Generalise: hurdle runners that won last time with a bottom-third Jump Index 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-95f8568d (Test-mechanism: the poor-jumping deficit should be absent in last-time) pointed the predicted way but failed the bar: A/E 1.05 on 902 bets (p 0.263), control 0.98. O-205a18e1
   • H-bfcb9669 (Chase-bound winners that jumped poorly when winning are over-bet) pointed the predicted way but failed the bar: A/E 0.78 on 240 bets (p 0.951), control 1.07. O-7acbc407
2. AskedDo last-time winners with a bottom-third Jump Index also run below their price over hurdles, which would give a larger pool to test the hidden-poor-jumping idea? answered
3. MechanismA horse that jumped poorly but still won had the race fall its way. The poor jumping costs ground again next time, less over hurdles than over fences but still some.
4. Why the market misses itThe form book shows a win, and comments rarely record small losses at each obstacle, so the market prices the win and not the jumping figure.
5. Predictednegative, A/E 0.85 to 0.95 on 400+ bets
6. Falsified ifA/E 0.98 or higher on 400+ bets, or no lower than hurdle runners that won last time with last_jump_index_pct >= 0.67
7. Rule sketchrace_type = 'Hurdle' AND last_position = 1 AND last_race_type IN ('Chase', 'Hurdle') AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct <= 0.33

Mechanism and rationale

A horse that jumped poorly but still won had the race fall its way. The poor jumping costs ground again next time, less over hurdles than over fences but still some.

The rule

race_type = 'Hurdle' AND last_position = 1 AND last_race_type IN ('Chase', 'Hurdle') AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct <= 0.33

Reads: race_type, last_position, last_race_type, last_jump_index_pct

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

4d5c6ea68d5cf0cead43426cf4df49477c359ab9abbb3eae683aea2c0620d4af

Frozen

not frozen

The definition can still change.

Confounders

• Favourite-longshot bias: the chase version's deficit sat at 16/1+ (0 from 20), so a deficit here may just be ordinary longshot under-performance. Check it against the same-race control by price band.
• Novice versus handicap mix: winners of novice hurdles who go up in grade, and handicap winners who are raised in the weights, may both run below their price for reasons unrelated to jumping.
• Winners' rise in the weights: the handicap mark goes up after a win. Earlier pools suggest this is weak (raised 0.84 vs unraised 0.82), but it is still a possible cause.
• Generic low-figure effect: slow-top-speed winners and losers both scored about 0.96. A poor jump figure may only be standing in for a weak overall performance.
• Country split: earlier jumping results differed sharply between GB and IRE (chase IRE 0.45 on a small sample). Results must point the same way in both countries.
• Code mismatch: a jump figure from a chase run used for a hurdle run measures a different skill.

Control

{
  "description": "Primary: the other runners in the same hurdle races should score about A/E 1.0, which rules out a race-level property. Second control: hurdle runners that won last time with a top-third Jump Index (>= 0.67). They show whether the deficit comes from the jumping figure or is just a general last-time-winner effect. The hypothesis needs the poor-jumping winners to score clearly below this group, at least 0.05 A/E lower.",
  "rule": "race_type = 'Hurdle' AND last_position = 1 AND last_race_type IN ('Chase', 'Hurdle') AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct >= 0.67"
}

Discovery result

fail broad rule · 16 Oct, 16:56

Bets

396

84 winners

A/E

1.07

95%: 0.86 to 1.33

SP return

+1.7%

level stakes

Betfair SP return

+29.2%

382 bets with a BSP

Strike rate 21.2%. p = 0.27. 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

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 · 30s · 16 Oct, 16:55

  Output (schema-checked)

  {
    "rule": "race_type = 'Hurdle' AND last_position = 1 AND last_race_type IN ('Chase', 'Hurdle') AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct <= 0.33",
    "refinements": [
      {
        "label": "Last run over hurdles",
        "rule": "race_type = 'Hurdle' AND last_position = 1 AND last_race_type = 'Hurdle' AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct <= 0.33"
      },
      {
        "label": "Bigger than evens",
        "rule": "race_type = 'Hurdle' AND last_position = 1 AND last_race_type IN ('Chase', 'Hurdle') AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct <= 0.33 AND sp_decimal > 2.0"
      }
    ],
    "control": {
      "description": "Primary: the other runners in the same hurdle races should score about A/E 1.0, which rules out a race-level property. Second control: hurdle runners that won last time with a top-third Jump Index (>= 0.67). They show whether the deficit comes from the jumping figure or is just a general last-time-winner effect. The hypothesis needs the poor-jumping winners to score clearly below this group, at least 0.05 A/E lower.",
      "rule": "race_type = 'Hurdle' AND last_position = 1 AND last_race_type IN ('Chase', 'Hurdle') AND last_jump_index_pct IS NOT NULL AND last_jump_index_pct >= 0.67"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Favourite-longshot bias: the chase version's deficit sat at 16/1+ (0 from 20), so a deficit here may just be ordinary longshot under-performance. Check it against the same-race control by price band.",
      "Novice versus handicap mix: winners of novice hurdles who go up in grade, and handicap winners who are raised in the weights, may both run below their price for reasons unrelated to jumping.",
      "Winners' rise in the weights: the handicap mark goes up after a win. Earlier pools suggest this is weak (raised 0.84 vs unraised 0.82), but it is still a possible cause.",
      "Generic low-figure effect: slow-top-speed winners and losers both scored about 0.96. A poor jump figure may only be standing in for a weak overall performance.",
      "Country split: earlier jumping results differed sharply between GB and IRE (chase IRE 0.45 on a small sample). Results must point the same way in both countries.",
      "Code mismatch: a jump figure from a chase run used for a hurdle run measures a different skill."
    ],
    "mechanism": "A horse that jumped poorly (bottom-third Jump Index) but still won probably had the race fall its way, through pace, rivals' mistakes or a weak field. Its true ability is lower than the win suggests. Poor jumping is a fairly stable trait, so it costs ground again next time. Hurdles cost less per mistake than fences, so the deficit should be smaller than the chase figure (0.78) but still there.",
    "market_miss": "The form book and the market price the win itself. Race comments rarely record small losses of momentum at each obstacle, and Jump Index sectional data is not widely priced in. So the market does not mark down a winner that jumped poorly.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.85 to 0.95 on 400+ bets, with the same-race control near 1.0",
      "falsified_if": "A/E 0.98 or higher on 400+ bets, or A/E no lower than the top-third Jump Index hurdle winner control (>= 0.67), or the deficit appears in only one country"
    },
    "notes": "I kept the rule sketch exactly as written. It mirrors the chase rule H-bfcb9669 (0.78 on 240, p 0.067) and uses only pre-off columns already proven in earlier rules. In this negative direction, the 1.05 A/E floor is applied to the inverse (the lab's convention for over-bet rules). I raised min_bets to 300 because the hurdle pool should be much larger than the chase pool and the prediction is for 400+ bets. There are two refinements. The first keeps only runs whose last run was over hurdles, so the jump figure comes from the same code. The second drops runners at evens or shorter, because short-priced class horses sat at par in the chase pool."
  }

Timeline

1. Job 2072 finished

   16 Oct, 16:56 · engine:compute-1

2. H-cf981b01 Discovery → KILLED fails: A/E 1.074, needs 0.95 or lower; p 0.7634 (lower tail), needs 0.05; periods [1.07, 1.08], each needs 0.99 or lower; A/E 1.074 against same-race control 0.983, needs 0.03 below

   16 Oct, 16:56 · engine:evaluator

3. discovery result for H-cf981b01 (broad rule): fail

   16 Oct, 16:56 · engine:evaluator

4. H-cf981b01 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 16:56 · engine:orchestrator

5. compute-1 took job 2072 (evaluate, attempt 1)

   16 Oct, 16:56 · compute-1

6. Job 2070 finished

   16 Oct, 16:55 · engine:ai-1

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

   16 Oct, 16:55 · engine:scheduler

8. H-cf981b01 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 16:55 · agent:designer@v1

9. ai-1 took job 2070 (design, attempt 1)

   16 Oct, 16:55 · ai-1

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

    16 Oct, 16:55 · engine:scheduler

11. New idea H-cf981b01: Generalise: hurdle runners that won last time with a bottom-third Jump Index are over-bet (generalise)

    16 Oct, 16:55 · 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.