Rival: quick-returning last-time winners beat their price whether or not they won easily

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
   • H34-won-easily-last-time (H34: won easily last time) was GREEN in the earlier research: A/E 1.05 on 2374 bets (periods 1.08 / 1.02); now PROMISING. O-65eec625
2. AskedIs the lift for easy winners back within 14 days caused by the easy-win flag, or do all last-time winners back within 14 days beat their price? answered
3. MechanismA yard that brings a winner straight back is signalling that the horse is well and is exploiting it before reassessment. That could produce the lift with or without an easy win.
4. Why the market misses itThe market sees a fresh winner but may underrate how much the yard's intent is worth. If this rival holds, the 'easy' label adds nothing.
5. Predictedpositive, A/E 1.08 to 1.15 on 1500+ bets if the rival is true; about 1.00 if the easy flag is the driver
6. Falsified ifWinners that did not win easily, back within 14 days, score A/E at or below 1.02 on 1000+ bets, which leaves the easy flag as the cause
7. Rule sketchlast_position = 1 AND last_won_easily = 0 AND days_since_last_run <= 14

Mechanism and rationale

A yard that brings a winner straight back is signalling that the horse is well and is exploiting it before reassessment. That could produce the lift with or without an easy win.

The rule

last_position = 1 AND last_won_easily = 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14

Reads: last_position, last_won_easily, days_since_last_run

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

80328ff7dfc68b28fa3461734ac462f747d819e2ea41b1d6b9b87650c778730c

Frozen

not frozen

The definition can still change.

Confounders

• Penalty-carrying runners in handicaps: a quick return often means racing under a penalty before the new mark applies, which can push the effect either way
• Race class of the last win: a quick return after a low-class win may simply be a class rise or drop
• Summer turf and all-weather calendar density: quick turnarounds bunch in busy periods with small fields
• Trainer type: a few high-volume yards that habitually turn winners out fast may drive the whole result
• Short data window: last_won_easily only exists from 2025-11-01, so the sample spans about one year and the season mix is uneven
• Same-horse repeats: one horse winning and going again several times can supply many of the bets

Control

{
  "description": "The lab's same-race control (other runners in the same races) should sit near 1.00; if it is also lifted, the effect is a race property (e.g. small summer fields). The question needs a second comparison: quick-returning EASY winners under the same window. If the rival is true, non-easy quick returners score about as well as easy ones (both above 1.05). If the easy flag is the driver, easy quick returners score clearly above non-easy ones, which sit near 1.00.",
  "rule": "last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14"
}

Discovery result

fail broad rule · 16 Oct, 16:16

Bets

2,015

432 winners

A/E

1.00

95%: 0.91 to 1.10

SP return

-19.5%

level stakes

Betfair SP return

-6.4%

2,001 bets with a BSP

Strike rate 21.4%. p = 0.52. 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 · 40s · 16 Oct, 16:15

  Output (schema-checked)

  {
    "rule": "last_position = 1 AND last_won_easily = 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14",
    "refinements": [
      {
        "label": "Handicaps only: exploit the mark before reassessment",
        "rule": "last_position = 1 AND last_won_easily = 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1"
      },
      {
        "label": "Very quick return (7 days or fewer): strongest yard-intent signal",
        "rule": "last_position = 1 AND last_won_easily = 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 7"
      },
      {
        "label": "14/1 or shorter: intent edges should sit with fancied runners, not big prices",
        "rule": "last_position = 1 AND last_won_easily = 0 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND sp_decimal <= 15"
      }
    ],
    "control": {
      "description": "The lab's same-race control (other runners in the same races) should sit near 1.00; if it is also lifted, the effect is a race property (e.g. small summer fields). The question needs a second comparison: quick-returning EASY winners under the same window. If the rival is true, non-easy quick returners score about as well as easy ones (both above 1.05). If the easy flag is the driver, easy quick returners score clearly above non-easy ones, which sit near 1.00.",
      "rule": "last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Penalty-carrying runners in handicaps: a quick return often means racing under a penalty before the new mark applies, which can push the effect either way",
      "Race class of the last win: a quick return after a low-class win may simply be a class rise or drop",
      "Summer turf and all-weather calendar density: quick turnarounds bunch in busy periods with small fields",
      "Trainer type: a few high-volume yards that habitually turn winners out fast may drive the whole result",
      "Short data window: last_won_easily only exists from 2025-11-01, so the sample spans about one year and the season mix is uneven",
      "Same-horse repeats: one horse winning and going again several times can supply many of the bets"
    ],
    "mechanism": "A yard that brings a winner back within 14 days is showing it thinks the horse is in form and wants to use that form (and, in handicaps, the old mark) before the handicapper or the horse's condition catches up. If that intent is the driver, the lift should appear in winners whatever the manner of the win, not only in the ones marked 'won easily'.",
    "market_miss": "Punters price a fresh winner off its form figure and the penalty, but they may give too little weight to what the quick turnaround says about the yard's confidence. If this rival holds, the 'won easily' label adds nothing to the quick-return signal. If it fails, the market prices plain quick-returning winners correctly and only the easy-win comment is underused.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.08 to 1.15 if the rival is true; about 1.00 if the easy flag is the driver",
      "falsified_if": "Winners that did not win easily and are back within 14 days score A/E at or below 1.02 on 300+ bets (or below the easy-winner control by 0.05 or more), which leaves the easy flag as the cause"
    },
    "notes": "last_won_easily is only populated from 2025-11-01 (GB and IRE). Using = 0 drops NULL rows, so the rule keeps to the window where the flag is defined. The hypothesis's 1500+ bet target may not be reachable there, so min_bets is set at 300 (stricter than the floor) and the falsification sample is scaled to match. Added an explicit days_since_last_run IS NOT NULL guard. The refinements come from the mechanism: handicaps carry the reassessment logic, a 7-day return is the sharpest intent signal, and the 14/1 price cap follows the family finding that edges die at big prices."
  }

Timeline

1. Job 1960 finished

   16 Oct, 16:16 · engine:compute-1

2. H-013024cb Discovery → KILLED fails: A/E 0.999, needs 1.05; p 0.5171, needs 0.05 or lower; periods [1.04, 0.96], each needs 1.0; A/E 0.999 against same-race control 1.001, needs 0.03 above (not a niche pass: A/E 0.999 below 1.05; p 0.5171364979311655 above 0.5; does not beat the same-race control 1.001)

   16 Oct, 16:16 · engine:evaluator

3. discovery result for H-013024cb (broad rule): fail

   16 Oct, 16:16 · engine:evaluator

4. H-013024cb Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 16:16 · engine:orchestrator

5. compute-1 took job 1960 (evaluate, attempt 1)

   16 Oct, 16:16 · compute-1

6. Job 1958 finished

   16 Oct, 16:15 · engine:ai-1

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

   16 Oct, 16:15 · engine:scheduler

8. H-013024cb Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 16:15 · agent:designer@v1

9. ai-1 took job 1958 (design, attempt 1)

   16 Oct, 16:15 · ai-1

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

    16 Oct, 16:15 · engine:scheduler

11. New idea H-013024cb: Rival: quick-returning last-time winners beat their price whether or not they won easily (rival-explanation)

    16 Oct, 16:15 · 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.