Test-mechanism: one-off chase error, well beaten, price drifted from last run

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-64933537 (Replicate: one-off chase error, well beaten, in Irish chases) passed discovery as a niche: A/E 1.19 on 189 bets (p 0.249), same-race control 0.98. O-46d844d8
   • H-64933537 (Replicate: one-off chase error, well beaten, in Irish chases): the price band 'up to 14/1' carries the effect (A/E 1.4) while the rest score up to 7/1 0.96, 16/1 or bigger 0.0. O-4a751487
2. AskedIs the one-off chase-error edge concentrated in runners whose price lengthened from last time (sp_decimal > last_sp_decimal), where the market has visibly marked them down for the bad run? answered
3. MechanismThe error was a one-off, so the horse's real ability is unchanged. A price drift since the last run means the market has marked it down for the bad run.
4. Why the market misses itThe market prices off the latest form figure and beaten distance. The cause, a single blunder after two clean rounds, sits in the in-running comments, which most bettors do not weigh.
5. Predictedpositive, A/E 1.30 to 1.45 on about 250 to 350 bets
6. Falsified ifA/E at or below 1.224, or runners whose price shortened scoring the same or higher
7. Rule sketch

Mechanism and rationale

The error was a one-off, so the horse's real ability is unchanged. A price drift since the last run means the market has marked it down for the bad run.

The rule

race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal

Reads: race_type, remarked_last3, jumping_errors_last3, last_jumping_error, last_position, last_beaten_lengths, last_sp_decimal, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

1d20a808ee08338286fb8e0e7706df6ac338e1f770b9c135e673b4466de027c2

Frozen

17 Oct, 01:35

Nothing in the definition can change after this.

Confounders

• Class rises and tougher races lengthen prices by themselves. A drift may mark a harder task rather than a market overreaction, so check the result by whether race_class went up against last_race_class (GB only, since IRE class data starts July 2026).
• Drift pushes runners into 16/1+, where chase error pools have scored near zero (IRE 0 from 64). The broad result can be dragged down or made noisy by longshots, which is why there is a 14/1-or-shorter refinement.
• Field size: bigger fields lengthen every price, so some of the drift is mechanical. Compare field_size with last_field_size.
• Period effect: remarked_last3 only exists from November 2025, so the window is short and one season could drive the result. Check the split before and after April 2026.
• Overlap: this is another subset of the one-off chase-error pool, which has already passed in several overlapping forms. A pass here is not independent confirmation of the base edge. It only adds information if the drift and shortened groups separate.
• Region mix: GB (1.24) and IRE (1.19) pools differ a little. Report both regions.

Discovery result

pass broad rule · 17 Oct, 01:28

Bets

263

31 winners

A/E

1.50

95%: 1.02 to 2.13

SP return

+4.2%

level stakes

Betfair SP return

+28.9%

262 bets with a BSP

Strike rate 11.8%. p = 0.020. Same-race control: the other runners in these races had A/E 0.96, so the selection beat its own races. Zone: discovery. Evaluator eval-1.0, data 6a4e98be18e83933

Falsification

Each attack is a deliberate attempt to kill the idea. 6 of 6 survived.

• perturbation: Perturb beaten-lengths cut to 8 and require a 10%+ driftpass

  271 bets · A/E 1.28 (0.84 to 1.88) · control 0.99 · 17 Oct, 01:35

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271281.49

  after 2026-04-271431.11

• perturbation: Perturb beaten-lengths cut to 12.5 with any driftpass

  246 bets · A/E 1.51 (1.01 to 2.18) · control 0.96 · 17 Oct, 01:35

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 · · 17 Oct, 01:36

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: > 10 -> 8 (-20%)pass

  287 bets · A/E 1.39 (0.95 to 1.97) · control 0.97 · 17 Oct, 01:28

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271331.44

  after 2026-04-271541.35

• neighbour: > 10 -> 12 (+20%)pass

  248 bets · A/E 1.54 (1.04 to 2.20) · control 0.96 · 17 Oct, 01:28

  Breakdown

Lineage

Agent runs

• designer v1ok · 35s · 17 Oct, 01:27

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal",
    "refinements": [
      {
        "label": "drifted, 14/1 or shorter",
        "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal AND sp_decimal <= 15"
      },
      {
        "label": "clear drift (25%+ longer than last SP)",
        "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND last_sp_decimal IS NOT NULL AND sp_decimal >= 1.25 * last_sp_decimal"
      }
    ],
    "control": {
      "description": "Other runners in the same races should score about 0.95 to 1.0, as in earlier chase-error tests, which shows the lift is not a property of the races. The mechanism's own test is the same one-off-error pool whose price shortened or stayed the same since its last run. If drift is what marks the market's overreaction, this group should score clearly below the drifted group, at or under the whole-pool A/E of about 1.22. If the shortened group scores the same or higher, the drift mechanism is falsified even if the broad rule passes.",
      "rule": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND last_sp_decimal IS NOT NULL AND sp_decimal <= last_sp_decimal"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.15,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Class rises and tougher races lengthen prices by themselves. A drift may mark a harder task rather than a market overreaction, so check the result by whether race_class went up against last_race_class (GB only, since IRE class data starts July 2026).",
      "Drift pushes runners into 16/1+, where chase error pools have scored near zero (IRE 0 from 64). The broad result can be dragged down or made noisy by longshots, which is why there is a 14/1-or-shorter refinement.",
      "Field size: bigger fields lengthen every price, so some of the drift is mechanical. Compare field_size with last_field_size.",
      "Period effect: remarked_last3 only exists from November 2025, so the window is short and one season could drive the result. Check the split before and after April 2026.",
      "Overlap: this is another subset of the one-off chase-error pool, which has already passed in several overlapping forms. A pass here is not independent confirmation of the base edge. It only adds information if the drift and shortened groups separate.",
      "Region mix: GB (1.24) and IRE (1.19) pools differ a little. Report both regions."
    ],
    "mechanism": "A single jumping error after two clean rounds, followed by a heavy defeat (beaten more than 10 lengths or failing to finish), says little about the horse's real ability. Its true chance today is close to what its earlier form implied. When the price has lengthened since the last run, the market has visibly marked the horse down for that bad run. Those runners should carry the largest gap between price and true chance, so the one-off-error edge should be concentrated in them.",
    "market_miss": "The market prices off the latest form figure and beaten distance. A bad figure lengthens the price. The cause, one blunder after two clean jumping rounds, sits in the in-running comments, which most bettors do not weigh. So the drift reflects the result and not the reason for it.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.30 to 1.45 on about 250 to 350 bets, against about 1.22 for the whole one-off-error pool",
      "falsified_if": "A/E at or below 1.224 for the drifted pool, or the control group (price shortened or unchanged) scoring the same or higher than the drifted pool"
    },
    "notes": "The rule sketch is kept unchanged. All columns are pre_off, and remarked_last3 limits the window to November 2025 onward in both regions. sp_decimal against last_sp_decimal compares today's starting price with the last run's. Both are known at the off, so the result is not read. min_ae is raised to 1.15 because the base pool already scores about 1.22. A pass only means something if the drift group clearly beats the shortened-price control."
  }

Timeline

1. Job 2299 finished

   17 Oct, 01:36 · engine:compute-1

2. forward result for H-373590f6 (forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}): info

   17 Oct, 01:36 · engine:evaluator

3. compute-1 took job 2299 (forward_check, attempt 1)

   17 Oct, 01:36 · compute-1

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

   17 Oct, 01:36 · engine:scheduler

5. Job 2283 finished

   17 Oct, 01:35 · engine:compute-1

6. H-373590f6 Frozen → SHADOW vault skipped: remarked_last3, jumping_errors_last3, last_jumping_error, last_beaten_lengths do not exist before the discovery zone; track it forward instead

   17 Oct, 01:35 · engine:orchestrator

7. compute-1 took job 2283 (vault, attempt 1)

   17 Oct, 01:35 · compute-1

8. Job 2274 finished

   17 Oct, 01:35 · engine:compute-1

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

   17 Oct, 01:35 · engine:scheduler

10. H-373590f6 Falsification → FROZEN survived falsification

    17 Oct, 01:35 · engine:falsification

11. falsification result for H-373590f6 (partition: Handicaps only): pass

    17 Oct, 01:35 · engine:evaluator

12. compute-1 took job 2274 (attack, attempt 1)

    17 Oct, 01:35 · compute-1

13. Job 2273 finished

    17 Oct, 01:35 · engine:compute-1

14. falsification result for H-373590f6 (partition: GB drifters not rising in class): pass

    17 Oct, 01:35 · engine:evaluator

15. compute-1 took job 2273 (attack, attempt 1)

    17 Oct, 01:35 · compute-1

16. Job 2272 finished

    17 Oct, 01:35 · engine:compute-1

17. falsification result for H-373590f6 (partition: Drift not caused by a bigger field): pass

    17 Oct, 01:35 · engine:evaluator

18. compute-1 took job 2272 (attack, attempt 1)

    17 Oct, 01:35 · compute-1

19. Job 2271 finished

    17 Oct, 01:35 · engine:compute-1

20. falsification result for H-373590f6 (control: Control: drifted after a heavy chase defeat with NO jumping error): pass

    17 Oct, 01:35 · engine:evaluator