region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND last_race_type = 'Hurdle'

Reads: region, race_type, is_handicap, month, distance_yards, last_race_type

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

48267c033b51ccc1144bcb437bed000cd87ccca4be600266144e7d1d62c81808

Frozen

not frozen

The definition can still change.

Confounders

• festival meetings (Galway, Listowel, Killarney etc.) concentrate these runners; check the lift is not only at those courses
• big jumps yards switching codes: trainer skill rather than mark lag
• long-price variance from a few winners at 16/1+ (earlier family rules lived or died on this)
• field size: big summer handicaps inflate long-price variance
• same-race control near 1.1 in this family: beating 1.00 is not enough if race-mates do as well
• hurdle-last runners may be the same horses as the count-based dual-purpose set, so overlap with H228 must be read, not counted as independent evidence

Control

{
  "description": "Same-race runners (lab default): if the effect is about the horse rather than the race, race-mates should score near the 1.08-1.10 seen in earlier family controls and the selection must beat them, not just beat 1.00. Second control isolates recency from being a hurdler: dual-purpose horses (3+ hurdle runs) in the same race type whose last run was NOT over hurdles. The mechanism predicts the hurdle-last group beats this group; if both score alike, the edge (if any) is a hurdler property, not a lag-after-hurdle-run effect.",
  "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_runs >= 3 AND last_race_type <> 'Hurdle'"
}

Discovery result

fail broad rule · 17 Oct, 03:35

Bets

130

7 winners

A/E

0.86

95%: 0.35 to 1.78

SP return

-54.0%

level stakes

Betfair SP return

-45.6%

130 bets with a BSP

Strike rate 5.4%. p = 0.70. Same-race control: the other runners in these races had A/E 1.02, so the selection did not beat its own races. Zone: discovery. Evaluator eval-1.0, data 6a4e98be18e83933

Falsification

No attacks yet. Only promising ideas are attacked.

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.

Lineage

Agent runs

• designer v1ok · 41s · 17 Oct, 03:33

  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 last_race_type = 'Hurdle'",
    "refinements": [
      {
        "label": "shorter than 16/1 (lift must not be a long-price artefact)",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND last_race_type = 'Hurdle' AND sp_decimal < 17"
      },
      {
        "label": "hurdle run within 35 days (lag largest just after the hurdle run)",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND last_race_type = 'Hurdle' AND days_since_last_run <= 35"
      }
    ],
    "control": {
      "description": "Same-race runners (lab default): if the effect is about the horse rather than the race, race-mates should score near the 1.08-1.10 seen in earlier family controls and the selection must beat them, not just beat 1.00. Second control isolates recency from being a hurdler: dual-purpose horses (3+ hurdle runs) in the same race type whose last run was NOT over hurdles. The mechanism predicts the hurdle-last group beats this group; if both score alike, the edge (if any) is a hurdler property, not a lag-after-hurdle-run effect.",
      "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_runs >= 3 AND last_race_type <> 'Hurdle'"
    },
    "pass_criteria": {
      "min_bets": 250,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 120,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "festival meetings (Galway, Listowel, Killarney etc.) concentrate these runners; check the lift is not only at those courses",
      "big jumps yards switching codes: trainer skill rather than mark lag",
      "long-price variance from a few winners at 16/1+ (earlier family rules lived or died on this)",
      "field size: big summer handicaps inflate long-price variance",
      "same-race control near 1.1 in this family: beating 1.00 is not enough if race-mates do as well",
      "hurdle-last runners may be the same horses as the count-based dual-purpose set, so overlap with H228 must be read, not counted as independent evidence"
    ],
    "mechanism": "A horse's Flat handicap mark is set from Flat form and reacts slowly to jumps form. Straight after a hurdle run the gap between current ability/fitness and the Flat mark is largest: the Flat handicapper has not reassessed, and the horse is race-fit from jumping. Career run counts (hurdle_runs, flat_runs) blur this timing, which may explain why count-based cuts failed in the vault; last_race_type = 'Hurdle' measures the timing directly.",
    "market_miss": "The Flat form figures on show are old and often moderate. The latest run is in a different code, and Flat punters give it little weight, so the price reflects stale Flat form rather than current fitness and ability.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.06 to 1.12 on 300+ bets, with the lift present below 16/1 and above the same-race control",
      "falsified_if": "A/E at or below 1.00 on 250+ bets; or a lift only at 16/1 or bigger (sub-16/1 A/E at or below 1.00); or A/E not above the same-race control and the non-hurdle-last dual-purpose control"
    },
    "notes": "Rule sketch kept unchanged: all columns are pre_off and available for IRE from 2020-07-01 (last_race_type, month, distance_yards checked). min_bets raised to 250 to match the falsification threshold, because earlier family results on 50-100 bets did not hold up in the vault. sp_decimal is the price at the off; the refinement uses it only as a price band, as earlier family price splits did. No course refinement, because the festival-course list was not fully confirmed and earlier course and field-size slices were killed."
  }

Timeline

1. Job 2707 finished

   17 Oct, 03:35 · engine:compute-1

2. H-4e463106 Discovery → KILLED fails: 130 bets, needs 250; A/E 0.862, needs 1.05; p 0.7006, needs 0.05 or lower; periods [0.86], each needs 1.0; A/E 0.862 against same-race control 1.021, needs 0.03 above (not a niche pass: A/E 0.862 below 1.05; p 0.700587397225847 above 0.5; does not beat the same-race control 1.021; a period points the other way [0.86])

   17 Oct, 03:35 · engine:evaluator

3. discovery result for H-4e463106 (broad rule): fail

   17 Oct, 03:35 · engine:evaluator

4. H-4e463106 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 03:35 · engine:orchestrator

5. compute-1 took job 2707 (evaluate, attempt 1)

   17 Oct, 03:35 · compute-1

6. Job 2706 finished

   17 Oct, 03:34 · engine:ai-1

7. Queued evaluate on the compute lane: discovery run of the broad rule

   17 Oct, 03:34 · engine:scheduler

8. H-4e463106 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 03:34 · agent:designer@v1

9. ai-1 took job 2706 (design, attempt 1)

   17 Oct, 03:33 · ai-1

10. Queued design on the ai lane: write the pre-spec

    17 Oct, 03:33 · engine:scheduler

11. New idea H-4e463106: Test-mechanism: straight from hurdles into an Irish summer staying Flat handicap (test-mechanism)

    17 Oct, 03:33 · agent:investigator@v1

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Killed.