Rival: quick-returning handicap non-winners at 9/1+ are over-bet too (quick-return effect, not winner tag)

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 handicap runners at 9/1+ that did not win last time but are back within 14 days also fall short, so that the quick-return deficit is about the quick return rather than the winner tag? answered
3. MechanismA horse turned out again within two weeks and sent off at a long price may simply be under-recovered or opportunistically placed, whatever its last result. Quick runs drain long-priced, marginal horses more.
4. Why the market misses itThe market prices recent form and the quick return as a sign of well-being. It does not discount the fatigue cost to a weak horse.
5. Predictednegative, A/E 0.85 to 0.93 on 3,000+ bets if the rival holds
6. Falsified ifA/E 0.97 or higher on 2,000+ bets. That would leave the deficit specific to last-time winners.
7. Rule sketchis_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND sp_decimal > 9

Mechanism and rationale

A horse turned out again within two weeks and sent off at a long price may simply be under-recovered or opportunistically placed, whatever its last result. Quick runs drain long-priced, marginal horses more.

The rule

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

Reads: is_handicap, last_position, days_since_last_run, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

8fa0e9007a58a933c5112f6dc473c796b1e5617631ca858a2f6d8f3799f03e1a

Frozen

not frozen

The definition can still change.

Confounders

• Price-band mix: non-winners skew to 16/1+, where longshot bias alone lowers A/E. Compare within the 9/1–14/1 and 16/1+ bands separately.
• Field size: big summer handicap fields carry more long-priced runners and higher overrounds.
• Flat summer turnover: quick returns cluster in Flat summer handicaps, so the season or code mix may drive the result rather than fatigue.
• Trainer placement patterns: some yards run horses quickly to exploit a mark, while others run them back as exercise.
• Last-run quality: recent non-winners include beaten-a-neck seconds and tailed-off runners. A heavy, tiring last run (or a non-finish) may matter more than the gap.
• Region: earlier negative shortfalls were often stronger in IRE, so check GB and IRE separately.

Control

{
  "description": "Same-race other runners should be near par (about 1.00). Fatigue is a property of the horse, not the race, so they should not share the deficit. The key comparison is the same population with a normal gap: long-priced handicap non-winners returning after more than 14 days. The prior 9/1–14/1 non-winner pool (gap-agnostic) scored 1.03 on 15,550, so the control is expected near 1.00–1.03. The rival holds only if the quick-return group is clearly below both this control and the same-race runners. If the quick-return non-winners sit at par while quick-return winners scored 0.65, the deficit is specific to the winner tag.",
  "rule": "is_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run > 14 AND sp_decimal > 9"
}

Discovery result

fail broad rule · 16 Oct, 17:02

Bets

7,481

309 winners

A/E

0.99

95%: 0.88 to 1.11

SP return

-32.1%

level stakes

Betfair SP return

-4.6%

7,120 bets with a BSP

Strike rate 4.1%. p = 0.59. Same-race control: the other runners in these races had A/E 1.00, 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 · 32s · 16 Oct, 17:01

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND sp_decimal > 9",
    "refinements": [
      {
        "label": "Up to 14/1 band (where the winner-tag shortfall sat in gap-agnostic pools)",
        "rule": "is_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND sp_decimal > 9 AND sp_decimal <= 15"
      },
      {
        "label": "16/1+ band (where quick-returning winners fell furthest, 0.448)",
        "rule": "is_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND sp_decimal > 15"
      },
      {
        "label": "Very quick return (7 days or less): dose-response for the fatigue mechanism",
        "rule": "is_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 7 AND sp_decimal > 9"
      }
    ],
    "control": {
      "description": "Same-race other runners should be near par (about 1.00). Fatigue is a property of the horse, not the race, so they should not share the deficit. The key comparison is the same population with a normal gap: long-priced handicap non-winners returning after more than 14 days. The prior 9/1–14/1 non-winner pool (gap-agnostic) scored 1.03 on 15,550, so the control is expected near 1.00–1.03. The rival holds only if the quick-return group is clearly below both this control and the same-race runners. If the quick-return non-winners sit at par while quick-return winners scored 0.65, the deficit is specific to the winner tag.",
      "rule": "is_handicap = 1 AND last_position IS NOT NULL AND last_position > 1 AND days_since_last_run IS NOT NULL AND days_since_last_run > 14 AND sp_decimal > 9"
    },
    "pass_criteria": {
      "min_bets": 1000,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 300,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Price-band mix: non-winners skew to 16/1+, where longshot bias alone lowers A/E. Compare within the 9/1–14/1 and 16/1+ bands separately.",
      "Field size: big summer handicap fields carry more long-priced runners and higher overrounds.",
      "Flat summer turnover: quick returns cluster in Flat summer handicaps, so the season or code mix may drive the result rather than fatigue.",
      "Trainer placement patterns: some yards run horses quickly to exploit a mark, while others run them back as exercise.",
      "Last-run quality: recent non-winners include beaten-a-neck seconds and tailed-off runners. A heavy, tiring last run (or a non-finish) may matter more than the gap.",
      "Region: earlier negative shortfalls were often stronger in IRE, so check GB and IRE separately."
    ],
    "mechanism": "A horse turned out again within 14 days and sent off at a long price (over 9/1) in a handicap may be under-recovered, or placed opportunistically, whatever its last result. Weak, marginal horses have the least in reserve, so a quick turnaround should cost them more than it costs well-backed horses. If quick-returning winners at 9/1+ (0.65 on 424) were showing a fatigue effect, quick-returning non-winners should show a similar shortfall.",
    "market_miss": "Punters read a quick return as the trainer showing intent and the horse being well. They price the recent form line but do not discount the physical cost of a short gap for a modest, long-priced horse. Few bettors buy these runners at big prices, so the error is not corrected.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.85 to 0.93 on 3,000+ bets, below the >14-day control and the same-race runners",
      "falsified_if": "A/E 0.97 or higher on 2,000+ bets, or no gap to the >14-day non-winner control. Either result would leave the 0.65 deficit specific to last-time winners (the winner tag) rather than to the quick return."
    },
    "notes": "The rule sketch is kept unchanged. It uses only pre-off columns, and last_position > 1 excludes winners. Runners that did not finish last time may have a NULL last_position, so they drop out; that is acceptable for this test. The min_bets bar is raised to 1,000 because the pool should be large (3,000+ expected), so a thin result would be a warning sign in itself. This is a rival-explanation test: a par result is informative, and it supports the winner-tag reading of H-9e92ea71. Prior evidence leans against the rival: gap-agnostic 9/1–14/1 handicap non-winners scored 1.03 on 15,550. The refinements test price bands, matching the band split seen in quick-returning winners, and a 7-day dose-response."
  }

Timeline

1. Job 2084 finished

   16 Oct, 17:02 · engine:compute-1

2. H-ab00e998 Discovery → KILLED fails: A/E 0.989, needs 0.95 or lower; p 0.4372 (lower tail), needs 0.05; periods [0.97, 1.0], each needs 0.99 or lower; A/E 0.989 against same-race control 1.005, needs 0.03 below

   16 Oct, 17:02 · engine:evaluator

3. discovery result for H-ab00e998 (broad rule): fail

   16 Oct, 17:02 · engine:evaluator

4. H-ab00e998 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 17:02 · engine:orchestrator

5. compute-1 took job 2084 (evaluate, attempt 1)

   16 Oct, 17:02 · compute-1

6. Job 2082 finished

   16 Oct, 17:01 · engine:ai-1

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

   16 Oct, 17:01 · engine:scheduler

8. H-ab00e998 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 17:01 · agent:designer@v1

9. ai-1 took job 2082 (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-ab00e998: Rival: quick-returning handicap non-winners at 9/1+ are over-bet too (quick-return effect, not winner tag) (rival-explanation)

    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.