H36: pulled up after being held up

Mechanism and rationale

Pulled up last time having been held up: looked after once the race had gone, so nothing was taken out of it.

The rule

last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield')

Reads: last_pulled_up, last_run_style

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

not designed yet

Frozen

not frozen

The definition can still change.

Broad

last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield')

Kind

remarks

Origin

research/experiments/overnight/results_v2.jsonl

Refinements

[
  {
    "name": "back within 35 days",
    "rule": "(last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield')) AND (days_since_last_run <= 35)"
  },
  {
    "name": "handicap",
    "rule": "(last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield')) AND (is_handicap = 1)"
  },
  {
    "name": "14/1 or shorter",
    "rule": "(last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield')) AND (sp_decimal <= 15)"
  },
  {
    "name": "no repeat jumping errors",
    "rule": "(last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield')) AND (jumping_errors_last3 <= 1)"
  }
]

Tier

A

Result (from the archive)

info no repeat jumping errors · 15 Oct, 23:10

Bets

1,329

A/E

1.20

winners vs the prices

SP return

-32.5%

level stakes

Betfair SP return

-2.1%

1,325 bets with a BSP

Strike rate 6.4%.

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

• investigator v1ok · 1m 2s · 17 Oct, 01:32

  Output (schema-checked)

  {
    "reflection": "H36 is the root of a large branch of work. Held-up or midfield horses that were pulled up last time score A/E 1.19 on 1451 bets. The two periods disagree: 1.07 on 931 bets, then 1.41 on 520. The money is weak: BSP ROI is -5.7% and SP ROI is -34.6%, so the edge shows up in A/E but is mostly eaten by the price, especially at SP. The 14/1-or-shorter cut (GREEN, 1.253 on 510) went live on 2026-10-13 with no forward bets yet.\n\nThe jumping overlays (IQ60–67) have mostly reshuffled the same pool. Of those overlays:\n- 'Jumped well' (1.35 on 156) and 'jumping as usual' (1.30 on 109) held up.\n- 'Out of character bad jumping' collapsed (0.52 on 63).\n- The 'poor jumper anyway' version failed its holdout outright (SYS1: 0.948 on 715).\n\nSo the cuts agree in direction: a pulled-up horse is forgivable when the PU was not caused by the horse failing. But every cut comes from the same twelve months and the same pool, and the holdout failure is a warning.\n\nWhat the chain has not done is test the core cause from a different angle. Two consequences should hold if the cause is real:\n- How soon the trainer brings the horse back (a horse that was looked after can return quickly).\n- Whether the PU was a race-wide event rather than one horse failing.\n\nThere is also the obvious negative control: front-runners who stopped. Rival explanations already killed elsewhere (going change in H101, generic PU/F longshots in H-8841a2b0) should not be retried.",
    "answers": [],
    "questions": [
      {
        "question": "Does the pulled-up lift depend on how quickly the horse is brought back, as it should if the run took nothing out of it?",
        "rationale": "If the horse was looked after, the trainer can bring it back quickly; a long absence after a PU points to injury or a problem the market is right to fear. Columns: days_since_last_run, last_pulled_up, last_run_style, sp_decimal. This tests the mechanism directly rather than re-cutting the pool by jumping figures.",
        "observation_ids": [
          "O-309e6a2d"
        ],
        "priority": 0.6
      },
      {
        "question": "Is the lift strongest when the last race was attritional, with many runners not finishing, so the PU says little about this horse?",
        "rationale": "When a large share of the field pulled up, the P is a property of the race (ground, pace, trip), not the horse, yet the form figure looks the same. Columns: last_race_dnf_share, last_field_size, last_pulled_up, last_run_style.",
        "observation_ids": [
          "O-309e6a2d"
        ],
        "priority": 0.5
      },
      {
        "question": "Do pulled-up horses that led or raced prominently (and so likely stopped) fail to show the lift, confirming that run style is doing the work?",
        "rationale": "This is a negative control for the H36 mechanism. A PU after making the running usually means the horse stopped while trying. If those runners also score about 1.19, then the lift is a general PU discount and not 'held up and looked after'. Columns: last_pulled_up, last_run_style, sp_decimal.",
        "observation_ids": [
          "O-309e6a2d"
        ],
        "priority": 0.55
      },
      {
        "question": "Is the A/E lift a spring/summer artefact, given the periods went 1.07 then 1.41?",
        "rationale": "The second period (from about April 2026) carries most of the lift. Smaller summer jumps fields and better ground could explain it rather than the PU story. Columns: race_date, field_size, going_group, race_type. Forward bets on the live system from October 2026 will partly answer it.",
        "observation_ids": [
          "O-309e6a2d"
        ],
        "priority": 0.35
      }
    ],
    "hypotheses": [
      {
        "title": "Test-mechanism: held-up PU horses brought back quickly beat their price",
        "family": "story",
        "operator": "test-mechanism",
        "question_id": null,
        "new_question": 0,
        "parent_id": "H36-pulled-up-after-being-held-up",
        "reasoning": {
          "observation_ids": [
            "O-309e6a2d"
          ],
          "question": "Does a quick return after a held-up pull-up carry the lift, as the 'nothing taken out of it' mechanism requires?",
          "mechanism": "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.",
          "market_miss": "The 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.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.20 to 1.35 on 500+ bets, higher than the same pool with days_since_last_run > 35",
            "falsified_if": "A/E below 1.05, or no higher than the long-gap complement"
          },
          "rule_sketch": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND days_since_last_run <= 35"
        },
        "confounders": [
          "quick returns cluster in summer jumps, where the period split already shows a lift",
          "yard policy: some trainers run horses often whatever the result",
          "price band mix differs between quick and slow returners"
        ],
        "expected_survival": 0.4
      },
      {
        "title": "Sharpen: held-up PU in an attritional race where many failed to finish",
        "family": "story",
        "operator": "sharpen",
        "question_id": null,
        "new_question": 1,
        "parent_id": "H36-pulled-up-after-being-held-up",
        "reasoning": {
          "observation_ids": [
            "O-309e6a2d"
          ],
          "question": "Is the lift concentrated where the PU was a race-wide event rather than one horse failing?",
          "mechanism": "When a large share of a field does not finish (testing ground, a strong pace, a long trip), held-up runners are pulled up because the race fell apart in front of them. That says nothing about their ability or wellbeing.",
          "market_miss": "The form figure is the same P whether 1 or 6 runners pulled up. The context sits only in the full result, which most punters do not read.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.25 to 1.45 on 300+ bets",
            "falsified_if": "A/E below 1.10, or not above the low-DNF-share complement (last_race_dnf_share < 0.25)"
          },
          "rule_sketch": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND last_race_dnf_share >= 0.25"
        },
        "confounders": [
          "attritional races are mostly soft or heavy, staying chases in winter",
          "the field size of the last race",
          "H101 going-change rival already killed, so ground may not explain it"
        ],
        "expected_survival": 0.35
      },
      {
        "title": "Rival: pulled up after leading or racing prominently shows the same lift",
        "family": "story",
        "operator": "rival-explanation",
        "question_id": null,
        "new_question": 2,
        "parent_id": "H36-pulled-up-after-being-held-up",
        "reasoning": {
          "observation_ids": [
            "O-309e6a2d"
          ],
          "question": "Is the H36 lift a general discount on any pulled-up horse rather than one specific to held-up horses?",
          "mechanism": "If the market simply over-punishes any P, front-runners who pulled up should beat their price as well. If H36's mechanism is right, they stopped while trying, and their price is fair or too short.",
          "market_miss": "Under the rival, punters treat every P alike. Under H36, the front-runner's P really does reflect the horse failing, so it is priced correctly.",
          "prediction": {
            "direction": "negative",
            "size": "A/E 0.85 to 1.00 on 600+ bets (the H36 mechanism predicts no lift)",
            "falsified_if": "A/E at or above 1.15, which would mean the held-up condition is not the cause"
          },
          "rule_sketch": "last_pulled_up = 1 AND last_run_style IN ('led', 'prominent')"
        },
        "confounders": [
          "front-runners pulled up may have had the most physical problems (bleeding, wind)",
          "price bands differ between the styles",
          "run style labels may be noisy for PU runs"
        ],
        "expected_survival": 0.55
      }
    ],
    "findings": [
      {
        "statement": "H36's broad lift (A/E 1.19 on 1451) is unstable across periods (1.07 then 1.41) and loses money at SP (-34.6%) and slightly at BSP (-5.7%). It is a forward-test candidate, not yet a proven edge.",
        "kind": "data",
        "observation_ids": [
          "O-309e6a2d"
        ]
      },
      {
        "statement": "The jumping overlays on the pulled-up pool all re-cut the same twelve months of runners. The live 14/1-or-shorter system has no forward bets, so its 1.253 on 510 remains a discovery figure.",
        "kind": "method",
        "observation_ids": [
          "O-309e6a2d"
        ]
      }
    ]
  }

Timeline

1. hypothesis note H36-pulled-up-after-being-held-up

   15 Oct, 23:10 · importer

2. hypothesis note H36-pulled-up-after-being-held-up

   15 Oct, 23:10 · importer

3. hypothesis note H36-pulled-up-after-being-held-up

   15 Oct, 23:10 · importer

4. H36-pulled-up-after-being-held-up Discovery → PROMISING import: tier A: strong candidate; best [14/1 or shorter] GREEN A/E 1.253 on 510 bets (broad AMBER)

   15 Oct, 23:10 · importer

5. import result for H36-pulled-up-after-being-held-up (no repeat jumping errors): info

   15 Oct, 23:10 · importer

6. import result for H36-pulled-up-after-being-held-up (14/1 or shorter): info

   15 Oct, 23:10 · importer

7. import result for H36-pulled-up-after-being-held-up (handicap): info

   15 Oct, 23:10 · importer

8. import result for H36-pulled-up-after-being-held-up (back within 35 days): info

   15 Oct, 23:10 · importer

9. import result for H36-pulled-up-after-being-held-up (broad): info

   15 Oct, 23:10 · importer

10. New idea H36-pulled-up-after-being-held-up: H36: pulled up after being held up (import)

    15 Oct, 23:10 · importer

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Promising.