Duel-burnt front runner: pressed for the lead, folded, in a race that did not collapse

Mechanism and rationale

A leader or prominent racer that is taken on by one rival and pressed throughout runs early fractions above its own aerobic threshold. Lactate builds up and it folds late. The cause is one rival's tactics in that race, not the horse's ability or stamina. Next time it is unlikely to meet the same duel and can set or track a pace it can sustain.

The rule

race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND l_pressed = 1 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5

Reads: race_type, l_same_code, l_pressed, l_folded, last_position, last_field_size

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

bd66b2e86557844a266cb5b58438af4e2a67f2a1aaaec1d8d5acf6f9d782a222

Frozen

not frozen

The definition can still change.

Confounders

• Genuine non-stayers: the fold may be the trip, not the duel; refinement 1 keeps only same-or-shorter trips today
• Horses in decline or with a fitness problem fold whether pressed or not; the folded-but-not-pressed control separates this
• Overlap with the pace-victim rule (H-22ef2116): the race-tilt filter only exists from Nov 2025, so it is a refinement, not part of the broad rule; report overlap with H-22ef2116 selections
• Coverage: RaceiQ flags start GB April 2023, IRE January 2024, jumps only; period split is 2023 to a year ago vs the last year, both must agree
• Price band: folded front runners drift next time, where A/E is noisy; refinement 2 caps at 14/1
• Being pressed may partly reflect quality of the rival (a stronger horse took it on), i.e. it was simply outclassed; check by class change today

Control

{
  "description": "The lab's same-race control: other runners in the same races today. They should come out near the usual RaceiQ control level (about 1.03-1.05), so the selection has to beat its own race-mates, not just the market. A second comparison: horses whose last run was folded but NOT pressed (they weakened without being harassed). The mechanism says the duel is the excuse, so the unpressed folders should run near or below 1.00. If they run as well as the pressed folders, the effect is about 'folded last time' being overpriced, not about the duel.",
  "rule": "race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND coalesce(l_pressed, 0) = 0 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5"
}

Discovery result

fail broad rule · 15 Oct, 23:28

Bets

2,591

320 winners

A/E

1.04

95%: 0.93 to 1.17

SP return

-15.6%

level stakes

Betfair SP return

+3.1%

2,526 bets with a BSP

Strike rate 12.4%. p = 0.22. Same-race control: the other runners in these races had A/E 0.99, so the selection beat its own races. Zone: discovery. Evaluator eval-1.0, data ef1163efa87ef7d2

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

No parent and no descendants yet. When an agent mutates, generalises or splits this idea, the new hypotheses appear here.

Agent runs

• designer v1ok · 38s · 15 Oct, 23:27

  Output (schema-checked)

  {
    "rule": "race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND l_pressed = 1 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5",
    "refinements": [
      {
        "label": "same or shorter trip today (rules out the non-stayer reading)",
        "rule": "race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND l_pressed = 1 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5 AND distance_yards <= last_distance_yards + 110"
      },
      {
        "label": "14/1 or shorter (edges die at big prices; folded front runners drift)",
        "rule": "race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND l_pressed = 1 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5 AND sp_decimal <= 15"
      },
      {
        "label": "race shape known and not collapsed (separates from pace-victim H-22ef2116; Nov 2025 onward only)",
        "rule": "race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND l_pressed = 1 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5 AND last_race_tilt IS NOT NULL AND last_race_tilt < 0.2"
      }
    ],
    "control": {
      "description": "The lab's same-race control: other runners in the same races today. They should come out near the usual RaceiQ control level (about 1.03-1.05), so the selection has to beat its own race-mates, not just the market. A second comparison: horses whose last run was folded but NOT pressed (they weakened without being harassed). The mechanism says the duel is the excuse, so the unpressed folders should run near or below 1.00. If they run as well as the pressed folders, the effect is about 'folded last time' being overpriced, not about the duel.",
      "rule": "race_type IN ('Hurdle', 'Chase') AND l_same_code = 1 AND coalesce(l_pressed, 0) = 0 AND l_folded = 1 AND last_position IS NOT NULL AND last_position > 1 AND last_field_size >= 5"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.08,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Genuine non-stayers: the fold may be the trip, not the duel; refinement 1 keeps only same-or-shorter trips today",
      "Horses in decline or with a fitness problem fold whether pressed or not; the folded-but-not-pressed control separates this",
      "Overlap with the pace-victim rule (H-22ef2116): the race-tilt filter only exists from Nov 2025, so it is a refinement, not part of the broad rule; report overlap with H-22ef2116 selections",
      "Coverage: RaceiQ flags start GB April 2023, IRE January 2024, jumps only; period split is 2023 to a year ago vs the last year, both must agree",
      "Price band: folded front runners drift next time, where A/E is noisy; refinement 2 caps at 14/1",
      "Being pressed may partly reflect quality of the rival (a stronger horse took it on), i.e. it was simply outclassed; check by class change today"
    ],
    "mechanism": "A front runner taken on and pressed throughout by a rival runs early fractions above its sustainable aerobic pace; lactate builds and it folds late. The cause is that one race's tactics, not the horse's ability or stamina. Next time it is unlikely to meet the same duel and can set or track a pace it can sustain, so it should run closer to its true level than the ugly last run suggests.",
    "market_miss": "The form book reads 'led, headed 2 out, weakened' with a big beaten distance that looks like a non-stayer or a horse out of form, and the next price (and handicap mark) treat the fold as the horse's limit. The RaceiQ 'pressed' flag, showing it was harassed rather than freely in front, is not on any public card, so the market cannot tell a duel-induced fold from a genuine collapse.",
    "notes": "Fixed the draft: last_run_style and last_race_tilt only exist from Nov 2025. With coalesce(tilt,0) the tilt filter would have done nothing before then and applied only in the last year, and last_run_style would have cut the sample to about a year. l_pressed already means the horse was at or near the front being taken on, so the run-style condition is redundant. The rule now uses only RaceiQ columns (from April 2023), limited to jumps races with a last run in the same code (as in H-d28065ea). The tilt and non-collapse condition moves to refinement 3. Raised min_ae to 1.08 because the RaceiQ control baseline runs about 1.03-1.05."
  }

Timeline

1. Job 52 finished

   15 Oct, 23:28 · engine:compute-2

2. H-1da60d84 Discovery → KILLED fails: A/E 1.045, needs 1.08; p 0.2207, needs 0.05 or lower (not a niche pass: A/E 1.045 below 1.08)

   15 Oct, 23:28 · engine:evaluator

3. discovery result for H-1da60d84 (broad rule): fail

   15 Oct, 23:28 · engine:evaluator

4. H-1da60d84 Designed → RESEARCHdiscovery measuring on the discovery zone

   15 Oct, 23:28 · engine:orchestrator

5. compute-2 took job 52 (evaluate, attempt 1)

   15 Oct, 23:28 · compute-2

6. Job 42 finished

   15 Oct, 23:28 · engine:ai-1

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

   15 Oct, 23:28 · engine:scheduler

8. H-1da60d84 Proposed → RESEARCHdesigned pre-spec written and hashed

   15 Oct, 23:28 · agent:designer@v1

9. ai-1 took job 42 (design, attempt 1)

   15 Oct, 23:27 · ai-1

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

    15 Oct, 23:27 · engine:scheduler

11. New idea H-1da60d84: Duel-burnt front runner: pressed for the lead, folded, in a race that did not collapse (scout)

    15 Oct, 23:27 · agent:scout@v1

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Killed.