Rival: the big-field lift is a class drop out of valuable 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
   • H-91a89a7c (Rival: the big-field lift is a drop into a smaller race today, not a m) passed discovery but was falsified: driven by one price band (up to 7/1): the rest score 0.88. O-4f7f9b7a
   • H-9ccf4a8d (Test-mechanism: big-field beatings where the horse still finished in t) passed discovery: A/E 1.29 on 642 bets (p 0.03), same-race control 0.98. O-6a8bcbc8
   • H-91a89a7c (Rival: the big-field lift is a drop into a smaller race today, not a m) passed discovery as a niche: A/E 1.15 on 526 bets (p 0.154), same-race control 0.99. O-ac67c459
2. AskedIs the big-field lift really a drop from a higher-class or festival handicap, rather than inflated beaten lengths? testing
3. MechanismBig fields cluster in valuable, higher-class handicaps. A horse well beaten there may simply face weaker rivals today.
4. Why the market misses itIf the class drop explains it, the market misses the drop rather than misreading the lengths. Class drops are visible, though, so this rival is expected to score no higher than the parent.
5. Predictedpositive, A/E 1.20 to 1.35 on 300+ bets if class carries the lift
6. Falsified ifA/E at or below the parent's 1.187, which would rule out class as the carrier

Mechanism and rationale

Big fields cluster in valuable, higher-class handicaps. A horse well beaten there may simply face weaker rivals today.

The rule

region = 'GB' AND is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND last_field_size IS NOT NULL AND last_field_size >= 14 AND last_race_class IS NOT NULL AND race_class IS NOT NULL AND last_race_class < race_class

Reads: region, is_handicap, race_type, last_beaten_lengths, last_field_size, last_race_class, race_class

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

3c47553a3f83f3c6b7d71f6d33363819ed3c0f8525c0aaa43e136a22b63b290d

Frozen

17 Oct, 03:42

Nothing in the definition can change after this.

Confounders

• GB only: IRE race_class only exists from 2026-07, so the comparison has to be with the parent's GB figure (1.215), not the pooled 1.187
• the 16/1-or-bigger band, which sits below par in every big-field cut and could drag the score down
• festival and big-meeting handicaps (Cheltenham, Aintree and so on) are both the source of big fields and of class drops, so the class drop may just be a stand-in for 'last ran at a festival'
• race_class is an ordinal label that tracks prize money and rating band; a nominal class drop can still mean the same rating band in a handicap
• seasonality: class drops cluster after the spring festivals

Control

{
  "description": "The key comparison is the parent's GB big-field heavy beatings that do NOT drop in class (same or higher class today). If class carries the lift, the drop group should clearly beat this control and the control should sit near 1.0. If the control scores about the same as the drop group (both around the parent's GB 1.215), class is not the carrier. Same-race other runners are expected near 1.0 or slightly below, as usual.",
  "rule": "region = 'GB' AND is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND last_field_size IS NOT NULL AND last_field_size >= 14 AND last_race_class IS NOT NULL AND race_class IS NOT NULL AND last_race_class >= race_class"
}

Discovery result

pass broad rule (niche) · 17 Oct, 03:35

Bets

176

25 winners

A/E

1.25

95%: 0.81 to 1.84

SP return

-6.0%

level stakes

Betfair SP return

+15.7%

176 bets with a BSP

Strike rate 14.2%. p = 0.16. 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. 4 of 6 survived.

• perturbation: Loosen thresholds: beaten 16+ lengths, last field 12+pass

  395 bets · A/E 1.27 (0.96 to 1.65) · control 0.97 · 17 Oct, 03:42

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-273141.18

  after 2026-04-27811.58

• perturbation: Tighten thresholds: beaten 24+ lengths, last field 16+fail

  65 bets · A/E 0.78 (0.29 to 1.71) · control 1.04 · 17 Oct, 03:42

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

Forward record

• forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}info

  0 bets · · 17 Oct, 03:42

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: >= 20 -> 16 (-20%)pass

  206 bets · A/E 1.15 (0.76 to 1.66) · control 0.99 · 17 Oct, 03:35

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271631.21

  after 2026-04-27430.92

• neighbour: >= 20 -> 24 (+20%)pass

  145 bets · A/E 1.11 (0.66 to 1.76) · control 1.00 · 17 Oct, 03:35

  Breakdown

Lineage

Agent runs

• designer v1ok · 35s · 17 Oct, 03:32

  Output (schema-checked)

  {
    "rule": "region = 'GB' AND is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND last_field_size IS NOT NULL AND last_field_size >= 14 AND last_race_class IS NOT NULL AND race_class IS NOT NULL AND last_race_class < race_class",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "region = 'GB' AND is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND last_field_size IS NOT NULL AND last_field_size >= 14 AND last_race_class IS NOT NULL AND race_class IS NOT NULL AND last_race_class < race_class AND sp_decimal < 16"
      },
      {
        "label": "dropping out of a valuable (class 1-3) race",
        "rule": "region = 'GB' AND is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND last_field_size IS NOT NULL AND last_field_size >= 14 AND last_race_class IS NOT NULL AND race_class IS NOT NULL AND last_race_class < race_class AND last_race_class <= 3"
      }
    ],
    "control": {
      "description": "The key comparison is the parent's GB big-field heavy beatings that do NOT drop in class (same or higher class today). If class carries the lift, the drop group should clearly beat this control and the control should sit near 1.0. If the control scores about the same as the drop group (both around the parent's GB 1.215), class is not the carrier. Same-race other runners are expected near 1.0 or slightly below, as usual.",
      "rule": "region = 'GB' AND is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND last_field_size IS NOT NULL AND last_field_size >= 14 AND last_race_class IS NOT NULL AND race_class IS NOT NULL AND last_race_class >= race_class"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "GB only: IRE race_class only exists from 2026-07, so the comparison has to be with the parent's GB figure (1.215), not the pooled 1.187",
      "the 16/1-or-bigger band, which sits below par in every big-field cut and could drag the score down",
      "festival and big-meeting handicaps (Cheltenham, Aintree and so on) are both the source of big fields and of class drops, so the class drop may just be a stand-in for 'last ran at a festival'",
      "race_class is an ordinal label that tracks prize money and rating band; a nominal class drop can still mean the same rating band in a handicap",
      "seasonality: class drops cluster after the spring festivals"
    ],
    "mechanism": "Big fields of 14 or more cluster in valuable, higher-class handicaps. A horse beaten 20 lengths or more in one of them may have run to a level that would be competitive in a weaker race. When it drops in class today it faces weaker rivals, and the heavy beating overstates how poor the horse is.",
    "market_miss": "If the class drop carries the lift, the market prices the raw beaten lengths and underrates how much easier today's race is. Class drops are plain to see, though, so this rival is not expected to beat the parent. Its main job is to show whether the lift sits in the drop group or the no-drop group.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.20 to 1.35 on 300+ GB bets if class carries the lift, with the no-drop control near 1.0",
      "falsified_if": "A/E at or below the parent's GB big-field score (about 1.215), or the no-drop control scoring the same or better, which would rule out class as the carrier of the big-field lift"
    },
    "notes": "The rule is restricted to region = 'GB' explicitly because IRE race_class is NULL before 2026-07, so IRE would drop out silently anyway. I added null guards. class 1 = top, so last_race_class < race_class means a drop in class. The parent to beat is the GB slice of the big-field rule (1.215), not 1.187. Refinements follow the reasoning: the lift lives under about 14/1, and the mechanism is about leaving valuable races."
  }

Timeline

1. Job 2743 finished

   17 Oct, 03:42 · engine:compute-1

2. forward result for H-0bdb9150 (forward vs bar {'min_bets': 100, 'min_ae': 1.0, 'max_p': 0.2}): info

   17 Oct, 03:42 · engine:evaluator

3. compute-1 took job 2743 (forward_check, attempt 1)

   17 Oct, 03:42 · compute-1

4. Queued forward_check on the compute lane: daily forward check of a shadow system

   17 Oct, 03:42 · engine:scheduler

5. Job 2739 finished

   17 Oct, 03:42 · engine:compute-1

6. H-0bdb9150 Frozen → SHADOW vault skipped: last_beaten_lengths do not exist before the discovery zone; track it forward instead

   17 Oct, 03:42 · engine:orchestrator

7. compute-1 took job 2739 (vault, attempt 1)

   17 Oct, 03:42 · compute-1

8. Job 2732 finished

   17 Oct, 03:42 · engine:compute-1

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

   17 Oct, 03:42 · engine:scheduler

10. H-0bdb9150 Falsification → FROZEN survived falsification

    17 Oct, 03:42 · engine:falsification

11. falsification result for H-0bdb9150 (price_band: Outside the 7/1 band: 8/1 or bigger only): pass

    17 Oct, 03:42 · engine:evaluator

12. compute-1 took job 2732 (attack, attempt 1)

    17 Oct, 03:42 · compute-1

13. Job 2731 finished

    17 Oct, 03:42 · engine:compute-1

14. falsification result for H-0bdb9150 (control: Control: heavy beating plus class drop, but last field under 14): pass

    17 Oct, 03:42 · engine:evaluator

15. compute-1 took job 2731 (attack, attempt 1)

    17 Oct, 03:42 · compute-1

16. Job 2730 finished

    17 Oct, 03:42 · engine:compute-1

17. falsification result for H-0bdb9150 (partition: Not from a premier race: last race class 2 or lower (excludes class 1 festival handicaps)): pass

    17 Oct, 03:42 · engine:evaluator

18. compute-1 took job 2730 (attack, attempt 1)

    17 Oct, 03:42 · compute-1

19. Job 2729 finished

    17 Oct, 03:42 · engine:compute-1

20. falsification result for H-0bdb9150 (partition: Dose-response: drop of two or more classes): info

    17 Oct, 03:42 · engine:evaluator