Test-mechanism: the heavy beating misleads most when ground or trip changed since

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-332b49e9 (Rival: the lift is a one-bad-run rebound after close recent form, need) passed discovery: A/E 1.23 on 1600 bets (p 0.004), same-race control 0.98. O-10a4d6e5
2. AskedIs the one-bad-run rebound strongest when the heavy beating came on different ground or at a different trip from today, as a temporary-cause mechanism predicts? answered
3. MechanismIf the 20-length beating had a temporary cause, the commonest causes are unsuitable ground or trip. A horse back on different going from the beating faces conditions closer to those of its close run, so the beating says least about today.
4. Why the market misses itThe form line shows the margin prominently. The going of each run is shown but is rarely used to discount a beating, so the latest heavy defeat drags the price whatever the conditions.
5. Predictedpositive, A/E 1.25 to 1.45 on 500+ bets, above the same-going complement
6. Falsified ifA/E at or below 1.15, or no higher than the same rule with last_going_group = going_group
7. Rule sketchis_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group <> going_group

Mechanism and rationale

If the 20-length beating had a temporary cause, the commonest causes are unsuitable ground or trip. A horse back on different going from the beating faces conditions closer to those of its close run, so the beating says least about today.

The rule

is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group <> going_group

Reads: is_handicap, race_type, last_beaten_lengths, best_last3_beaten, last_going_group, going_group

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

ec7c6c1f630f8588761b5ae6007aeb5dc979dd71ebcdf1dc5fe2476ca2379d0e

Frozen

17 Oct, 04:22

Nothing in the definition can change after this.

Confounders

• Season change shifts the going and the yard's form together, so a going change may stand in for a stable coming into form
• Going groups are coarse; a change across a group boundary may be small and one inside a group may be large
• best_last3_beaten exists only from November 2025, so the window is short and one season
• A few longshot winners can drive the A/E; the 16/1-or-shorter refinement checks this
• Overlaps the known close-finish rebound (1.23 on 1600); the result must beat the same-going complement, not just the floor
• Recent winners carry much of the GB lift in this family, so the close run may be a win rather than a near-miss

Control

{
  "description": "The key comparison is the same-going complement: the identical rule but with last_going_group = going_group. The mechanism says the rebound is concentrated where the ground changed, so the changed-going group should beat the same-going group clearly (the family baseline for beaten-20+ close finishers is about 1.23 on 1600). If both score alike, going change carries nothing and the lift belongs to the close-finish rebound already known. Other runners in the same races should be about 1.0; a high reading there would point to a race-level effect (e.g. going-change meetings with odd results) rather than the horse.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group = going_group"
}

Discovery result

pass broad rule · 17 Oct, 04:14

Bets

917

105 winners

A/E

1.26

95%: 1.03 to 1.53

SP return

+8.2%

level stakes

Betfair SP return

+54.9%

914 bets with a BSP

Strike rate 11.5%. p = 0.012. 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. 6 of 6 survived.

• control: Same-going complement: identical rule with last_going_group = going_grouppass

  682 bets · A/E 1.20 (0.94 to 1.50) · control 0.99 · 17 Oct, 04:22

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272701.10

  after 2026-04-274121.26

• price_band: 16/1 or shorter only: is the lift a handful of longshot winners?pass

  650 bets · A/E 1.21 (0.97 to 1.48) · control 0.98 · 17 Oct, 04:22

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': 1.05, 'max_p': 0.2}info

  0 bets · · 17 Oct, 04:24

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: >= 20 -> 16 (-20%)pass

  1,182 bets · A/E 1.19 (0.99 to 1.41) · control 0.98 · 17 Oct, 04:15

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-277431.19

  after 2026-04-274391.19

• neighbour: >= 20 -> 24 (+20%)pass

  735 bets · A/E 1.25 (0.99 to 1.56) · control 0.98 · 17 Oct, 04:15

  Breakdown

Lineage

Agent runs

• designer v1ok · 41s · 17 Oct, 04:13

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group <> going_group",
    "refinements": [
      {
        "label": "going and trip both changed (10%+)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group <> going_group AND last_distance_yards IS NOT NULL AND distance_yards IS NOT NULL AND (distance_yards >= 1.1 * last_distance_yards OR distance_yards <= 0.9 * last_distance_yards)"
      },
      {
        "label": "16/1 or shorter, to check the lift is not a few longshot winners",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group <> going_group AND sp_decimal <= 17"
      },
      {
        "label": "excluding recent winners (close run was a defeat, not a win)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten > 0 AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group <> going_group"
      }
    ],
    "control": {
      "description": "The key comparison is the same-going complement: the identical rule but with last_going_group = going_group. The mechanism says the rebound is concentrated where the ground changed, so the changed-going group should beat the same-going group clearly (the family baseline for beaten-20+ close finishers is about 1.23 on 1600). If both score alike, going change carries nothing and the lift belongs to the close-finish rebound already known. Other runners in the same races should be about 1.0; a high reading there would point to a race-level effect (e.g. going-change meetings with odd results) rather than the horse.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten <= 6 AND last_going_group IS NOT NULL AND going_group IS NOT NULL AND last_going_group = going_group"
    },
    "pass_criteria": {
      "min_bets": 200,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Season change shifts the going and the yard's form together, so a going change may stand in for a stable coming into form",
      "Going groups are coarse; a change across a group boundary may be small and one inside a group may be large",
      "best_last3_beaten exists only from November 2025, so the window is short and one season",
      "A few longshot winners can drive the A/E; the 16/1-or-shorter refinement checks this",
      "Overlaps the known close-finish rebound (1.23 on 1600); the result must beat the same-going complement, not just the floor",
      "Recent winners carry much of the GB lift in this family, so the close run may be a win rather than a near-miss"
    ],
    "mechanism": "A 20-length beating that sits next to a close finish in the last three runs is likely to have had a temporary cause, and the commonest are unsuitable ground or trip. When today's going group differs from the going of the beating, the horse is no longer facing the conditions that beat it, so that beating tells us least about today. The rebound seen in close-finishing heavily beaten jumps handicappers should therefore be concentrated where the going has changed.",
    "market_miss": "The form line shows the latest margin prominently, and the price is set largely off that. The going of each run is shown but is rarely used to discount a beating, so the heavy defeat drags the price the same whether or not today's ground differs from the ground it came on.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.25 to 1.45 on 500+ bets, clearly above the same-going complement",
      "falsified_if": "A/E at or below 1.15, or no higher than the same-going control rule (last_going_group = going_group) on comparable bets"
    },
    "notes": "I kept the rule sketch and added an explicit best_last3_beaten IS NOT NULL. The mechanism names trip as well as going, so trip goes in as a refinement and not into the broad rule. Pass criteria are stricter than the floor (200 bets, A/E 1.10) because this overlaps a known family effect. The real test is the gap against the same-going control."
  }

Timeline

1. Job 2899 finished

   17 Oct, 04:24 · engine:compute-1

2. forward result for H-c36bf81d (forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}): info

   17 Oct, 04:24 · engine:evaluator

3. compute-1 took job 2899 (forward_check, attempt 1)

   17 Oct, 04:24 · compute-1

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

   17 Oct, 04:24 · engine:scheduler

5. Job 2867 finished

   17 Oct, 04:22 · engine:compute-1

6. H-c36bf81d Frozen → SHADOW vault skipped: last_beaten_lengths, best_last3_beaten do not exist before the discovery zone; track it forward instead

   17 Oct, 04:22 · engine:orchestrator

7. compute-1 took job 2867 (vault, attempt 1)

   17 Oct, 04:22 · compute-1

8. Job 2856 finished

   17 Oct, 04:22 · engine:compute-1

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

   17 Oct, 04:22 · engine:scheduler

10. H-c36bf81d Falsification → FROZEN survived falsification

    17 Oct, 04:22 · engine:falsification

11. falsification result for H-c36bf81d (perturbation: Perturb looser: beaten 16+ last time, best of last 3 within 5 lengths): pass

    17 Oct, 04:22 · engine:evaluator

12. compute-1 took job 2856 (attack, attempt 1)

    17 Oct, 04:22 · compute-1

13. Job 2855 finished

    17 Oct, 04:22 · engine:compute-1

14. falsification result for H-c36bf81d (perturbation: Perturb tighter: beaten 25+ last time, best of last 3 within 7.5 lengths): pass

    17 Oct, 04:22 · engine:evaluator

15. compute-1 took job 2855 (attack, attempt 1)

    17 Oct, 04:22 · compute-1

16. Job 2854 finished

    17 Oct, 04:22 · engine:compute-1

17. falsification result for H-c36bf81d (partition: Quick return (60 days or less): going change without a season/yard-form shift): pass

    17 Oct, 04:22 · engine:evaluator

18. compute-1 took job 2854 (attack, attempt 1)

    17 Oct, 04:22 · compute-1

19. Job 2853 finished

    17 Oct, 04:22 · engine:compute-1