Fell or unseated while going well: hidden form at normal prices

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-da5bbd88 (Unseated or brought down and back within four weeks: the fresh-horse c) survived falsification: 5 of 6 attacks left it standing; it is frozen for its one vault query. O-bc59f653
   • H-da5bbd88 (Unseated or brought down and back within four weeks: the fresh-horse c) passed discovery as a niche: A/E 1.17 on 355 bets (p 0.188), same-race control 0.98. O-ff73b33f
2. AskedDo unseated runners that were in contention when they departed beat their price at under 16/1, where the plain BD/UR rule does not? answered
3. MechanismA horse travelling well when it fell or unseated ran to a level the result never recorded. Its true form is better than the line shows, and if it jumped clear before, it is not a chronic faller.
4. Why the market misses itThe result shows F or UR with no position, so a horse that was likely to place or win gets the same reading as one that was tailed off. Comment-reading is patchy, and the price at normal odds reflects the blank.
5. Predictedpositive, A/E 1.08 to 1.15 on 150+ bets at 14/1 or shorter
6. Falsified ifA/E below 1.03, or no better than the capped BD/UR rule's 1.021
7. Rule sketchlast_fell_going_well = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 15
8. question id

The question's answer so far: Not robustly. Going-well departures at 14/1 or shorter scored 1.099 on 245 but were falsified (second period 0.96, Ireland 0.80), and the not-going-well complement did as well.

Mechanism and rationale

A horse travelling well when it fell or unseated ran to a level the result never recorded. Its true form is better than the line shows, and if it jumped clear before, it is not a chronic faller.

The rule

last_fell_going_well = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 15

Reads: last_fell_going_well, race_type, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

30d64f41cbd51a19ad65e3c5d3005e69af48639bb935d222cbdba033d6d87d35

Frozen

not frozen

The definition can still change.

Confounders

• The flag exists only from 2025-11-01 (GB and IRE), so discovery covers about one jumps season and may fall short of 150 bets. A thin result that points the right way goes to falsification and forward tracking, not to a loosened bar.
• The meaning of last_fell_going_well is given only as 'fell going well'. It may leave out unseated riders or be read loosely from the comments. Refinement 1 checks for a real F/UR departure.
• A fall can leave physical or confidence damage, which works against the mechanism and may show up as more jumping errors next time.
• Chronic fallers: one 'going well' fall says nothing about a horse that keeps making errors. Refinement 2 uses jumping_errors_last3, whose definition (including whether it counts the last fall) is not documented.
• Trainer and yard quality: good yards may run horses back quickly after a fall and also have more horses travelling well.
• The market may already read race comments ('fell when challenging'), in which case the price reflects the hidden form and A/E sits near 1.0.
• The 15.0 cap is a pre-off filter on the SP. It matches the capped BD/UR rival (H-799d5af4) so the two can be compared directly.

Discovery result

pass broad rule (niche) · 16 Oct, 13:28

Bets

245

42 winners

A/E

1.10

95%: 0.79 to 1.49

SP return

-8.9%

level stakes

Betfair SP return

+12.1%

245 bets with a BSP

Strike rate 17.1%. p = 0.29. Same-race control: the other runners in these races had A/E 0.98, so the selection beat its own races. Zone: discovery. Evaluator eval-1.0, data 6a4e98be18e83933

Falsification

Each attack is a deliberate attempt to kill the idea. 6 of 6 survived.

• perturbation: Price cap tightened to 11/1 (sp_decimal <= 12)pass

  204 bets · A/E 1.11 (0.79 to 1.51) · control 0.98 · 16 Oct, 13:34

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271321.15

  after 2026-04-27721.02

• perturbation: Price cap loosened to 17/1 (sp_decimal <= 18)pass

  258 bets · A/E 1.14 (0.82 to 1.52) · control 0.98 · 16 Oct, 13:34

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.

Other experiments

• neighbour: <= 15 -> 12 (-20%)pass

  204 bets · A/E 1.11 (0.79 to 1.51) · control 0.98 · 16 Oct, 13:28

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271321.15

  after 2026-04-27721.02

• neighbour: <= 15 -> 18 (+20%)pass

  258 bets · A/E 1.14 (0.82 to 1.52) · control 0.98 · 16 Oct, 13:28

  Breakdown

Lineage

Agent runs

• designer v1ok · 45s · 16 Oct, 13:26

  Output (schema-checked)

  {
    "rule": "last_fell_going_well = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 15",
    "refinements": [
      {
        "label": "departure confirmed as F or UR",
        "rule": "last_fell_going_well = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 15 AND last_non_finish_code IN ('F', 'UR')"
      },
      {
        "label": "not a chronic faller (at most the one error in last three)",
        "rule": "last_fell_going_well = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 15 AND jumping_errors_last3 <= 1"
      }
    ],
    "control": {
      "description": "Same-race other runners should show about 1.0, so any lift is not a property of the race. The comparison that matters is the other half of the F/UR group: capped jumps runners that fell or unseated last time without the going-well flag. On the capped BD/UR result (1.021) they should sit near 1.0. The hypothesis needs the flagged group to beat this control by about 0.05 A/E or more. If both groups show a similar lift, the effect belongs to the F/UR blank in general and the comment information adds nothing.",
      "rule": "last_non_finish_code IN ('F', 'UR') AND last_fell_going_well = 0 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 15"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "The flag exists only from 2025-11-01 (GB and IRE), so discovery covers about one jumps season and may fall short of 150 bets. A thin result that points the right way goes to falsification and forward tracking, not to a loosened bar.",
      "The meaning of last_fell_going_well is given only as 'fell going well'. It may leave out unseated riders or be read loosely from the comments. Refinement 1 checks for a real F/UR departure.",
      "A fall can leave physical or confidence damage, which works against the mechanism and may show up as more jumping errors next time.",
      "Chronic fallers: one 'going well' fall says nothing about a horse that keeps making errors. Refinement 2 uses jumping_errors_last3, whose definition (including whether it counts the last fall) is not documented.",
      "Trainer and yard quality: good yards may run horses back quickly after a fall and also have more horses travelling well.",
      "The market may already read race comments ('fell when challenging'), in which case the price reflects the hidden form and A/E sits near 1.0.",
      "The 15.0 cap is a pre-off filter on the SP. It matches the capped BD/UR rival (H-799d5af4) so the two can be compared directly."
    ],
    "mechanism": "A horse that was travelling well, or in contention, when it fell or unseated ran to a level its result never records. On the run it really put up, its form is better than its blank line. If its record shows clean jumping otherwise, the departure was a one-off and not chronic. When it next runs, it should run closer to that hidden level than the market expects.",
    "market_miss": "The result line shows F or UR with no position. A horse that was likely to place or win gets the same mark as one that was tailed off. Form figures and most quick price-making are built on positions, and comment-reading is patchy. So at normal prices (14/1 or shorter) the SP is set partly on the blank and undervalues the flagged horse. Earlier lab work found the general BD/UR edge only at longshot prices (capped: 1.021 on 312). This hypothesis says the edge at normal prices sits in the 'going well' subset.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.08 to 1.15 at 14/1 or shorter (SP 15.0 or less), on jumps runners flagged fell going well last time",
      "falsified_if": "A/E below 1.03, or no better than the capped BD/UR rule's 1.021 / the not-going-well F/UR control (flagged group fails to beat the control by at least 0.03 A/E)"
    },
    "notes": "The rule is the investigator's sketch, kept because it is sound. Every column is pre_off. The flag is available from 2025-11-01 in both regions, which caps the sample, so a sub-150 discovery count is expected and should be passed on for forward tracking, not loosened. The forward bar is stricter than the floor (1.03 on A/E, not 1.0) because the falsification line sits at 1.03 and the capped BD/UR baseline is about 1.02. Each refinement is only looked at if the broad rule shows evidence."
  }

Timeline

1. Job 1630 finished

   16 Oct, 13:34 · engine:compute-1

2. H-9f384f7c Falsification → KILLED falsified: period after 2026-04-27 A/E 0.96; IRE A/E 0.802; driven by one price band (up to 7/1): the rest score 0.97

   16 Oct, 13:34 · engine:falsification

3. falsification result for H-9f384f7c (partition: Handicaps only): pass

   16 Oct, 13:34 · engine:evaluator

4. compute-1 took job 1630 (attack, attempt 1)

   16 Oct, 13:34 · compute-1

5. Job 1629 finished

   16 Oct, 13:34 · engine:compute-1

6. falsification result for H-9f384f7c (partition: Chases only): pass

   16 Oct, 13:34 · engine:evaluator

7. compute-1 took job 1629 (attack, attempt 1)

   16 Oct, 13:34 · compute-1

8. Job 1628 finished

   16 Oct, 13:34 · engine:compute-1

9. falsification result for H-9f384f7c (partition: Departure confirmed as F or UR): pass

   16 Oct, 13:34 · engine:evaluator

10. compute-1 took job 1628 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

11. Job 1627 finished

    16 Oct, 13:34 · engine:compute-1

12. falsification result for H-9f384f7c (control: Control: F/UR last time without the going-well flag, capped): pass

    16 Oct, 13:34 · engine:evaluator

13. compute-1 took job 1627 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

14. Job 1626 finished

    16 Oct, 13:34 · engine:compute-1

15. falsification result for H-9f384f7c (perturbation: Price cap loosened to 17/1 (sp_decimal <= 18)): pass

    16 Oct, 13:34 · engine:evaluator

16. compute-1 took job 1626 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

17. Job 1625 finished

    16 Oct, 13:34 · engine:compute-1

18. falsification result for H-9f384f7c (perturbation: Price cap tightened to 11/1 (sp_decimal <= 12)): pass

    16 Oct, 13:34 · engine:evaluator

19. compute-1 took job 1625 (attack, attempt 1)

    16 Oct, 13:34 · compute-1

20. Job 1611 finished

    16 Oct, 13:34 · engine:ai-1