Rival: last-time chase non-finishers with clean jumping remarks beat their price too

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-3cc36f8a (Rival: the 8/1 to 14/1 chase lift is generic to well-beaten chasers wi) pointed the predicted way but failed the bar: A/E 1.18 on 302 bets (p 0.237), control 0.99. O-026ae1a6
   • H-c66cb8b5 (Rival: chase-error lift is an 8/1 to 14/1 price-band effect, not a dri) survived falsification: 6 of 6 attacks left it standing; it is frozen for its one vault query. O-929a781d
2. AskedDoes a last-time chase non-completion after two clean-jumping runs beat its price whether or not the remark records a jumping error? answered
3. MechanismRival cause: any chase non-completion (mostly pull-ups) leaves no beaten distance. The market reads the P as lost form, and it overreacts whatever caused it. On this account the jumping-error remark plays no part in the lift.
4. Why the market misses itA P in the form figures looks as bad as a heavy defeat, but it often reflects tactics, ground or a horse that was looked after. The held-up pull-up thread showed a lift of 1.2 to 1.3 from the same blind spot.
5. Predictedpositive, If the rival holds: A/E 1.25 or higher on 300+ bets, both periods above 1.05. If the error story holds: A/E 1.10 or lower.
6. Falsified ifA/E of 1.10 or lower, or a lift only after 27 April 2026, which would leave the error non-completion pool's lift specific to the error

Mechanism and rationale

Rival cause: any chase non-completion (mostly pull-ups) leaves no beaten distance. The market reads the P as lost form, and it overreacts whatever caused it. On this account the jumping-error remark plays no part in the lift.

The rule

race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal < 17

Reads: race_type, remarked_last3, jumping_errors_last3, last_position, days_since_last_run, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

eb58fba4c5b8e57830b5d222a143c6077747c807201c8f8db8d4a786b649d2c0

Frozen

not frozen

The definition can still change.

Confounders

• Overlaps with the held-up pull-up thread: any lift may just repeat that thread's 1.2 to 1.3 rather than be a new effect
• Pulled-up horses include ones with known physical problems that the market rightly marks down, which dilutes the pool towards par
• Late-period skew: the generic well-beaten chaser bounce-back appears only after 27 April 2026, so a lift confined to that period is a calendar effect
• GB/IRE mix differs from the error pool, and GB carried the error-free finisher lift
• remarked_last3 only exists from 2025-11-01, so the window is short and the sample thin
• Last run may have been a hurdle or point, not a chase. Refinement 1 isolates last-time chase non-completions
• Before 2025 every non-finisher is coded PU. Inside this window F/UR are mostly excluded by jumping_errors_last3 = 0, but BD/SU may remain

Control

{
  "description": "Same-race other runners should be near par (about 0.95 to 1.0), as in the error-pool work (0.97). The rival account is decided by comparing against the one-off-error non-completion pool, built on the same frame. If the clean pool matches that pool's lift (about 1.3 or more), the error remark adds nothing. If the clean pool sits at or below 1.10 while the error pool is lifted, the error story holds. Both pools should be split at 27 April 2026 and by region.",
  "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal < 17"
}

Discovery result

fail broad rule · 17 Oct, 01:59

Bets

208

20 winners

A/E

0.71

95%: 0.44 to 1.10

SP return

-29.8%

level stakes

Betfair SP return

-16.3%

208 bets with a BSP

Strike rate 9.6%. p = 0.95. Same-race control: the other runners in these races had A/E 1.06, 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 · 46s · 17 Oct, 01:57

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal < 17",
    "refinements": [
      {
        "label": "last run was itself a chase",
        "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal < 17 AND last_race_type = 'Chase'"
      },
      {
        "label": "pull-ups only (the P the market misreads)",
        "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal < 17 AND last_pulled_up = 1"
      },
      {
        "label": "8/1 to 14/1 band, matching the error pool's strongest band",
        "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 0 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal >= 9 AND sp_decimal < 16"
      }
    ],
    "control": {
      "description": "Same-race other runners should be near par (about 0.95 to 1.0), as in the error-pool work (0.97). The rival account is decided by comparing against the one-off-error non-completion pool, built on the same frame. If the clean pool matches that pool's lift (about 1.3 or more), the error remark adds nothing. If the clean pool sits at or below 1.10 while the error pool is lifted, the error story holds. Both pools should be split at 27 April 2026 and by region.",
      "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND last_position IS NULL AND days_since_last_run IS NOT NULL AND sp_decimal < 17"
    },
    "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": [
      "Overlaps with the held-up pull-up thread: any lift may just repeat that thread's 1.2 to 1.3 rather than be a new effect",
      "Pulled-up horses include ones with known physical problems that the market rightly marks down, which dilutes the pool towards par",
      "Late-period skew: the generic well-beaten chaser bounce-back appears only after 27 April 2026, so a lift confined to that period is a calendar effect",
      "GB/IRE mix differs from the error pool, and GB carried the error-free finisher lift",
      "remarked_last3 only exists from 2025-11-01, so the window is short and the sample thin",
      "Last run may have been a hurdle or point, not a chase. Refinement 1 isolates last-time chase non-completions",
      "Before 2025 every non-finisher is coded PU. Inside this window F/UR are mostly excluded by jumping_errors_last3 = 0, but BD/SU may remain"
    ],
    "mechanism": "A chase non-completion (mostly a pull-up) leaves no beaten distance. The market reads the P as lost form and overreacts, whatever the cause: tactics, ground, or a horse that was looked after. If this rival cause is right, the one-off-error non-completion lift is not about the jumping error. Clean-jumping non-finishers at up to 15/1 should beat their price by a similar amount.",
    "market_miss": "Form figures show a P that looks as bad as a heavy defeat or worse, with no beaten distance to put it in context. Punters and compilers mark the horse down as out of form. Clean jumping remarks across the last three runs suggest the non-completion was not a sign of a jumping problem. The held-up pull-up thread showed a 1.2 to 1.3 lift from the same blind spot.",
    "prediction": {
      "direction": "positive",
      "size": "Rival holds: A/E 1.25 or higher, both periods (split at 27 April 2026) above 1.05, comparable to the error non-completion control. Error story holds: A/E 1.10 or lower.",
      "falsified_if": "A/E of 1.10 or lower on the broad rule, or a lift present only after 27 April 2026 (below 1.05 before). Either would leave the error non-completion pool's lift specific to the error."
    },
    "notes": "Kept the sketch, adding only a days_since_last_run IS NOT NULL guard so that a NULL last_position means a non-finish and not a debut. sp_decimal < 17 excludes 16/1 and bigger, in line with the thread-wide finding that outsiders never beat par. remarked_last3 starts 2025-11-01, so 300+ bets may not be reached. A thin result in the right direction goes to falsification and forward tracking rather than being killed. Pass criteria are left at the floors."
  }

Timeline

1. Job 2379 finished

   17 Oct, 01:59 · engine:compute-1

2. H-b4454f9c Discovery → KILLED fails: A/E 0.715, needs 1.05; p 0.9515, needs 0.05 or lower; periods [0.74, 0.67], each needs 1.0; A/E 0.715 against same-race control 1.06, needs 0.03 above (not a niche pass: A/E 0.715 below 1.05; p 0.9514758984503698 above 0.5; does not beat the same-race control 1.06; a period points the other way [0.74, 0.67])

   17 Oct, 01:59 · engine:evaluator

3. discovery result for H-b4454f9c (broad rule): fail

   17 Oct, 01:59 · engine:evaluator

4. H-b4454f9c Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 01:59 · engine:orchestrator

5. compute-1 took job 2379 (evaluate, attempt 1)

   17 Oct, 01:59 · compute-1

6. Job 2377 finished

   17 Oct, 01:58 · engine:ai-1

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

   17 Oct, 01:58 · engine:scheduler

8. H-b4454f9c Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 01:58 · agent:designer@v1

9. ai-1 took job 2377 (design, attempt 1)

   17 Oct, 01:57 · ai-1

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

    17 Oct, 01:56 · engine:scheduler

11. New idea H-b4454f9c: Rival: last-time chase non-finishers with clean jumping remarks beat their price too (rival-explanation)

    17 Oct, 01:56 · 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.