Test-mechanism: big-field beatings where the horse still finished in the top half

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-23b2b028 (Sharpen: heavily beaten held-up handicappers, nothing taken out of the) pointed the predicted way but failed the bar: A/E 1.06 on 4329 bets (p 0.14), control 1.0. O-627a99c1
   • H-d97a030e (Rival: big-field beatings, not the heavy beating itself, carry the lif) passed discovery: A/E 1.19 on 2097 bets (p 0.022), same-race control 0.98. O-93b1b6e2
2. AskedIs the broad heavy-beating lift carried by runners at 16/1 or bigger (1.174 on 3156), against the family pattern that longshots sit at or below par? testing
3. MechanismIn big, strung-out jumps handicaps the margins pile up all the way down the field. A horse finishing in the top half can still be beaten 20 lengths without having run badly, so its margin overstates how much it was inferior.
4. Why the market misses itForm readers and price compilers lean on beaten lengths. A top-half finish beaten 20 lengths reads as a bad run when it was a fair one in a strung-out race.
5. Predictedpositive, A/E 1.20 to 1.35 on 400+ bets, at or above the big-field parent's 1.187
6. Falsified ifA/E below 1.10, or no higher than big-field heavily beaten runners that finished in the bottom half (last_position * 2 > last_field_size)
7. Rule sketchis_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 <= last_field_size

Mechanism and rationale

In big, strung-out jumps handicaps the margins pile up all the way down the field. A horse finishing in the top half can still be beaten 20 lengths without having run badly, so its margin overstates how much it was inferior.

The rule

is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 <= last_field_size

Reads: is_handicap, race_type, last_beaten_lengths, last_field_size, last_position

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

959584b9b1b2eb62dfb721f4578e2b73ae7b0ff2f56c7bf3b3714202d673a958

Frozen

not frozen

The definition can still change.

Confounders

• Testing ground at the last run stretching every margin (check last_going_group split)
• Race class of the last run (big fields in higher-class handicaps are deeper)
• Region mix (GB carries more 14+ runner jumps handicaps than IRE; the lift must show in both)
• Price band shift: top-half finishers are shorter in the betting, and the big-field lift sits under 16/1, so compare within price bands
• Hurdle vs chase mix (chase margins grow with jumping errors and fatigue)

Control

{
  "description": "Same-race other runners should be near par (about 0.99, as in the parent's control), showing the lift is not a property of the races. The mechanism-specific comparison is big-field heavily beaten runners that finished in the bottom half: the mechanism predicts the top-half group beats them clearly. If the bottom-half group scores the same or higher, the lift is about big-field heavy beatings in general, not misleading margins for respectable finishes.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 > last_field_size"
}

Designer

designer@v1

Direction

positive

Discovery result

pass broad rule · 17 Oct, 03:30

Bets

642

64 winners

A/E

1.29

95%: 0.99 to 1.64

SP return

-19.3%

level stakes

Betfair SP return

+5.7%

642 bets with a BSP

Strike rate 10.0%. p = 0.030. Same-race control: the other runners in these races had A/E 0.98, 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. 3 of 5 survived.

• perturbation: Perturb looser: beaten 16+ lengths, last field 12+pass

  1,254 bets · A/E 1.20 (1.00 to 1.43) · control 0.98 · 17 Oct, 03:30

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-277121.15

  after 2026-04-275421.30

• perturbation: Perturb tighter: beaten 24+ lengths, last field 16+pass

  291 bets · A/E 1.41 (0.95 to 2.01) · control 0.98 · 17 Oct, 03:30

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 · 22s · 17 Oct, 03:27

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 <= last_field_size",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 <= last_field_size AND sp_decimal <= 15"
      },
      {
        "label": "Top third finish (stronger form of the mechanism)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 3 <= last_field_size"
      },
      {
        "label": "Last run also a handicap (same strung-out race shape)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 <= last_field_size AND last_is_handicap = 1"
      }
    ],
    "control": {
      "description": "Same-race other runners should be near par (about 0.99, as in the parent's control), showing the lift is not a property of the races. The mechanism-specific comparison is big-field heavily beaten runners that finished in the bottom half: the mechanism predicts the top-half group beats them clearly. If the bottom-half group scores the same or higher, the lift is about big-field heavy beatings in general, not misleading margins for respectable finishes.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND last_field_size >= 14 AND last_position IS NOT NULL AND last_position * 2 > last_field_size"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Testing ground at the last run stretching every margin (check last_going_group split)",
      "Race class of the last run (big fields in higher-class handicaps are deeper)",
      "Region mix (GB carries more 14+ runner jumps handicaps than IRE; the lift must show in both)",
      "Price band shift: top-half finishers are shorter in the betting, and the big-field lift sits under 16/1, so compare within price bands",
      "Hurdle vs chase mix (chase margins grow with jumping errors and fatigue)"
    ],
    "mechanism": "In big, strung-out jumps handicaps (14+ runners) the margins pile up all the way down the field. A horse that finishes in the top half can still be beaten 20 lengths or more without having run badly. Its margin overstates how inferior it was, and it may still be well handicapped next time.",
    "market_miss": "Form readers and price compilers lean on raw beaten lengths. A top-half finish beaten 20+ lengths in a big field reads as a bad run when it was a fair one in a strung-out race, so the horse goes off at a bigger price than its form deserves.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.20 to 1.35 on 400+ bets, at or above the big-field parent's 1.187",
      "falsified_if": "A/E below 1.10, or no higher than the bottom-half control (big-field runners beaten 20+ lengths with last_position * 2 > last_field_size)"
    },
    "notes": "The rule sketch is kept as written: every column is pre-off (last_* columns describe the previous run), the rule does not read today's result, and the top-half condition is the mechanism itself. Non-finishers are excluded by last_position IS NOT NULL. The 14/1 refinement follows the finding that the big-field lift sits under 16/1. Pass A/E is tightened to 1.10 to match the falsification line."
  }

Timeline

1. Job 2698 finished

   17 Oct, 03:30 · engine:compute-1

2. H-9ccf4a8d Falsification → KILLED falsified: 2 of 5 attacks broke it

   17 Oct, 03:30 · engine:falsification

3. falsification result for H-9ccf4a8d (partition: Partition: hurdles only (strung out without jumping errors)): pass

   17 Oct, 03:30 · engine:evaluator

4. compute-1 took job 2698 (attack, attempt 1)

   17 Oct, 03:30 · compute-1

5. Job 2697 finished

   17 Oct, 03:30 · engine:compute-1

6. falsification result for H-9ccf4a8d (control: Control: big-field top-half finishers beaten under 20 lengths): fail

   17 Oct, 03:30 · engine:evaluator

7. compute-1 took job 2697 (attack, attempt 1)

   17 Oct, 03:30 · compute-1

8. Job 2696 finished

   17 Oct, 03:30 · engine:compute-1

9. falsification result for H-9ccf4a8d (control: Control: top-half finish beaten 20+ lengths in small fields (8-13)): fail

   17 Oct, 03:30 · engine:evaluator

10. compute-1 took job 2696 (attack, attempt 1)

    17 Oct, 03:30 · compute-1

11. Job 2695 finished

    17 Oct, 03:30 · engine:compute-1

12. falsification result for H-9ccf4a8d (perturbation: Perturb tighter: beaten 24+ lengths, last field 16+): pass

    17 Oct, 03:30 · engine:evaluator

13. compute-1 took job 2695 (attack, attempt 1)

    17 Oct, 03:30 · compute-1

14. Job 2694 finished

    17 Oct, 03:30 · engine:compute-1

15. falsification result for H-9ccf4a8d (perturbation: Perturb looser: beaten 16+ lengths, last field 12+): pass

    17 Oct, 03:30 · engine:evaluator

16. compute-1 took job 2694 (attack, attempt 1)

    17 Oct, 03:30 · compute-1

17. Job 2693 finished

    17 Oct, 03:30 · engine:ai-1

18. Queued attack on the compute lane: attack: Partition: hurdles only (strung out without jumping errors)

    17 Oct, 03:30 · engine:scheduler

19. Queued attack on the compute lane: attack: Control: big-field top-half finishers beaten under 20 lengths

    17 Oct, 03:30 · engine:scheduler