Test-mechanism: held-up PU horses brought back quickly beat their price

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
   • H36-pulled-up-after-being-held-up (H36: pulled up after being held up) was GREEN in the earlier research: A/E 1.19 on 1451 bets (periods 1.07 / 1.41); now PROMISING. O-309e6a2d
2. AskedDoes the pulled-up lift depend on how quickly the horse is brought back, as it should if the run took nothing out of it? answered
3. MechanismA horse that was held up, never got into the race and was pulled up while its chance was gone had a light, short race. The trainer can run it again within a few weeks. A long gap after a PU more often means the horse had a problem.
4. Why the market misses itThe form line shows a P whatever the gap. Punters see a recent bad run and mark the horse down, while the trainer's quick return is a signal they overlook.
5. Predictedpositive, A/E 1.20 to 1.35 on 500+ bets, higher than the same pool with days_since_last_run > 35
6. Falsified ifA/E below 1.05, or no higher than the long-gap complement
7. Rule sketchlast_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run <= 35

Mechanism and rationale

A horse that was held up, never got into the race and was pulled up while its chance was gone had a light, short race. The trainer can run it again within a few weeks. A long gap after a PU more often means the horse had a problem.

The rule

last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run IS NOT NULL AND days_since_last_run <= 35

Reads: last_pulled_up, last_run_style, days_since_last_run

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

d749bfafaec5a2352d83b247912deded63b6042a506440db1cb82e9fdd30893e

Frozen

not frozen

The definition can still change.

Confounders

• Season: quick returns cluster in summer jumps, where the pulled-up pool's period split already shows a lift; check the result by month or season
• Yard policy: some trainers run horses often whatever the result, so a quick return may reflect the yard rather than how hard the race was; check that no few trainers dominate the lift
• Price mix: quick and slow returners may sit in different price bands, and the PU lift has been concentrated at long prices before; compare within bands (the 14/1-or-shorter refinement)
• Overlap with the one-off chase-error pool (also in the PU/beaten set); confirm the lift is not just those runners
• Short data window: last_run_style and last_pulled_up exist only from 2025-11-01, so discovery covers about one year and one summer

Control

{
  "description": "Two comparisons. (1) Same-race other runners: expected to score near 1.00; if they also show a lift, the effect belongs to the races (e.g. summer jumps cards), not to the horses. (2) The mechanism's own complement: the same held-up/midfield pull-up pool returning after more than 35 days. The mechanism says a long gap after a PU more often signals a problem, so this group should score clearly lower (near or below 1.00). If the long-gap group scores as high as or higher than the quick returners, the 'light race, quick return' mechanism is falsified even if the broad rule passes.",
  "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run IS NOT NULL AND days_since_last_run > 35"
}

Discovery result

fail broad rule · 17 Oct, 01:35

Bets

777

39 winners

A/E

1.03

95%: 0.73 to 1.40

SP return

-43.2%

level stakes

Betfair SP return

-16.4%

775 bets with a BSP

Strike rate 5.0%. p = 0.46. Same-race control: the other runners in these races had A/E 1.00, 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 · 31s · 17 Oct, 01:33

  Output (schema-checked)

  {
    "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run IS NOT NULL AND days_since_last_run <= 35",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run IS NOT NULL AND days_since_last_run <= 35 AND sp_decimal <= 15"
      },
      {
        "label": "held up only (purest 'never got involved' case)",
        "rule": "last_pulled_up = 1 AND last_run_style = 'held_up' AND days_since_last_run IS NOT NULL AND days_since_last_run <= 35"
      },
      {
        "label": "very quick return, 21 days or fewer",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run IS NOT NULL AND days_since_last_run <= 21"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) Same-race other runners: expected to score near 1.00; if they also show a lift, the effect belongs to the races (e.g. summer jumps cards), not to the horses. (2) The mechanism's own complement: the same held-up/midfield pull-up pool returning after more than 35 days. The mechanism says a long gap after a PU more often signals a problem, so this group should score clearly lower (near or below 1.00). If the long-gap group scores as high as or higher than the quick returners, the 'light race, quick return' mechanism is falsified even if the broad rule passes.",
      "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run IS NOT NULL AND days_since_last_run > 35"
    },
    "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": [
      "Season: quick returns cluster in summer jumps, where the pulled-up pool's period split already shows a lift; check the result by month or season",
      "Yard policy: some trainers run horses often whatever the result, so a quick return may reflect the yard rather than how hard the race was; check that no few trainers dominate the lift",
      "Price mix: quick and slow returners may sit in different price bands, and the PU lift has been concentrated at long prices before; compare within bands (the 14/1-or-shorter refinement)",
      "Overlap with the one-off chase-error pool (also in the PU/beaten set); confirm the lift is not just those runners",
      "Short data window: last_run_style and last_pulled_up exist only from 2025-11-01, so discovery covers about one year and one summer"
    ],
    "mechanism": "A horse that was held up or raced midfield, never got into contention and was pulled up once its chance had gone had a short, easy race and took little out of itself. That is why the trainer can run it again within about five weeks. A long gap after a pull-up more often points to a physical problem. So a quick return after a held-up pull-up marks a sound horse whose P misrepresents its ability.",
    "market_miss": "The form line shows the same 'P' whatever the gap and however the horse was ridden. Punters mark down a fresh bad run. They overlook two signals: the horse was never asked for an effort, and the trainer brought it back quickly, which shows confidence that it is sound.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.20 to 1.35, ideally on 500+ bets (the data window starts 2025-11, so fewer is possible), clearly above the >35-day complement",
      "falsified_if": "A/E below 1.05 on the broad rule, or the broad rule's A/E is no higher than the >35-day complement's, or same-race other runners show a comparable lift"
    },
    "notes": "Kept the rule sketch and added an explicit days_since_last_run IS NOT NULL guard. The run-style values 'held_up' and 'midfield' are taken from the sketch. The catalogue has no value list, so if either value matches zero rows, the rule should be treated as malformed rather than as a null result. All key columns start on 2025-11-01, so expect a thin sample. A directionally correct result under 150 bets goes to falsification and forward tracking as lab policy says."
  }

Timeline

1. Job 2279 finished

   17 Oct, 01:35 · engine:compute-1

2. H-ce4188c2 Discovery → KILLED fails: A/E 1.027, needs 1.05; p 0.4565, needs 0.05 or lower; periods [0.97, 1.22], each needs 1.0; A/E 1.027 against same-race control 1.0, needs 0.03 above (not a niche pass: A/E 1.027 below 1.05; does not beat the same-race control 1.0)

   17 Oct, 01:35 · engine:evaluator

3. discovery result for H-ce4188c2 (broad rule): fail

   17 Oct, 01:35 · engine:evaluator

4. H-ce4188c2 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 01:35 · engine:orchestrator

5. compute-1 took job 2279 (evaluate, attempt 1)

   17 Oct, 01:35 · compute-1

6. Job 2276 finished

   17 Oct, 01:34 · engine:ai-1

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

   17 Oct, 01:34 · engine:scheduler

8. H-ce4188c2 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 01:34 · agent:designer@v1

9. ai-1 took job 2276 (design, attempt 1)

   17 Oct, 01:33 · ai-1

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

    17 Oct, 01:33 · engine:scheduler

11. New idea H-ce4188c2: Test-mechanism: held-up PU horses brought back quickly beat their price (test-mechanism)

    17 Oct, 01:33 · 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.