Rival: short-priced jumps runners that completed last time beat their price as much as short-priced departures

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) passed discovery but was falsified: driven by one price band (up to 3/1): the rest score 0.92. O-7f64a209
   • H-567e3266 (Mid-priced jumps departures (8/1 to 14/1) are over-bet on the 'unlucky) pointed the predicted way but failed the bar: A/E 0.82 on 280 bets (p 0.814), control 1.02. O-eb56a99e
2. AskedIs the up-to-3/1 lift in jumps departure rules just the general short-price end of the jumps market in this window (October 2025 on), shared by runners that completed last time? testing
3. MechanismIn this window, jumps runners at the front of the market may simply be priced slightly generously relative to the rest (the favourite-longshot bias). Any subset of runners, departures included, would then show its lift in the shortest band and a deficit at mid prices.
4. Why the market misses itThe market spreads its overround toward bigger prices. Favourites carry less of the margin, so any rule that takes in short-priced runners picks up a small lift without any hidden-form story.
5. Predictedpositive, A/E 1.03 to 1.10 on 2000+ bets, close to the fallers' up-to-3/1 band
6. Falsified ifA/E 1.00 or below on 2000+ bets, which would leave the short-price departure lift unexplained by market structure

Mechanism and rationale

In this window, jumps runners at the front of the market may simply be priced slightly generously relative to the rest (the favourite-longshot bias). Any subset of runners, departures included, would then show its lift in the shortest band and a deficit at mid prices.

The rule

race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 4 AND prior_runs >= 1 AND last_position IS NOT NULL

Reads: race_type, sp_decimal, prior_runs, last_position

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

e76ca189ce929c2fe954d9ed9f6fc062e62b6d48ba2bb11b86d4933d79cd41db

Frozen

not frozen

The definition can still change.

Confounders

• Field size: short-priced runners cluster in small fields, where the overround is lower and how it is spread differs
• Race class and novice or graded races, where short-priced runners are over-represented
• Overlap with the same-race control used to judge departure rules: these completers make up much of that control, so the two are not independent
• Window: the favourite-longshot slope may differ between 2020-2024 and 2025 onward, when the departure rules were measured (refinement 1 checks this)
• Region: GB and IRE books may spread their margin differently

Control

{
  "description": "The usual same-race control is the other runners in these races. They are mostly at longer prices, so if the mechanism holds they should show A/E slightly below 1 (about 0.95 to 0.99) because they carry the overround. The comparison that answers the question is short-priced jumps runners that did NOT finish last time, at the same price cap. If completers at 3/1 or shorter match that group's A/E (within the CIs), the departure lift up to 3/1 is a general short-price effect. If completers sit near 1.00 while departures are clearly higher, the lift is specific to departures.",
  "rule": "race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 4 AND prior_runs >= 1 AND last_position IS NULL"
}

Designer

Discovery result

fail broad rule · 16 Oct, 13:51

Bets

6,303

2,160 winners

A/E

0.99

95%: 0.95 to 1.04

SP return

-9.8%

level stakes

Betfair SP return

-1.8%

6,093 bets with a BSP

Strike rate 34.3%. p = 0.61. Same-race control: the other runners in these races had A/E 0.99, 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 · 42s · 16 Oct, 13:49

  Output (schema-checked)

  {
    "rule": "race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 4 AND prior_runs >= 1 AND last_position IS NOT NULL",
    "refinements": [
      {
        "label": "Same window as the departure rules (2025 onward)",
        "rule": "race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 4 AND prior_runs >= 1 AND last_position IS NOT NULL AND race_date >= '2025-01-01'"
      },
      {
        "label": "Shortest end only (6/4 or shorter), where the favourite-longshot bias should be strongest",
        "rule": "race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 2.5 AND prior_runs >= 1 AND last_position IS NOT NULL"
      }
    ],
    "control": {
      "description": "The usual same-race control is the other runners in these races. They are mostly at longer prices, so if the mechanism holds they should show A/E slightly below 1 (about 0.95 to 0.99) because they carry the overround. The comparison that answers the question is short-priced jumps runners that did NOT finish last time, at the same price cap. If completers at 3/1 or shorter match that group's A/E (within the CIs), the departure lift up to 3/1 is a general short-price effect. If completers sit near 1.00 while departures are clearly higher, the lift is specific to departures.",
      "rule": "race_type IN ('Hurdle', 'Chase') AND sp_decimal <= 4 AND prior_runs >= 1 AND last_position IS NULL"
    },
    "pass_criteria": {
      "min_bets": 2000,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 300,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Field size: short-priced runners cluster in small fields, where the overround is lower and how it is spread differs",
      "Race class and novice or graded races, where short-priced runners are over-represented",
      "Overlap with the same-race control used to judge departure rules: these completers make up much of that control, so the two are not independent",
      "Window: the favourite-longshot slope may differ between 2020-2024 and 2025 onward, when the departure rules were measured (refinement 1 checks this)",
      "Region: GB and IRE books may spread their margin differently"
    ],
    "mechanism": "Favourite-longshot bias over jumps: the market puts too little of its overround on the front of the market, so runners at 3/1 or shorter win slightly more often than their SP implies, whatever their last-time story. If that holds, the up-to-3/1 lift seen in fallers, BD/UR and going-well departure rules is a feature of price, not of hidden form. Short-priced runners that completed last time should then show the same lift.",
    "market_miss": "Bookmakers load their margin onto bigger prices, where recreational money goes, and price favourites tighter to the true odds. Any rule that takes in short-priced jumps runners picks up a small structural lift with no informational edge. It is a property of how the book is built, not something the market misreads about the horse.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.03 to 1.10 on 2000+ bets, close to the fallers' up-to-3/1 band and within the CI of the short-priced non-finisher control",
      "falsified_if": "A/E 1.00 or below on 2000+ bets, or clearly below the short-priced non-finisher control (no CI overlap). Either would mean the short-price departure lift is not explained by market structure."
    },
    "notes": "I replaced the sketch's last_fell_or_unseated = 0 AND last_pulled_up = 0 with last_position IS NOT NULL AND prior_runs >= 1. Both flags only exist from 2025-11-01, so the sketch would have silently dropped every earlier runner (NULL = 0 is false). It would also have let in BD/RO/REF non-finishers. Using last_position IS NOT NULL gives true completers across the whole window. Refinement 1 restricts to 2025 onward to match the window of the departure rules. sp_decimal <= 4 matches the 'up to 3/1' band. The predicted size (1.03 to 1.10) overlaps the lab floor of 1.05: a true rival effect of about 1.03 could be real yet fail the discovery bar. A result between 1.00 and 1.05 should be read as partial support for the rival, not as refutation. min_bets is raised to 2000 to match the prediction's sample size."
  }

Timeline

1. Job 1693 finished

   16 Oct, 13:51 · engine:compute-1

2. H-7de20b37 Discovery → KILLED fails: A/E 0.994, needs 1.05; p 0.6058, needs 0.05 or lower; periods [1.01, 0.96], each needs 1.0; A/E 0.994 against same-race control 0.995, needs 0.03 above (not a niche pass: A/E 0.994 below 1.05; p 0.6058069056779608 above 0.5; does not beat the same-race control 0.995)

   16 Oct, 13:51 · engine:evaluator

3. discovery result for H-7de20b37 (broad rule): fail

   16 Oct, 13:51 · engine:evaluator

4. H-7de20b37 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 13:51 · engine:orchestrator

5. compute-1 took job 1693 (evaluate, attempt 1)

   16 Oct, 13:51 · compute-1

6. Job 1692 finished

   16 Oct, 13:50 · engine:ai-1

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

   16 Oct, 13:50 · engine:scheduler

8. H-7de20b37 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 13:50 · agent:designer@v1

9. ai-1 took job 1692 (design, attempt 1)

   16 Oct, 13:49 · ai-1

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

    16 Oct, 13:49 · engine:scheduler

11. New idea H-7de20b37: Rival: short-priced jumps runners that completed last time beat their price as much as short-priced departures (rival-explanation)

    16 Oct, 13:49 · 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.