Mid-priced jumps departures (8/1 to 14/1) are over-bet on the 'unlucky last time' story

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-49fa55f7 (Jumps fallers at 14/1 or shorter beat their price: the market over-pen): the country 'IRE' carries the effect (A/E 1.21) while the rest score GB 1.0. O-7fe21bbd
   • H-49fa55f7 (Jumps fallers at 14/1 or shorter beat their price: the market over-pen) passed discovery as a niche: A/E 1.07 on 389 bets (p 0.302), same-race control 0.99. O-ea451163
   • H-e7675af7 (Going-well departure with weak visible form: the hidden form is larges) passed discovery but was falsified: IRE A/E 0.96; driven by one price band (up to 7/1): the rest score 0.77; 2 of 5 attacks broke it. O-fb274822
2. AskedAre jumps runners that fell, unseated or were brought down last time over-bet in the 8/1 to 14/1 band, while those up to 7/1 match or beat their price? answered
3. MechanismA fall or unseat lets backers imagine the horse would have been placed. Public and each-way money therefore shortens mid-priced departures, whose price is set by public money more than by informed money. Fancied departures (up to 7/1) are priced by connections' confidence. Longshots get no such story.
4. Why the market misses itThe F or U in the form figures is read as hidden form ('was going well'), but most departures in this band had no real chance. Punters over-trust the narrative, so these runners start shorter than their true chance.
5. Predictednegative, A/E 0.75 to 0.90 on 300+ bets
6. Falsified ifA/E 0.97 or higher, or either period above 1.0
7. Rule sketch

The question's answer so far: Not supported. Pooled jumps departures at 8/1 to 14/1 scored A/E 0.82 on 280 bets (CI 0.47 to 1.34, p 0.81, control 1.02). The earlier 0.395 and 0.579 band figures were small segments chosen because they looked extreme, and they shrank toward 1 when pooled.

Mechanism and rationale

A fall or unseat lets backers imagine the horse would have been placed. Public and each-way money therefore shortens mid-priced departures, whose price is set by public money more than by informed money. Fancied departures (up to 7/1) are priced by connections' confidence. Longshots get no such story.

The rule

last_fell_or_unseated = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal > 8 AND sp_decimal <= 15

Reads: last_fell_or_unseated, race_type, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

595a6720d23eca8eb8928b8d4f35896c2986e9f061cd4291e548369614099ef6

Frozen

not frozen

The definition can still change.

Confounders

• Field size: mid-priced runners cluster in big handicaps. The each-way refinement and the same-band non-departure control separate this.
• Price-band bias: A/E at 15/2 to 14/1 may be low for every runner. Answered by the same-band control (last_fell_or_unseated = 0).
• Sample overlap with H-49fa55f7 (fallers <=14/1, A/E 1.07) and H-9f384f7c/H-9519a139 (F/U <=14/1, 1.10 and 1.13). The combined 14/1-or-shorter F/U pool already beat its price, so a negative result here needs the lift to sit wholly below 15/2.
• Novice chases with high fall rates: a repeat fall risk could produce a low A/E through jumping, not through the narrative.
• Irish big-field handicaps: earlier family rules split by country in opposite directions, so a country concentration is not read as mechanism.
• Thin winners: at about 10/1 average, 300 bets give only about 25 to 30 expected winners. The period split is required.

Control

{
  "description": "Same-race runners should score about 1.0, so a low A/E is not just a feature of the race. Mid-price bands can carry their own favourite-longshot bias, so the main control is jumps runners at the same 15/2 to 14/1 prices whose last run did NOT end in a fall or unseat. We expect them to score about 0.95 to 1.0. The hypothesis only stands if the departure group scores clearly below this control (by at least 0.07) as well as below 1.0.",
  "rule": "last_fell_or_unseated = 0 AND prior_runs >= 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal > 8 AND sp_decimal <= 15"
}

Discovery result

fail broad rule · 16 Oct, 13:46

Bets

280

16 winners

A/E

0.82

95%: 0.47 to 1.33

SP return

-35.5%

level stakes

Betfair SP return

-13.8%

280 bets with a BSP

Strike rate 5.7%. p = 0.81. 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 · 37s · 16 Oct, 13:42

  Output (schema-checked)

  {
    "rule": "last_fell_or_unseated = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal > 8 AND sp_decimal <= 15",
    "refinements": [
      {
        "label": "each-way handicaps (field 12+)",
        "rule": "last_fell_or_unseated = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal > 8 AND sp_decimal <= 15 AND is_handicap = 1 AND field_size >= 12"
      },
      {
        "label": "story still fresh (back within 60 days)",
        "rule": "last_fell_or_unseated = 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal > 8 AND sp_decimal <= 15 AND days_since_last_run <= 60"
      }
    ],
    "control": {
      "description": "Same-race runners should score about 1.0, so a low A/E is not just a feature of the race. Mid-price bands can carry their own favourite-longshot bias, so the main control is jumps runners at the same 15/2 to 14/1 prices whose last run did NOT end in a fall or unseat. We expect them to score about 0.95 to 1.0. The hypothesis only stands if the departure group scores clearly below this control (by at least 0.07) as well as below 1.0.",
      "rule": "last_fell_or_unseated = 0 AND prior_runs >= 1 AND race_type IN ('Hurdle', 'Chase') AND sp_decimal > 8 AND sp_decimal <= 15"
    },
    "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": [
      "Field size: mid-priced runners cluster in big handicaps. The each-way refinement and the same-band non-departure control separate this.",
      "Price-band bias: A/E at 15/2 to 14/1 may be low for every runner. Answered by the same-band control (last_fell_or_unseated = 0).",
      "Sample overlap with H-49fa55f7 (fallers <=14/1, A/E 1.07) and H-9f384f7c/H-9519a139 (F/U <=14/1, 1.10 and 1.13). The combined 14/1-or-shorter F/U pool already beat its price, so a negative result here needs the lift to sit wholly below 15/2.",
      "Novice chases with high fall rates: a repeat fall risk could produce a low A/E through jumping, not through the narrative.",
      "Irish big-field handicaps: earlier family rules split by country in opposite directions, so a country concentration is not read as mechanism.",
      "Thin winners: at about 10/1 average, 300 bets give only about 25 to 30 expected winners. The period split is required."
    ],
    "mechanism": "A fall or unseat blanks the form line and lets backers imagine a placing ('would have been involved'). In the 15/2 to 14/1 band, prices are formed mostly by public and each-way money rather than by connections' confidence. That narrative money shortens these runners beyond their true chance. Fancied departures (7/1 or shorter) are priced by informed money. Longshots attract no story.",
    "market_miss": "Recreational backers read F/U in the figures as hidden form. But most mid-priced departures were not travelling like winners and carry repeat-error risk. Each-way support in big jumps fields compresses their price, so they start shorter than their win chance.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.75 to 0.90 on 300+ bets; same-band non-departure control near 0.95-1.0",
      "falsified_if": "A/E 0.97 or higher, or either time period above 1.0, or the departure group not at least 0.07 below the same-band non-departure control"
    },
    "notes": "The direction is negative (over-bet), so min_ae 1.05 is meant in the predicted direction, i.e. inverse strength: A/E <= 1/1.05 ≈ 0.95. The forward bar likewise needs A/E <= 1.0 in the predicted direction. min_bets is raised to 300 to match the prediction. sp_decimal > 8 excludes 7/1 and covers 15/2 to 14/1, consistent with the earlier '<= 15' rules, so the bands fit together exactly. Prior evidence leans against this hypothesis: F/U runners at 14/1 or shorter scored 1.10 and 1.13 in total. But the quick-return BD/UR 14/1 band scored 0.579 on 76, which motivates the test. The rule uses only pre-off columns; no result columns are read."
  }

Timeline

1. Job 1672 finished

   16 Oct, 13:46 · engine:compute-1

2. H-567e3266 Discovery → KILLED fails: 280 bets, needs 300; p 0.258 (lower tail), needs 0.05

   16 Oct, 13:46 · engine:evaluator

3. discovery result for H-567e3266 (broad rule): fail

   16 Oct, 13:46 · engine:evaluator

4. H-567e3266 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 13:46 · engine:orchestrator

5. compute-1 took job 1672 (evaluate, attempt 1)

   16 Oct, 13:46 · compute-1

6. Job 1671 finished

   16 Oct, 13:43 · engine:ai-1

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

   16 Oct, 13:43 · engine:scheduler

8. H-567e3266 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 13:43 · agent:designer@v1

9. ai-1 took job 1671 (design, attempt 1)

   16 Oct, 13:42 · ai-1

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

    16 Oct, 13:42 · engine:scheduler

11. New idea H-567e3266: Mid-priced jumps departures (8/1 to 14/1) are over-bet on the 'unlucky last time' story (test-mechanism)

    16 Oct, 13:42 · 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.