Replicate: quick-returning last-time winners at 9/1+ in non-handicaps are over-bet

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-9e92ea71 (Quick-returning last-time winners of any race running in a handicap at) pointed the predicted way but failed the bar: A/E 0.65 on 424 bets (p 0.97), control 1.02. O-3e924322
2. AskedDo quick-returning last-time winners at 9/1+ fall short in non-handicaps too, where the winner tag is equally visible but no mark rise applies? testing
3. MechanismA winner sent off at 9/1+ soon after winning has usually won a weak race or is up in grade. The quick turnout adds little, but punters back the 'in form, out again' story.
4. Why the market misses itThe visible win draws money and shortens the price. The form book does not show how weak the race was.
5. Predictednegative, A/E 0.70 to 0.88 on 300+ bets
6. Falsified ifA/E 0.95 or higher on 300+ bets.
7. Rule sketchlast_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14
8. question idQ-486ae6ea
9. operatorreplicate
10. agent

Mechanism and rationale

A winner sent off at 9/1+ soon after winning has usually won a weak race or is up in grade. The quick turnout adds little, but punters back the 'in form, out again' story.

The rule

last_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14

Reads: last_position, is_handicap, sp_decimal, days_since_last_run

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

52b7b5cfe0f40758dceb44e992e902b18e60f0f62af1638f863e21f2681dd4f2

Frozen

not frozen

The definition can still change.

Confounders

• Class rises into pattern or listed races: the 9/1+ price may correctly reflect a step up in grade, not over-betting
• Penalties in conditions races and novice events carried by recent winners
• Two-year-old flat races, where fast turnarounds and uncertain form are routine and longshot bias is strong
• Small sample: non-handicap 9/1+ quick-returning winners may number only a few hundred
• General longshot under-performance at 9/1+, which the same-race control must separate out
• Period effect: the handicap version weakened in later periods

Control

{
  "description": "Other runners in the same races (non-handicaps), expected near par (A/E about 0.97 to 1.03). That would show the deficit belongs to the quick-returning winners and not to the race. A secondary comparison: last-time winners at 9/1+ in non-handicaps returning after more than 14 days. These should show a smaller shortfall or none. If they fall short just as much, the effect is the winner tag and not the quick return.",
  "rule": "last_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run > 14"
}

Designer

designer@v1

Direction

negative

Discovery result

fail broad rule · 16 Oct, 17:02

Bets

106

2 winners

A/E

0.46

95%: 0.06 to 1.65

SP return

-35.8%

level stakes

Betfair SP return

+19.2%

98 bets with a BSP

Strike rate 1.9%. p = 0.93. Same-race control: the other runners in these races had A/E 1.03, so the selection did not 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 · 27s · 16 Oct, 17:01

  Output (schema-checked)

  {
    "rule": "last_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14",
    "refinements": [
      {
        "label": "Exclude two-year-olds",
        "rule": "last_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND age IS NOT NULL AND age >= 3"
      },
      {
        "label": "9/1 to 14/1 band",
        "rule": "last_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND sp_decimal <= 15 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14"
      },
      {
        "label": "Last win also a non-handicap",
        "rule": "last_position = 1 AND is_handicap = 0 AND last_is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14"
      }
    ],
    "control": {
      "description": "Other runners in the same races (non-handicaps), expected near par (A/E about 0.97 to 1.03). That would show the deficit belongs to the quick-returning winners and not to the race. A secondary comparison: last-time winners at 9/1+ in non-handicaps returning after more than 14 days. These should show a smaller shortfall or none. If they fall short just as much, the effect is the winner tag and not the quick return.",
      "rule": "last_position = 1 AND is_handicap = 0 AND sp_decimal > 9 AND days_since_last_run IS NOT NULL AND days_since_last_run > 14"
    },
    "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": [
      "Class rises into pattern or listed races: the 9/1+ price may correctly reflect a step up in grade, not over-betting",
      "Penalties in conditions races and novice events carried by recent winners",
      "Two-year-old flat races, where fast turnarounds and uncertain form are routine and longshot bias is strong",
      "Small sample: non-handicap 9/1+ quick-returning winners may number only a few hundred",
      "General longshot under-performance at 9/1+, which the same-race control must separate out",
      "Period effect: the handicap version weakened in later periods"
    ],
    "mechanism": "A last-time winner sent off at 9/1+ within 14 days of winning has usually either won a weak race or is stepping up in grade (or carrying a penalty) in a non-handicap. The quick turnout adds no real ability, but punters back the 'in form, out again' story and the winner tag. That shortens the price below the horse's true chance. This replicates the handicap finding (A/E 0.651 on 424) in races where there is no handicap mark to absorb the win.",
    "market_miss": "The recent win and quick return are visible in the form figures and draw money. The weakness of the race won and the size of the step up in grade are not obvious from the figures, so the price stays shorter than the horse's chance warrants.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.70 to 0.88 on 300+ bets, with the same-race control near 1.0",
      "falsified_if": "A/E 0.95 or higher on 300+ bets, or a shortfall that the same-race control matches (control A/E within 0.05 of the rule's)."
    },
    "notes": "The rule sketch is kept unchanged. It uses only pre-off columns and does not read the result. The refinements address the named confounders (two-year-olds, conditions or penalty races) and the price band where the earlier winner-tag shortfall sat. race_class is avoided because it only covers IRE from 2026. For this negative hypothesis, the A/E threshold applies in the over-bet direction: the effective bar is A/E at or below 1/1.05."
  }

Timeline

1. Job 2085 finished

   16 Oct, 17:02 · engine:compute-1

2. H-4f18f696 Discovery → KILLED fails: 106 bets, needs 300; p 0.1881 (lower tail), needs 0.05; periods [0.68], each needs 0.99 or lower

   16 Oct, 17:02 · engine:evaluator

3. discovery result for H-4f18f696 (broad rule): fail

   16 Oct, 17:02 · engine:evaluator

4. H-4f18f696 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 17:02 · engine:orchestrator

5. compute-1 took job 2085 (evaluate, attempt 1)

   16 Oct, 17:02 · compute-1

6. Job 2083 finished

   16 Oct, 17:02 · engine:ai-1

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

   16 Oct, 17:02 · engine:scheduler

8. H-4f18f696 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 17:02 · agent:designer@v1

9. ai-1 took job 2083 (design, attempt 1)

   16 Oct, 17:01 · ai-1

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

    16 Oct, 17:01 · engine:scheduler

11. New idea H-4f18f696: Replicate: quick-returning last-time winners at 9/1+ in non-handicaps are over-bet (replicate)

    16 Oct, 17:01 · 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.