Generalise: a fall or unseat followed by a switch of race code

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-070b2026 (Rival: held-up PU lift comes from a switch of race code since the pull) survived falsification: 6 of 6 attacks left it standing; it is frozen for its one vault query. O-4aec4e49
   • H-070b2026 (Rival: held-up PU lift comes from a switch of race code since the pull) passed discovery as a niche: A/E 1.36 on 304 bets (p 0.092), same-race control 0.98. O-9a6df59c
2. AskedDoes any last-time non-completion lose its meaning once the horse switches code, or is the effect specific to a held-up pull-up? testing
3. MechanismA non-completion in one code says little about ability in another, especially chase falls followed by a drop back to hurdles, where the obstacle that caused the failure is removed.
4. Why the market misses itForm readers treat F or U as a black mark whatever the code, so the horse drifts in a discipline where the failure does not apply.
5. Predictedpositive, A/E 1.10 to 1.30 on 250+ bets
6. Falsified ifA/E at or below 1.05, which would make the held-up pull-up the essential part
7. Rule sketchlast_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type <> last_race_type

Mechanism and rationale

A non-completion in one code says little about ability in another, especially chase falls followed by a drop back to hurdles, where the obstacle that caused the failure is removed.

The rule

last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type <> last_race_type

Reads: last_fell_or_unseated, last_race_type, race_type

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

2db4e3f1f2fbd77e0993f5f924a480bcbb6d95d9d2776b565bb1442a6fd9c268

Frozen

not frozen

The definition can still change.

Confounders

• Fallers going well when they came down may already be backed, so the market may have priced the excuse (last_fell_going_well)
• Same-code F/U non-finishers may be over-bet (error-free chase non-finishers scored 0.71), so the comparison with the control may flatter the switch group
• Irish non-finishers sat below par in the chase-error thread; a GB/IRE mix could hide or dilute the effect
• Outsiders at 16/1 or bigger never beat par in this thread; a long-priced tail could drag the A/E down
• last_fell_or_unseated exists only from 2025-11-01, so the window is short and one season; a late-period skew (seen around 27 April 2026) could produce a calendar effect
• Code switches to Flat or Bumper are a different population (dual-purpose or lightly raced horses) from Chase/Hurdle switches
• Hurdle-to-chase switchers after a hurdle fall are often novice chasers with stable intent behind the move, separate from the removed-obstacle story

Control

{
  "description": "Same-race rivals should sit near par (A/E about 1.0); if they show a similar lift the effect belongs to the race type, not the horse. The key separating comparison is last-time fallers/unseaters who stay in the same code: the mechanism says the F/U keeps its meaning there, so they should be at or below par (the lab found error-free chase non-finishers over-bet at 0.71). A switch-group lift with a same-code control also lifted would mean the market simply over-penalises every F/U, not that the code switch matters.",
  "rule": "last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type = last_race_type"
}

Discovery result

fail broad rule · 17 Oct, 02:09

Bets

188

11 winners

A/E

0.75

95%: 0.38 to 1.35

SP return

-55.5%

level stakes

Betfair SP return

-45.0%

187 bets with a BSP

Strike rate 5.9%. p = 0.86. Same-race control: the other runners in these races had A/E 1.02, 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 · 48s · 17 Oct, 02:04

  Output (schema-checked)

  {
    "rule": "last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type <> last_race_type",
    "refinements": [
      {
        "label": "Chase faller dropping back to hurdles",
        "rule": "last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type <> last_race_type AND last_race_type = 'Chase' AND race_type = 'Hurdle'"
      },
      {
        "label": "14/1 or shorter (outsiders never beat par in this thread)",
        "rule": "last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type <> last_race_type AND sp_decimal <= 15"
      },
      {
        "label": "GB only (Irish non-finishers sat below par in the chase-error thread)",
        "rule": "last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type <> last_race_type AND region = 'GB'"
      }
    ],
    "control": {
      "description": "Same-race rivals should sit near par (A/E about 1.0); if they show a similar lift the effect belongs to the race type, not the horse. The key separating comparison is last-time fallers/unseaters who stay in the same code: the mechanism says the F/U keeps its meaning there, so they should be at or below par (the lab found error-free chase non-finishers over-bet at 0.71). A switch-group lift with a same-code control also lifted would mean the market simply over-penalises every F/U, not that the code switch matters.",
      "rule": "last_fell_or_unseated = 1 AND last_race_type IS NOT NULL AND race_type = last_race_type"
    },
    "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": [
      "Fallers going well when they came down may already be backed, so the market may have priced the excuse (last_fell_going_well)",
      "Same-code F/U non-finishers may be over-bet (error-free chase non-finishers scored 0.71), so the comparison with the control may flatter the switch group",
      "Irish non-finishers sat below par in the chase-error thread; a GB/IRE mix could hide or dilute the effect",
      "Outsiders at 16/1 or bigger never beat par in this thread; a long-priced tail could drag the A/E down",
      "last_fell_or_unseated exists only from 2025-11-01, so the window is short and one season; a late-period skew (seen around 27 April 2026) could produce a calendar effect",
      "Code switches to Flat or Bumper are a different population (dual-purpose or lightly raced horses) from Chase/Hurdle switches",
      "Hurdle-to-chase switchers after a hurdle fall are often novice chasers with stable intent behind the move, separate from the removed-obstacle story"
    ],
    "mechanism": "A fall or unseat is an obstacle-specific failure. It says little about ability once the horse runs in a different code, and least of all when a chase faller drops back to hurdles (or goes to the Flat), where the fence that caused the failure is gone. The question is whether the code-switch lift seen in held-up pull-ups extends to any non-completion, or whether the held-up pull-up is the essential part.",
    "market_miss": "Form readers and the market treat an F or U in the form figures as a black mark whatever the code. The horse drifts in a discipline where the failure does not apply, and its underlying ability is under-priced.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.30 on 250+ bets",
      "falsified_if": "A/E at or below 1.05 on the broad rule, or a same-code faller control lifted about as much, which would make the held-up pull-up (not the switch after any non-completion) the essential part"
    },
    "notes": "Kept the rule sketch: it is broad, uses only pre-off columns and matches the generalise question. last_fell_or_unseated starts 2025-11-01, so the sample may be thin. If it is, the lab should send it to falsification and forward tracking rather than kill it for size. The refinements come straight from the reasoning (chase-to-hurdle obstacle removal, the outsider price pattern, GB vs IRE). They are read only if the broad rule shows evidence."
  }

Timeline

1. Job 2411 finished

   17 Oct, 02:09 · engine:compute-1

2. H-525d10d2 Discovery → KILLED fails: A/E 0.752, needs 1.05; p 0.8622, needs 0.05 or lower; periods [0.69, 0.85], each needs 1.0; A/E 0.752 against same-race control 1.019, needs 0.03 above (not a niche pass: A/E 0.752 below 1.05; p 0.8621942564032626 above 0.5; does not beat the same-race control 1.019; a period points the other way [0.69, 0.85])

   17 Oct, 02:09 · engine:evaluator

3. discovery result for H-525d10d2 (broad rule): fail

   17 Oct, 02:09 · engine:evaluator

4. H-525d10d2 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 02:09 · engine:orchestrator

5. compute-1 took job 2411 (evaluate, attempt 1)

   17 Oct, 02:09 · compute-1

6. Job 2408 finished

   17 Oct, 02:05 · engine:ai-1

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

   17 Oct, 02:05 · engine:scheduler

8. H-525d10d2 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 02:05 · agent:designer@v1

9. ai-1 took job 2408 (design, attempt 1)

   17 Oct, 02:04 · ai-1

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

    17 Oct, 02:03 · engine:scheduler

11. New idea H-525d10d2: Generalise: a fall or unseat followed by a switch of race code (generalise)

    17 Oct, 02:03 · 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.