Rival: the no-drop lift is quick returners racing before the new mark applies

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-4392b730 (Rival: the jumps winner rebound is mark relief, strongest when the han) passed discovery as a niche: A/E 1.25 on 380 bets (p 0.08), same-race control 0.99. O-4822a69b
2. AskedIs the stronger rebound of undropped recent winners (rating_change >= 0) an official-excuse signal, or a quick return that came before the handicapper's reassessment? answered
3. MechanismA yard that brings a recent winner back within about 10 days of a heavy beating knows the horse is well, and runs it before any rating change. The quick turnaround, not the handicapper's view, is the information.
4. Why the market misses itPunters see a 20-length beating a few days ago and assume the horse needs time. The yard's quick return signals that nothing is wrong.
5. Predictedpositive, A/E 1.30+ on 80 to 150 bets
6. Falsified ifA/E at or below 1.10, or more than 0.10 below the days_since_last_run >= 14 version, which rules the quick-return explanation out
7. Rule sketchis_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change >= 0 AND days_since_last_run <= 10
8. question idQ-d13a01e9
9. operator

The question's answer so far: It is not a timing artefact. Only 12 undropped winners came back before reassessment, while horses given time to be reassessed scored 1.42 on 249 (H-6243bb06, 6/6 attacks held).

Mechanism and rationale

A yard that brings a recent winner back within about 10 days of a heavy beating knows the horse is well, and runs it before any rating change. The quick turnaround, not the handicapper's view, is the information.

The rule

is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 10

Reads: is_handicap, race_type, last_beaten_lengths, best_last3_beaten, rating_change, days_since_last_run

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

77f85d789c8395c04d879bc2f7d94d9a131a243243b87ec88a85145fac1e26fa

Frozen

not frozen

The definition can still change.

Confounders

• Small sample: rating_change and best_last3_beaten only start in Nov 2025, so this window will probably give fewer than 150 bets, and a few longshot winners could carry the result
• Summer jumping: quick turnarounds bunch up on summer jumps cards with weaker fields, so a seasonal effect could pass for a timing effect
• Penalty-carrying winners: a horse that won elsewhere since its last handicap may carry a penalty, so rating_change >= 0 may reflect the penalty rather than the handicapper's view
• IRE share: earlier findings put the undropped rebound in IRE (1.71 on 88), so the lift may be regional rather than about timing
• Mark lag: a horse back within 10 days races off its old mark by rule, so rating_change >= 0 is close to automatic here and the rule partly restates days_since_last_run
• Yard effects: a few trainers who turn horses out quickly may dominate the sample

Control

{
  "description": "The deciding comparison is the slow-return version of the same profile: rating held or raised, but back 14 days or more after the heavy beating. If the quick-return explanation is right, the <=10-day rule beats that group by more than 0.10 A/E. If the slow group scores about the same or better, the no-drop lift is not about timing. Same-race rivals should sit near 1.0 (about 0.90 to 1.00), which shows the lift is not a property of the races.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run >= 14"
}

Discovery result

fail broad rule · 17 Oct, 04:47

Bets

12

2 winners

A/E

1.31

95%: 0.16 to 4.73

SP return

+20.8%

level stakes

Betfair SP return

–

12 bets with a BSP

Strike rate 16.7%. p = 0.45. Same-race control: the other runners in these races had A/E 0.96, 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 · 35s · 17 Oct, 04:45

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 10",
    "refinements": [
      {
        "label": "very quick return (7 days or fewer)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 7"
      },
      {
        "label": "14/1 or shorter, because the family's rebound lifts have leaned on a few 16/1+ winners",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 10 AND sp_decimal <= 15"
      },
      {
        "label": "GB only, because earlier findings split the rebound lift by region",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 10 AND region = 'GB'"
      }
    ],
    "control": {
      "description": "The deciding comparison is the slow-return version of the same profile: rating held or raised, but back 14 days or more after the heavy beating. If the quick-return explanation is right, the <=10-day rule beats that group by more than 0.10 A/E. If the slow group scores about the same or better, the no-drop lift is not about timing. Same-race rivals should sit near 1.0 (about 0.90 to 1.00), which shows the lift is not a property of the races.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths IS NOT NULL AND last_beaten_lengths >= 20 AND best_last3_beaten IS NOT NULL AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0 AND days_since_last_run IS NOT NULL AND days_since_last_run >= 14"
    },
    "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": [
      "Small sample: rating_change and best_last3_beaten only start in Nov 2025, so this window will probably give fewer than 150 bets, and a few longshot winners could carry the result",
      "Summer jumping: quick turnarounds bunch up on summer jumps cards with weaker fields, so a seasonal effect could pass for a timing effect",
      "Penalty-carrying winners: a horse that won elsewhere since its last handicap may carry a penalty, so rating_change >= 0 may reflect the penalty rather than the handicapper's view",
      "IRE share: earlier findings put the undropped rebound in IRE (1.71 on 88), so the lift may be regional rather than about timing",
      "Mark lag: a horse back within 10 days races off its old mark by rule, so rating_change >= 0 is close to automatic here and the rule partly restates days_since_last_run",
      "Yard effects: a few trainers who turn horses out quickly may dominate the sample"
    ],
    "mechanism": "When a recent handicap jumps winner (a win in its last three runs) is brought back within 10 days of a 20+ length beating, the yard is signalling that the horse is well and that the beating was a one-off. The horse runs before the handicapper can drop its mark. The quick return carries the information. The theory is that the no-drop lift found earlier comes from these horses, not from the handicapper's view of them.",
    "market_miss": "Punters weight the fresh 20-length beating heavily and assume a horse turned out days later is either short of peak or being run out of desperation. They fail to read the yard's quick turnaround as a sign that the horse is in good health, so its price drifts more than its true chance justifies.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.30+ on about 80 to 150 bets",
      "falsified_if": "A/E at or below 1.10. Also falsified if it is more than 0.10 below the control version (days_since_last_run >= 14, otherwise identical), which would rule out the quick-return explanation for the no-drop lift."
    },
    "notes": "The rule sketch is kept as written, with explicit IS NOT NULL guards added. All columns are pre_off. rating_change and best_last3_beaten both start on 2025-11-01, so the sample is likely to fall short of 150 bets. Under lab policy a thin result pointing the right way goes on to falsification and forward tracking. Pass criteria are set slightly above the floor (A/E 1.10) to match the falsification line."
  }

Timeline

1. Job 3028 finished

   17 Oct, 04:47 · engine:compute-1

2. H-0b37c0ca Discovery → KILLED fails: 12 bets, needs 150; p 0.4515, needs 0.05 or lower; periods [], each needs 1.0 (not a niche pass: only 12 bets)

   17 Oct, 04:47 · engine:evaluator

3. discovery result for H-0b37c0ca (broad rule): fail

   17 Oct, 04:47 · engine:evaluator

4. H-0b37c0ca Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 04:47 · engine:orchestrator

5. compute-1 took job 3028 (evaluate, attempt 1)

   17 Oct, 04:47 · compute-1

6. Job 3025 finished

   17 Oct, 04:46 · engine:ai-1

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

   17 Oct, 04:46 · engine:scheduler

8. H-0b37c0ca Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 04:46 · agent:designer@v1

9. ai-1 took job 3025 (design, attempt 1)

   17 Oct, 04:45 · ai-1

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

    17 Oct, 04:45 · engine:scheduler

11. New idea H-0b37c0ca: Rival: the no-drop lift is quick returners racing before the new mark applies (rival-explanation)

    17 Oct, 04:45 · 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.