Wide last time, drawn low today on the Flat

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
   • IQ140a-wide-trip-with-figures (IQ140a: wide trip with figures) was GREEN in the earlier research: A/E 1.27 on 1080 bets (periods 1.15 / 1.41); now PROMISING. O-943a9f88
2. AskedOn the Flat, does a low draw today (unlikely to be forced wide again) carry more of the edge than a high draw? answered
3. MechanismThe value is that the ground lost last time will not repeat. From a low (inside) stall the horse can save ground today. From a wide stall it may lose the same ground again and run to its bare form.
4. Why the market misses itThe market prices the draw in general, but not as a release from a disadvantage the form figures hide. The horse's last run looks ordinary, and nothing in the form flags the draw today as a change.
5. Predictedpositive, A/E 1.30+ on 200+ bets, above the high-draw wide runners
6. Falsified ifA/E below 1.15, or no better than wide runners drawn in the top half of the field
7. Rule sketchrace_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw * 2 <= field_size

Mechanism and rationale

The value is that the ground lost last time will not repeat. From a low (inside) stall the horse can save ground today. From a wide stall it may lose the same ground again and run to its bare form.

The rule

race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 6 AND draw * 2 <= field_size

Reads: race_type, last_raced_wide, last_position, last_beaten_lengths, last_slow_start, draw, field_size

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

64dcd69809f191c192a5ba1453b4e4510ea46fb3e1a11ebe6c14a62524832ecd

Frozen

16 Oct, 12:11

Nothing in the definition can change after this.

Confounders

• Course draw bias: at some tracks high stalls are favoured (track_draw_bias), so 'low' is not always the good side. Split by track_draw_bias.
• Straight-course races have no bends, so there is no ground to save from a low draw; on straight courses low can mean either side. Refinement 1 isolates races with a bend.
• Stall 1 is not on the inside rail at every course or every distance (rail placement, direction of track). Split by track_direction.
• Field size: in small fields the draw hardly matters and 'bottom half' covers almost everyone. Guarded by field_size >= 6; refinement 2 uses 10+.
• Longshot artefact: beaten horses show A/E above 1.10 at 16/1+ in general. Refinement 3 keeps 14/1 or shorter.
• Thin data window: last_raced_wide exists only from November 2025, so about a year in two halves; both halves must agree.
• Overlap with the parent wide-trip system (LIVE): this is a subset of it, so a pass is only news if it beats the high-drawn half.

Control

{
  "description": "The question is whether a low draw carries MORE of the wide-trip edge, so the main comparison is the same wide-trip runners drawn in the top half of the field (control rule). If the mechanism is right, the low-drawn group beats the high-drawn group clearly, and the high-drawn group sits close to its bare form (A/E near 1.0 to 1.15). If both groups show the same A/E (about the 1.35 of the parent wide-trip rule), the draw adds nothing and the hypothesis is falsified. The lab's usual same-race other runners should sit near 1.00. Low-drawn runners who did not race wide last time should also be near 1.00; if they are well above 1.00, the lift comes from draw bias at the track, not from release from a wide trip.",
  "rule": "race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 6 AND draw * 2 > field_size"
}

Discovery result

pass broad rule (niche) · 16 Oct, 11:16

Bets

456

74 winners

A/E

1.12

95%: 0.88 to 1.40

SP return

+7.7%

level stakes

Betfair SP return

+29.3%

454 bets with a BSP

Strike rate 16.2%. p = 0.19. 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 2887c9abd664f46f

Falsification

Each attack is a deliberate attempt to kill the idea. 4 of 5 survived.

• price_band: Low draw, 14/1 or shorter only (strip longshot artefact)pass

  368 bets · A/E 1.06 (0.83 to 1.35) · control 0.98 · 16 Oct, 12:11

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271611.00

  after 2026-04-272071.11

• control: Control: low-drawn runners NOT wide last time (draw-bias check)pass

  9,384 bets · A/E 1.01 (0.96 to 1.07) · control 0.99 · 16 Oct, 12:11

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

Forward record

• forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}info

  0 bets · · 16 Oct, 12:12

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: > 1 -> 0 (-20%)pass

  583 bets · A/E 1.06 (0.86 to 1.30) · control 0.99 · 16 Oct, 11:16

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272541.00

  after 2026-04-273291.11

• neighbour: > 1 -> 2 (+20%)pass

  340 bets · A/E 1.09 (0.80 to 1.46) · control 0.99 · 16 Oct, 11:16

  Breakdown

Lineage

Agent runs

• designer v1ok · 43s · 16 Oct, 11:15

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 6 AND draw * 2 <= field_size",
    "refinements": [
      {
        "label": "race has a bend (round course)",
        "rule": "race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 6 AND draw * 2 <= field_size AND distance_yards > track_straight_furlongs * 220"
      },
      {
        "label": "big field (10+), where the draw decides who goes wide",
        "rule": "race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 10 AND draw * 2 <= field_size"
      },
      {
        "label": "14/1 or shorter",
        "rule": "race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 6 AND draw * 2 <= field_size AND sp_decimal <= 15"
      }
    ],
    "control": {
      "description": "The question is whether a low draw carries MORE of the wide-trip edge, so the main comparison is the same wide-trip runners drawn in the top half of the field (control rule). If the mechanism is right, the low-drawn group beats the high-drawn group clearly, and the high-drawn group sits close to its bare form (A/E near 1.0 to 1.15). If both groups show the same A/E (about the 1.35 of the parent wide-trip rule), the draw adds nothing and the hypothesis is falsified. The lab's usual same-race other runners should sit near 1.00. Low-drawn runners who did not race wide last time should also be near 1.00; if they are well above 1.00, the lift comes from draw bias at the track, not from release from a wide trip.",
      "rule": "race_type = 'Flat' AND last_raced_wide = 1 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND draw >= 1 AND field_size >= 6 AND draw * 2 > field_size"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Course draw bias: at some tracks high stalls are favoured (track_draw_bias), so 'low' is not always the good side. Split by track_draw_bias.",
      "Straight-course races have no bends, so there is no ground to save from a low draw; on straight courses low can mean either side. Refinement 1 isolates races with a bend.",
      "Stall 1 is not on the inside rail at every course or every distance (rail placement, direction of track). Split by track_direction.",
      "Field size: in small fields the draw hardly matters and 'bottom half' covers almost everyone. Guarded by field_size >= 6; refinement 2 uses 10+.",
      "Longshot artefact: beaten horses show A/E above 1.10 at 16/1+ in general. Refinement 3 keeps 14/1 or shorter.",
      "Thin data window: last_raced_wide exists only from November 2025, so about a year in two halves; both halves must agree.",
      "Overlap with the parent wide-trip system (LIVE): this is a subset of it, so a pass is only news if it beats the high-drawn half."
    ],
    "mechanism": "A horse forced wide last time ran worse than its ability because it covered extra ground, so its form figure understates it. The edge only pays if that loss does not repeat. On the Flat, a stall in the lower (inside) half of the field gives the horse a fair chance of a rail position and of saving ground on the bends. From a high (outside) stall it is likely to be forced wide again and to run only to its bare form.",
    "market_miss": "The market prices the draw in general and reads last time's form at face value. It does not join the two: it does not see that a low draw today removes the specific handicap that made the last run look ordinary. Nothing in the form figures flags the wide trip, so the change in circumstances goes unpriced.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.30+ on 200+ bets, clearly above the high-drawn wide-trip runners in the control",
      "falsified_if": "A/E below 1.15 on the broad rule; or the low-drawn group is no better than the high-drawn control (difference under 0.10 A/E); or the two halves of the data disagree in direction."
    },
    "notes": "This kept the sketch rule. It adds draw >= 1 so that missing or zero stalls are excluded, and field_size >= 6 because the draw means little in tiny fields. Low draw is defined as being in the bottom half of the stalls (draw * 2 <= field_size). The comparison that matters is with the high-drawn control, not with 1.00: the parent wide-trip rule already runs at about 1.35. If the broad rule passes but does not beat the control, this is no evidence for the draw mechanism. It would just be the parent edge again. Each refinement comes straight from the mechanism: races with a bend (no ground to save on a straight course), big fields, and 14/1 or shorter (the longshot artefact). The refinements are only looked at if the broad rule shows evidence."
  }

Timeline

1. Job 1280 finished

   16 Oct, 12:12 · engine:compute-1

2. forward result for H-baf747c3 (forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}): info

   16 Oct, 12:12 · engine:evaluator

3. compute-1 took job 1280 (forward_check, attempt 1)

   16 Oct, 12:12 · compute-1

4. Queued forward_check on the compute lane: daily forward check of a shadow system

   16 Oct, 12:12 · engine:scheduler

5. Job 1263 finished

   16 Oct, 12:11 · engine:compute-1

6. H-baf747c3 Frozen → SHADOW vault skipped: last_raced_wide, last_beaten_lengths, last_slow_start do not exist before the discovery zone; track it forward instead

   16 Oct, 12:11 · engine:orchestrator

7. compute-1 took job 1263 (vault, attempt 1)

   16 Oct, 12:11 · compute-1

8. Job 1254 finished

   16 Oct, 12:11 · engine:compute-1

9. Queued vault on the compute lane: frozen: one sealed vault query

   16 Oct, 12:11 · engine:scheduler

10. H-baf747c3 Falsification → FROZEN survived falsification

    16 Oct, 12:11 · engine:falsification

11. falsification result for H-baf747c3 (perturbation: Big fields 10+ with a tighter beaten-distance cut (4L)): pass

    16 Oct, 12:11 · engine:evaluator

12. compute-1 took job 1254 (attack, attempt 1)

    16 Oct, 12:11 · compute-1

13. Job 1253 finished

    16 Oct, 12:11 · engine:compute-1

14. falsification result for H-baf747c3 (perturbation: Stronger dose: bottom third of the draw): pass

    16 Oct, 12:11 · engine:evaluator

15. compute-1 took job 1253 (attack, attempt 1)

    16 Oct, 12:11 · compute-1

16. Job 1252 finished

    16 Oct, 12:11 · engine:compute-1

17. falsification result for H-baf747c3 (partition: Round course only: race longer than the straight): fail

    16 Oct, 12:11 · engine:evaluator

18. compute-1 took job 1252 (attack, attempt 1)

    16 Oct, 12:11 · compute-1

19. Job 1251 finished

    16 Oct, 12:11 · engine:compute-1

20. falsification result for H-baf747c3 (control: Control: low-drawn runners NOT wide last time (draw-bias check)): pass

    16 Oct, 12:11 · engine:evaluator