Sharpen: heavily beaten held-up handicappers, nothing taken out of them

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-1a646ca5 (Generalise: jumps handicappers beaten 20 lengths or more last time bea) passed discovery: A/E 1.08 on 6743 bets (p 0.041), same-race control 0.99. O-58c96d7b
   • H-1a646ca5 (Generalise: jumps handicappers beaten 20 lengths or more last time bea) survived falsification: 6 of 6 attacks left it standing; it is frozen for its one vault query. O-f18ec549
2. AskedAre big beaten distances in large fields less informative, so that the lift concentrates where last_field_size was big? answered
3. MechanismA held-up horse that never got into the race is often eased down. The beaten distance overstates its effort, and the run took little out of it.
4. Why the market misses itThe form book records the full beaten distance, and the market treats 25 lengths eased as 25 lengths of inability.
5. Predictedpositive, A/E 1.12 to 1.22 on 1500+ bets
6. Falsified ifA/E at or below the broad rule's 1.08, or below 1.05 against the same-race control
7. Rule sketchis_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IN ('held_up', 'midfield')
8. question idQ-35af2794

The question's answer so far: Largely yes. Heavy beatings in last fields of 14 or more scored 1.187 on 2097, fields of 11 to 13 scored 1.116 on 1509, and fields of 10 or fewer scored 1.022 on 3281. The lift rises with field size and held in both countries and both periods.

Mechanism and rationale

A held-up horse that never got into the race is often eased down. The beaten distance overstates its effort, and the run took little out of it.

The rule

is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IN ('held_up', 'midfield')

Reads: is_handicap, race_type, last_beaten_lengths, last_run_style

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

bd31ec82495339e27f96dae65efc64d3f8ee0209f6316999795c0b6d5603c302

Frozen

not frozen

The definition can still change.

Confounders

• Field size last time: big fields make 20+ lengths easier to reach by chance and push more runners to the back
• Going last time: heavy ground stretches beaten distances, and tired horses are eased
• Region: earlier narrow cuts in this family were GB-only, so the lift must appear in both GB and IRE
• Price band: the broad rule is above par at 16/1+ (1.174), so check that the lift is not just the longshot part of the broad rule
• Overlap with the parent rule: held-up/midfield may make up most of the 6743 broad-rule bets, which would make this a relabel rather than a sharpening
• Race code: hurdle against chase mix, since chase heavy beatings can come from jumping errors rather than easing
• Mark relief: heavily beaten handicappers drop in the weights, which affects both run-style groups

Control

{
  "description": "Same-race control: the other runners in these races should sit near 0.99, as in the broad heavy-beating finding. The sharpening claim also needs a second comparison: heavily beaten jumps handicappers that raced prominently or led last time. Under the mechanism, those horses were in the race and probably had a hard race, so they should score at or below the broad rule's 1.08. The held-up/midfield group must beat them clearly.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IS NOT NULL AND last_run_style NOT IN ('held_up', 'midfield')"
}

Discovery result

fail broad rule · 17 Oct, 03:18

Bets

4,329

309 winners

A/E

1.06

95%: 0.95 to 1.19

SP return

-23.0%

level stakes

Betfair SP return

+7.5%

4,326 bets with a BSP

Strike rate 7.1%. p = 0.14. Same-race control: the other runners in these races had A/E 1.00, so the selection 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 · 36s · 17 Oct, 03:13

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IN ('held_up', 'midfield')",
    "refinements": [
      {
        "label": "eased last time (mechanism directly)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IN ('held_up', 'midfield') AND last_eased = 1"
      },
      {
        "label": "never involved last time (never got into the race)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IN ('held_up', 'midfield') AND last_never_involved = 1"
      }
    ],
    "control": {
      "description": "Same-race control: the other runners in these races should sit near 0.99, as in the broad heavy-beating finding. The sharpening claim also needs a second comparison: heavily beaten jumps handicappers that raced prominently or led last time. Under the mechanism, those horses were in the race and probably had a hard race, so they should score at or below the broad rule's 1.08. The held-up/midfield group must beat them clearly.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_run_style IS NOT NULL AND last_run_style NOT IN ('held_up', 'midfield')"
    },
    "pass_criteria": {
      "min_bets": 400,
      "min_ae": 1.1,
      "max_p": 0.02
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Field size last time: big fields make 20+ lengths easier to reach by chance and push more runners to the back",
      "Going last time: heavy ground stretches beaten distances, and tired horses are eased",
      "Region: earlier narrow cuts in this family were GB-only, so the lift must appear in both GB and IRE",
      "Price band: the broad rule is above par at 16/1+ (1.174), so check that the lift is not just the longshot part of the broad rule",
      "Overlap with the parent rule: held-up/midfield may make up most of the 6743 broad-rule bets, which would make this a relabel rather than a sharpening",
      "Race code: hurdle against chase mix, since chase heavy beatings can come from jumping errors rather than easing",
      "Mark relief: heavily beaten handicappers drop in the weights, which affects both run-style groups"
    ],
    "mechanism": "A horse held up or in midfield that never got into contention is often eased once its chance has gone. Its official beaten distance (20+ lengths) then overstates how little ability it showed. Because it was not ridden out, the run also took little out of it. Next time in a handicap, often off a lower mark, it runs closer to its true level than the beaten distance suggests.",
    "market_miss": "The form book and ratings record the full beaten distance. Punters treat 25 lengths eased as 25 lengths of inability and do not separate a horse that was looked after from a prominent horse that was run into the ground. So the price lengthens too far for the held-up type.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.12 to 1.22 on 1500+ bets, clearly above the broad heavy-beating rule's 1.08 and above the prominent/led control",
      "falsified_if": "Falsified if: A/E is at or below 1.08 (the broad rule); or the lift over the same-race control is less than 1.05; or the prominent/led heavy-beating control scores as high or higher; or the lift appears in only one country."
    },
    "notes": "Every column is pre-off. last_beaten_lengths and last_run_style are available from 2025-11 in both countries. The draft rule is kept as it stands. Refinements test the mechanism directly (eased, never involved). No price cap is used, because the family finding shows the broad rule is above par at 16/1+. Pass bar is stricter than the floor because the claim is a sharpening and has to clear the parent's 1.08."
  }

Timeline

1. Job 2613 finished

   17 Oct, 03:18 · engine:compute-1

2. H-23b2b028 Discovery → KILLED fails: A/E 1.065, needs 1.1; p 0.14, needs 0.02 or lower (not a niche pass: A/E 1.065 below 1.1)

   17 Oct, 03:18 · engine:evaluator

3. discovery result for H-23b2b028 (broad rule): fail

   17 Oct, 03:18 · engine:evaluator

4. H-23b2b028 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 03:18 · engine:orchestrator

5. compute-1 took job 2613 (evaluate, attempt 1)

   17 Oct, 03:18 · compute-1

6. Job 2610 finished

   17 Oct, 03:14 · engine:ai-1

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

   17 Oct, 03:14 · engine:scheduler

8. H-23b2b028 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 03:14 · agent:designer@v1

9. ai-1 took job 2610 (design, attempt 1)

   17 Oct, 03:13 · ai-1

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

    17 Oct, 03:13 · engine:scheduler

11. New idea H-23b2b028: Sharpen: heavily beaten held-up handicappers, nothing taken out of them (sharpen)

    17 Oct, 03:13 · 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.