Test-mechanism: chase fallers that drift from their last SP also beat their price

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-f5df804d (Test-mechanism: chase unseaters that drift from their last SP carry th) passed discovery as a niche: A/E 2.1 on 110 bets (p 0.004), same-race control 0.91. O-1c66e5b5
2. AskedDoes the 'unearned discount for a blank figure' extend to other chase non-finishes (F, PU) that drift from their last SP, or is it unique to UR? answered
3. MechanismAn F, like a U, leaves no finishing position to rate, and casual punters and compilers lengthen the price. A fall says more about the horse's jumping than an unseat does, so part of the discount is earned. The lift should be present but smaller than for unseaters.
4. Why the market misses itThe form line shows F with no position or beaten distance. A drift driven by the missing figure, and not by today's race, is a discount for a run that was often competitive until the fall.
5. Predictedpositive, A/E 1.15 to 1.40 on 80 to 120 bets, below the unseater drifters' 2.10
6. Falsified ifDrifting chase fallers score A/E at or below 1.02, or no higher than fallers that shortened. That would mean the drift discount is specific to UR.
7. Rule sketchrace_type = 'Chase' AND last_non_finish_code = 'F' AND sp_decimal > last_sp_decimal
8. question idQ-069480c7

The question's answer so far: No, the drift lift does not extend to other chase non-finishes. Drifting fallers scored 1.058 on 130, below fallers that shortened (1.159), and drifting pulled-up chasers scored 0.963 on 363 against a control of 1.012. The drift effect is specific to unseaters.

Mechanism and rationale

An F, like a U, leaves no finishing position to rate, and casual punters and compilers lengthen the price. A fall says more about the horse's jumping than an unseat does, so part of the discount is earned. The lift should be present but smaller than for unseaters.

The rule

race_type = 'Chase' AND last_non_finish_code = 'F' AND sp_decimal > last_sp_decimal

Reads: race_type, last_non_finish_code, sp_decimal, last_sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

db92bec0afd5be60f55195dc63ba87741c561f1e5c621665ac7bde69a309172e

Frozen

not frozen

The definition can still change.

Confounders

• Drifting fallers may have a real loss of jumping confidence that the market prices correctly, so part of the discount is earned (this is why the predicted lift is smaller than for unseaters)
• Change in class or field strength today: a drift can reflect a harder race, not the blank form figure
• Thin pool carried by a few long-priced winners. The earlier going-well faller result (1.256 on 85) rested on 2 long-priced winners, so check the 16/1-or-shorter refinement and the win count
• Selection on the same family: the drift split was found on unseaters after several UR cuts, so the faller test is the out-of-sample check of that split and should be read on its own
• last_non_finish_code is missing from the vault, so a vault verdict carries no information; the real test is forward tracking
• Rows with NULL last_sp_decimal drop out silently through the comparison, which may bias the pool towards horses with fully recorded histories
• Short discovery window for non-finish codes: results by period and by country should agree

Control

Discovery result

pass broad rule (niche) · 17 Oct, 03:42

Bets

130

13 winners

A/E

1.06

95%: 0.56 to 1.81

SP return

+14.4%

level stakes

Betfair SP return

+91.5%

130 bets with a BSP

Strike rate 10.0%. p = 0.46. Same-race control: the other runners in these races had A/E 1.01, 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. 2 of 6 survived.

• perturbation: Perturbation: meaningful drift only (SP at least 1.2x last SP)pass

  115 bets · A/E 1.01 (0.48 to 1.85) · control 1.01 · 17 Oct, 03:50

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27791.05

  after 2026-04-2736–

• price_band: Price-band: drifting fallers 16/1 or shorterfail

  82 bets · A/E 0.99 (0.49 to 1.78) · control 1.02 · 17 Oct, 03:50

  Breakdown

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: drop: race_type = 'Chase'fail

  348 bets · A/E 0.80 (0.49 to 1.24) · control 1.02 · 17 Oct, 03:50

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272130.97

  after 2026-04-271350.47

• neighbour: drop: last_non_finish_code = 'F'fail

  6,218 bets · A/E 0.99 (0.91 to 1.08) · control 1.01 · 17 Oct, 03:50

  Breakdown

Lineage

Agent runs

• designer v1ok · 29s · 17 Oct, 03:40

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND last_non_finish_code = 'F' AND sp_decimal > last_sp_decimal",
    "refinements": [
      {
        "label": "16/1 or shorter (guards against a few long-priced winners)",
        "rule": "race_type = 'Chase' AND last_non_finish_code = 'F' AND sp_decimal > last_sp_decimal AND sp_decimal < 17"
      },
      {
        "label": "Fell when going well (the run was competitive until the fall)",
        "rule": "race_type = 'Chase' AND last_non_finish_code = 'F' AND sp_decimal > last_sp_decimal AND last_fell_going_well = 1"
      }
    ],
    "control": {
      "description": "Main comparison: last-time chase fallers whose SP shortened or held against their last SP. The hypothesis is falsified if drifters score no higher than these, and the UR split gives a guide (0.95 for unseaters that shortened or held). The same-race other runners should sit near 1.0. If they are well above 1.0, the lift belongs to the race and not the horse. A further reference is all chase drifters, at 0.995 on 6218, which shows drift alone carries no edge.",
      "rule": "race_type = 'Chase' AND last_non_finish_code = 'F' AND sp_decimal <= last_sp_decimal"
    },
    "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": [
      "Drifting fallers may have a real loss of jumping confidence that the market prices correctly, so part of the discount is earned (this is why the predicted lift is smaller than for unseaters)",
      "Change in class or field strength today: a drift can reflect a harder race, not the blank form figure",
      "Thin pool carried by a few long-priced winners. The earlier going-well faller result (1.256 on 85) rested on 2 long-priced winners, so check the 16/1-or-shorter refinement and the win count",
      "Selection on the same family: the drift split was found on unseaters after several UR cuts, so the faller test is the out-of-sample check of that split and should be read on its own",
      "last_non_finish_code is missing from the vault, so a vault verdict carries no information; the real test is forward tracking",
      "Rows with NULL last_sp_decimal drop out silently through the comparison, which may bias the pool towards horses with fully recorded histories",
      "Short discovery window for non-finish codes: results by period and by country should agree"
    ],
    "mechanism": "An F, like a UR, leaves no finishing position or beaten distance to rate. Casual punters and odds-compilers fill that blank cautiously and lengthen the price. When the drift from the last SP comes from the missing figure and not from today's race, it is a mispricing. A fall says more about the horse's jumping than an unseat does, so part of the discount is earned. The lift should therefore be present but smaller than the unseater drifters' 2.10.",
    "market_miss": "The form line shows 'F' with no position or beaten distance, and ratings-based compilers and casual money treat it as a bad run. Many fallers were competitive until the fall. A drift against the last SP that comes from this missing information, and not from today's conditions, leaves a discount the market fails to correct.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.15 to 1.40 on roughly 80 to 120 bets, clearly below the unseater drifters' 2.10",
      "falsified_if": "Drifting chase fallers score A/E at or below 1.02, or no higher than chase fallers that shortened or held (the control). Either result would mean the drift discount is specific to unseaters."
    },
    "notes": "The rule sketch is sound: all columns are pre-off and nothing is sliced beyond the mechanism, so it is kept unchanged. The expected pool (80 to 120) is below the 150-bet discovery floor. A thin result in the right direction goes on to falsification and forward tracking rather than being killed. The vault cannot evaluate last_non_finish_code rules (H-4cf8f914 returned 0 bets), so forward tracking is the real test. Refinements are looked at only if the broad rule shows evidence."
  }

Timeline

1. Job 2749 finished

   17 Oct, 03:50 · engine:compute-1

2. hypothesis note H-3ea4ba62

   17 Oct, 03:50 · engine:perturb

3. perturbation result for H-3ea4ba62 (neighbour: drop: sp_decimal > last_sp_decimal): pass

   17 Oct, 03:50 · engine:perturb

4. perturbation result for H-3ea4ba62 (neighbour: drop: last_non_finish_code = 'F'): fail

   17 Oct, 03:50 · engine:perturb

5. perturbation result for H-3ea4ba62 (neighbour: drop: race_type = 'Chase'): fail

   17 Oct, 03:50 · engine:perturb

6. compute-1 took job 2749 (mutate, attempt 1)

   17 Oct, 03:50 · compute-1

7. Job 2766 finished

   17 Oct, 03:50 · engine:compute-1

8. H-3ea4ba62 Falsification → KILLED falsified: GB A/E 0.704; driven by one price band (16/1 or bigger): the rest score 0.99; 3 of 5 attacks broke it

   17 Oct, 03:50 · engine:falsification

9. falsification result for H-3ea4ba62 (country: Country: GB drifting fallers 16/1 or shorter): fail

   17 Oct, 03:50 · engine:evaluator

10. compute-1 took job 2766 (attack, attempt 1)

    17 Oct, 03:50 · compute-1

11. Job 2765 finished

    17 Oct, 03:50 · engine:compute-1

12. falsification result for H-3ea4ba62 (partition: Partition: handicap chases only): pass

    17 Oct, 03:50 · engine:evaluator

13. compute-1 took job 2765 (attack, attempt 1)

    17 Oct, 03:50 · compute-1

14. Job 2764 finished

    17 Oct, 03:50 · engine:compute-1

15. falsification result for H-3ea4ba62 (partition: Partition: fell when going well, among drifters): info

    17 Oct, 03:50 · engine:evaluator

16. compute-1 took job 2764 (attack, attempt 1)

    17 Oct, 03:50 · compute-1

17. Job 2763 finished

    17 Oct, 03:50 · engine:compute-1

18. falsification result for H-3ea4ba62 (control: Control: drifting chase fallers vs same-race non-fallers who drifted): fail

    17 Oct, 03:50 · engine:evaluator