Hung or jinked last time on a track going the other way round, today the opposite direction

Mechanism and rationale

A horse that hangs, lugs or jinks is often reacting to the track's camber and bends, or to discomfort on one lead. When the wayward run came on a left-handed track and today's track goes right (or the other way round), the physical cause is gone and the energy wasted moving sideways comes back.

The rule

last_wayward = 1 AND last_position > 1 AND lower(track_direction) IN ('left', 'right') AND lower(last_track_direction) IN ('left', 'right') AND lower(track_direction) != lower(last_track_direction)

Reads: last_wayward, last_position, track_direction, last_track_direction

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

1d4205a2168b0982330e84631fa37903b0cdac52ab3c137c0ca290355a4c1d15

Frozen

not frozen

The definition can still change.

Confounders

• Real temperament: wayward horses tend to do it again wherever they run, so the trait (which the market marks down fairly) could swamp any direction effect. The same-direction control tests this directly.
• The wayward tag ('edged left/right') may come from tiredness late in the race rather than from the track. Refinement 2 (beaten 8 lengths or less) tests this; samples of the tag should be read by eye.
• Hanging caused by interference rather than the track; refinement 3 removes runs with trouble in running.
• Earlier direction-preference rules were mixed (H185 1.02, H186 negative 0.92), so any edge has to come from the wayward trigger. The control is the same trigger with no change of direction.
• Straight and figure-of-eight tracks: excluded by requiring both directions to be left or right, but how the course guides label tracks is not checked (hence lower()).
• Data window: last_wayward only exists from 2025-11-01, so there is under a year of discovery data. Split into two halves rather than three periods; the sample will probably be thin, so it may need forward tracking rather than passing in discovery.

Discovery result

fail broad rule · 15 Oct, 23:24

Bets

2,692

332 winners

A/E

0.94

95%: 0.84 to 1.05

SP return

-24.9%

level stakes

Betfair SP return

-3.5%

2,676 bets with a BSP

Strike rate 12.3%. p = 0.87. Same-race control: the other runners in these races had A/E 1.01, so the selection did not 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 · 55s · 15 Oct, 23:23

  Output (schema-checked)

  {
    "rule": "last_wayward = 1 AND last_position > 1 AND lower(track_direction) IN ('left', 'right') AND lower(last_track_direction) IN ('left', 'right') AND lower(track_direction) != lower(last_track_direction)",
    "refinements": [
      {
        "label": "14/1 or shorter (edges die at big prices; a horse whose wayward run hid real ability should still be in the market)",
        "rule": "last_wayward = 1 AND last_position > 1 AND lower(track_direction) IN ('left', 'right') AND lower(last_track_direction) IN ('left', 'right') AND lower(track_direction) != lower(last_track_direction) AND sp_decimal <= 15"
      },
      {
        "label": "Beaten 8 lengths or less last time (removes 'edged left when tired'; ground lost sideways only matters if it was close)",
        "rule": "last_wayward = 1 AND last_position > 1 AND lower(track_direction) IN ('left', 'right') AND lower(last_track_direction) IN ('left', 'right') AND lower(track_direction) != lower(last_track_direction) AND last_beaten_lengths <= 8"
      },
      {
        "label": "No interference last time (hanging after a bump is about traffic, not the track)",
        "rule": "last_wayward = 1 AND last_position > 1 AND lower(track_direction) IN ('left', 'right') AND lower(last_track_direction) IN ('left', 'right') AND lower(track_direction) != lower(last_track_direction) AND coalesce(last_trouble_in_running, 0) = 0"
      }
    ],
    "control": {
      "description": "Main control: wayward, beaten horses whose today's track goes the SAME way as the wayward run. If the mechanism is real, the opposite-direction group should clearly beat this group. If both score about the same (both near or both above 1.00), the result is about how the market prices wayward horses in general, not about the change of direction. The lab's usual same-race comparison (other runners in the same races) should be about 1.00 to 1.05, as usual for RaceiQ-era controls. Anything above that would point to a race-level property rather than the horse.",
      "rule": "last_wayward = 1 AND last_position > 1 AND lower(track_direction) IN ('left', 'right') AND lower(last_track_direction) IN ('left', 'right') AND lower(track_direction) = lower(last_track_direction)"
    },
    "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": [
      "Real temperament: wayward horses tend to do it again wherever they run, so the trait (which the market marks down fairly) could swamp any direction effect. The same-direction control tests this directly.",
      "The wayward tag ('edged left/right') may come from tiredness late in the race rather than from the track. Refinement 2 (beaten 8 lengths or less) tests this; samples of the tag should be read by eye.",
      "Hanging caused by interference rather than the track; refinement 3 removes runs with trouble in running.",
      "Earlier direction-preference rules were mixed (H185 1.02, H186 negative 0.92), so any edge has to come from the wayward trigger. The control is the same trigger with no change of direction.",
      "Straight and figure-of-eight tracks: excluded by requiring both directions to be left or right, but how the course guides label tracks is not checked (hence lower()).",
      "Data window: last_wayward only exists from 2025-11-01, so there is under a year of discovery data. Split into two halves rather than three periods; the sample will probably be thin, so it may need forward tracking rather than passing in discovery.",
      "Moving from a left-handed to a right-handed track often means a change of region or track type (e.g. many right-handed tracks are sharp jumps tracks), so the track character could be the real cause. Check by race_type and region."
    ],
    "mechanism": "A horse that hangs, lugs or jinks is often reacting to the camber and bends of a track going one way, or to discomfort when leading on one leg. That makes the bad run partly a product of the track. When today's track goes the other way (left after right, or right after left), that cause is gone. The energy it wasted moving sideways goes into going forward, so it should run better than its last-time form suggests. A horse that is simply ungenuine should not improve.",
    "market_miss": "'Hung badly right' reads as temperament or lack of resolution, and the market fairly marks down horses with a bad character. It does not separate a horse that misbehaves anywhere from one that struggled with that track's direction. The change of direction is shown on the racecard, not in the form figures or the in-running comment, so the price treats every wayward horse alike.",
    "notes": "The draft rule was kept; the only change is lower() on both direction columns, because I could not confirm from the catalogue how the direction values are written (case). There is no use of the result or post-race columns. last_wayward only starts on 2025-11-01, so the discovery sample is short: split it in two halves and expect a thin result. The same-direction control is the key test: the opposite-direction group has to beat it as well as reaching A/E 1.05. Before trusting the result, read about 30 last_wayward remarks by eye to check the tag catches hanging, lugging and jinking, not tiredness late in the race."
  }

Timeline

1. Job 31 finished

   15 Oct, 23:24 · engine:compute-1

2. H-6d38d88b Discovery → KILLED fails: A/E 0.942, needs 1.05; p 0.8668, needs 0.05 or lower; periods [0.84, 1.0], each needs 1.0; A/E 0.942 against same-race control 1.01, needs 0.03 above (not a niche pass: A/E 0.942 below 1.05; p 0.8668323334758147 above 0.5; does not beat the same-race control 1.01; a period points the other way [0.84, 1.0])

   15 Oct, 23:24 · engine:evaluator

3. discovery result for H-6d38d88b (broad rule): fail

   15 Oct, 23:24 · engine:evaluator

4. H-6d38d88b Designed → RESEARCHdiscovery measuring on the discovery zone

   15 Oct, 23:24 · engine:orchestrator

5. compute-1 took job 31 (evaluate, attempt 1)

   15 Oct, 23:24 · compute-1

6. Job 26 finished

   15 Oct, 23:24 · engine:ai-1

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

   15 Oct, 23:24 · engine:scheduler

8. H-6d38d88b Proposed → RESEARCHdesigned pre-spec written and hashed

   15 Oct, 23:24 · agent:designer@v1

9. ai-1 took job 26 (design, attempt 1)

   15 Oct, 23:23 · ai-1

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

    15 Oct, 23:22 · engine:scheduler

11. New idea H-6d38d88b: Hung or jinked last time on a track going the other way round, today the opposite direction (scout)

    15 Oct, 23:22 · 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.