Discovery: error-free chase pull-ups that had led last time 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-b4454f9c (Rival: last-time chase non-finishers with clean jumping remarks beat t) was predicted positive but came out the other way: A/E 0.71 on 208 bets (p 0.074 for the opposite direction). O-1f8ceedf
2. AskedIs the under-performance of error-free chase non-finishers concentrated in horses that led or raced prominently before stopping, where pulling up most likely means a physical problem rather than a horse being given an easy time? answered
3. MechanismA chaser that led and then pulled up without any jumping error usually stopped for a physical reason (wind, bleeding, a lack of fitness). That problem often persists into the next run.
4. Why the market misses itThe form line reads 'PU' with no figure, and punters treat it as a throw-out run. Having led looks like a sign of ability, so the horse is backed to show its earlier form. The remark contains no excuse, and that absence is the warning sign.
5. Predictednegative, A/E 0.70 to 0.88 on 250+ bets
6. Falsified ifA/E of 0.97 or higher, or held-up error-free pull-ups scoring as low as this pool
7. Rule sketchrace_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 1
8. question idQ-9ba41adc

The question's answer so far: No: chase pull-ups with clean jumping that led last time score 0.68 on 43 (CI 0.14 to 1.97), the same as the whole clean pool (0.71 on 208). The broad led/prominent pull-up rival (H-a2870576) was killed, so run style does not concentrate the under-performance.

Mechanism and rationale

A chaser that led and then pulled up without any jumping error usually stopped for a physical reason (wind, bleeding, a lack of fitness). That problem often persists into the next run.

The rule

race_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 1

Reads: race_type, last_race_type, last_pulled_up, last_jumping_error, last_led_in_running

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

36f52790626154eb3b97e837f688adcc4c906f5c13f5dbebda65161dc6d49756

Frozen

not frozen

The definition can still change.

Confounders

• Outsiders at 16/1 or bigger never beat par anywhere in this thread, so a negative A/E driven by long prices is generic and not this mechanism. The under-16/1 refinement checks for this.
• Age and long absences: older chasers and horses off for a long time after a physical problem may underperform for reasons the market already prices. The days-since refinement and a check of the age mix cover this.
• The late-period (after 27 April 2026) summer-jumping skew that has shifted other ground and chase cuts in this thread. Report results split before and after that date.
• Irish heavy-ground pull-up culture: Irish non-finishers sit below par in both the error and the clean pools (0.43, 0.50), so an Irish-heavy pool could produce the shortfall on its own. The GB refinement checks for this.
• The window is short: the remark-derived columns start on 2025-11-01, so the pool covers roughly one jumps season and may be thin.
• Front-runners that went off too fast in a contested pace may have pulled up for tactical rather than physical reasons. That would weaken the effect, not create it.

Discovery result

fail broad rule · 17 Oct, 02:09

Bets

43

3 winners

A/E

0.68

95%: 0.14 to 1.97

SP return

-57.0%

level stakes

Betfair SP return

-51.0%

43 bets with a BSP

Strike rate 7.0%. p = 0.82. Same-race control: the other runners in these races had A/E 1.04, 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 · 42s · 17 Oct, 02:00

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 1",
    "refinements": [
      {
        "label": "under 16/1 (backed runners only)",
        "rule": "race_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 1 AND sp_decimal < 17"
      },
      {
        "label": "GB only (strip Irish pull-up culture)",
        "rule": "race_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 1 AND region = 'GB'"
      },
      {
        "label": "back within 60 days (problem not yet fixed by a break)",
        "rule": "race_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 60"
      }
    ],
    "control": {
      "description": "Same-race other runners should sit near par (about 1.0), so the shortfall is not a property of the races. The main comparison is error-free chase pull-ups that did NOT lead last time (held up or tracked). The mechanism says leading and then stopping with no excuse points to a physical problem, so these runners should score clearly higher than the led pool, at about 0.85 to 1.0. If the control scores as low as the led pool, the effect is generic to clean pull-ups and the 'led' part of the mechanism is falsified.",
      "rule": "race_type = 'Chase' AND last_race_type = 'Chase' AND last_pulled_up = 1 AND last_jumping_error = 0 AND last_led_in_running = 0"
    },
    "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": [
      "Outsiders at 16/1 or bigger never beat par anywhere in this thread, so a negative A/E driven by long prices is generic and not this mechanism. The under-16/1 refinement checks for this.",
      "Age and long absences: older chasers and horses off for a long time after a physical problem may underperform for reasons the market already prices. The days-since refinement and a check of the age mix cover this.",
      "The late-period (after 27 April 2026) summer-jumping skew that has shifted other ground and chase cuts in this thread. Report results split before and after that date.",
      "Irish heavy-ground pull-up culture: Irish non-finishers sit below par in both the error and the clean pools (0.43, 0.50), so an Irish-heavy pool could produce the shortfall on its own. The GB refinement checks for this.",
      "The window is short: the remark-derived columns start on 2025-11-01, so the pool covers roughly one jumps season and may be thin.",
      "Front-runners that went off too fast in a contested pace may have pulled up for tactical rather than physical reasons. That would weaken the effect, not create it."
    ],
    "mechanism": "A chaser that led and then pulled up with no jumping error in the remark usually stopped for a physical reason: a wind problem, bleeding, a heart irregularity or a lack of fitness. It was clearly travelling, then stopped abruptly with no excuse. These problems often persist into the next run, especially after a short break, so the horse wins less often than its price implies.",
    "market_miss": "The form figure 'P' reads as a throw-out run. Having led looks like proof of ability and current fitness, so punters back the horse to show its earlier form. The excuse-free remark is the actual warning sign: nothing explains the stop except the horse itself. Error-free chase non-finishers as a group are already over-bet (A/E 0.71 on 208), and this tests whether the front-runners carry that shortfall.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.70 to 0.88 on roughly 150 to 300 bets",
      "falsified_if": "A/E of 0.97 or higher on the broad rule, or the held-up/non-leading error-free pull-up control scoring as low or lower (control A/E within 0.05 of the led pool)."
    },
    "notes": "I checked all columns with query_catalog. They are pre_off, and the remark-derived ones are available from 2025-11-01 in both GB and IRE, so discovery covers about one season. The draft rule was sound and is kept unchanged. The direction is negative: pass_criteria min_ae 1.05 is to be read on the over-bet side (A/E at or below about 0.952, i.e. 1/1.05). The forward bar's A/E of 1.0 means A/E below 1.0. The floors are kept, not tightened, because the window is short. Report the GB/IRE split and the split before and after 27 April 2026 alongside the headline result."
  }

Timeline

1. Job 2405 finished

   17 Oct, 02:09 · engine:compute-1

2. H-2d1da157 Discovery → KILLED fails: 43 bets, needs 300; p 0.3517 (lower tail), needs 0.05; periods [], each needs 0.99 or lower

   17 Oct, 02:09 · engine:evaluator

3. discovery result for H-2d1da157 (broad rule): fail

   17 Oct, 02:09 · engine:evaluator

4. H-2d1da157 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 02:09 · engine:orchestrator

5. compute-1 took job 2405 (evaluate, attempt 1)

   17 Oct, 02:09 · compute-1

6. Job 2403 finished

   17 Oct, 02:01 · engine:ai-1

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

   17 Oct, 02:01 · engine:scheduler

8. H-2d1da157 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 02:01 · agent:designer@v1

9. ai-1 took job 2403 (design, attempt 1)

   17 Oct, 02:00 · ai-1

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

    17 Oct, 02:00 · engine:scheduler

11. New idea H-2d1da157: Discovery: error-free chase pull-ups that had led last time are over-bet (test-mechanism)

    17 Oct, 02:00 · 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.