Rival: GB no-drop chase lift is a GB-region effect, not class choice

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-5a97af88 (Replicate: one-off jumping error, well beaten, in hurdles under 16/1): the country 'GB' carries the effect (A/E 1.16) while the rest score IRE 0.94. O-7080bb8f
   • H-c2928b4e (Generalise: one-off chase error, well beaten, not dropping in class (G) passed discovery: A/E 1.36 on 253 bets (p 0.041), same-race control 0.96. O-bc889498
2. AskedIs the class 'no-drop' lift really a GB-only effect, because race_class is missing for IRE before July 2026 and the class condition removes Irish runners? answered
3. Mechanismrace_class is null for IRE before July 2026, so the no-drop condition also removes Irish runners. If Irish chasers in this story are priced efficiently, or the Irish error remarks are less reliable, removing them alone would raise A/E. The class condition would then be getting the credit for a country split.
4. Why the market misses itUnder this rival, there is no class-specific miss: GB books under-react to the error remark in general, and the class filter adds nothing.
5. Predictedpositive, A/E 1.30 to 1.36 on about 300 bets if region does the work
6. Falsified ifThe GB full pool scores 1.25 or lower while no-drop holds at 1.36. That would rule the region rival out and leave the class choice standing.

Mechanism and rationale

race_class is null for IRE before July 2026, so the no-drop condition also removes Irish runners. If Irish chasers in this story are priced efficiently, or the Irish error remarks are less reliable, removing them alone would raise A/E. The class condition would then be getting the credit for a country split.

The rule

race_type = 'Chase' AND region = 'GB' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)

Reads: race_type, region, remarked_last3, jumping_errors_last3, last_jumping_error, last_position, last_beaten_lengths

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

eaa56fce2c6218027b8c944317946f44743c8ab4b1ddd3d4c5140c34378429df

Frozen

17 Oct, 01:26

Nothing in the definition can change after this.

Confounders

• Small class-dropper sample (about 40 to 60). The full-pool vs no-drop gap may be noise, since 40 runners move the pooled A/E only a little.
• Heavy overlap with the five or six frozen chase-error subsets. This is the same evidence re-cut, not an independent confirmation.
• IRE race_class only starts in July 2026, so class and region cannot be separated inside Ireland in the discovery window.
• Irish error remarks may be coded or written differently, so remarked_last3 and jumping_errors_last3 may not mean the same thing across regions.
• Earlier partitions found the close-run lift concentrated in recent winners and GB placed losers at 0.52 on 62. Sub-pool composition can drive the result more than region does.

Control

{
  "description": "Same-race other runners should score about 0.95 to 1.00, as they have in earlier chase-error tests (0.96), which shows the lift is not a property of the race. The comparison that matters for this rival is the same story in Ireland: if region does the work, IRE one-off-error chasers should score near 1.0 or below while the GB full pool holds about 1.30. If IRE also scores about 1.3, then removing Irish runners does not explain the no-drop lift, and the rival fails on that side as well.",
  "rule": "race_type = 'Chase' AND region = 'IRE' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)"
}

Discovery result

pass broad rule (niche) · 17 Oct, 01:23

Bets

347

53 winners

A/E

1.24

95%: 0.93 to 1.62

SP return

+8.1%

level stakes

Betfair SP return

+28.1%

346 bets with a BSP

Strike rate 15.3%. p = 0.073. Same-race control: the other runners in these races had A/E 0.97, 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.

• control: Irish twin of the same story (the region ingredient)pass

  189 bets · A/E 1.19 (0.73 to 1.84) · control 0.98 · 17 Oct, 01:26

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27611.16

  after 2026-04-271281.20

• partition: GB class droppers only (the slice the no-drop filter removed)pass

  83 bets · A/E 1.01 (0.56 to 1.70) · control 1.00 · 17 Oct, 01:26

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, 01:26

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: > 10 -> 8 (-20%)pass

  381 bets · A/E 1.13 (0.85 to 1.47) · control 0.99 · 17 Oct, 01:23

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272181.04

  after 2026-04-271631.24

• neighbour: > 10 -> 12 (+20%)pass

  328 bets · A/E 1.22 (0.91 to 1.62) · control 0.97 · 17 Oct, 01:23

  Breakdown

Lineage

Agent runs

• designer v1ok · 45s · 17 Oct, 01:21

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND region = 'GB' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)",
    "refinements": [
      {
        "label": "GB class droppers only (the part the no-drop filter removed)",
        "rule": "race_type = 'Chase' AND region = 'GB' 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 race_class IS NOT NULL AND last_race_class IS NOT NULL AND race_class > last_race_class"
      },
      {
        "label": "GB no-drop subset, re-measured alongside the full pool for a direct comparison",
        "rule": "race_type = 'Chase' AND region = 'GB' 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 race_class IS NOT NULL AND last_race_class IS NOT NULL AND race_class <= last_race_class"
      }
    ],
    "control": {
      "description": "Same-race other runners should score about 0.95 to 1.00, as they have in earlier chase-error tests (0.96), which shows the lift is not a property of the race. The comparison that matters for this rival is the same story in Ireland: if region does the work, IRE one-off-error chasers should score near 1.0 or below while the GB full pool holds about 1.30. If IRE also scores about 1.3, then removing Irish runners does not explain the no-drop lift, and the rival fails on that side as well.",
      "rule": "race_type = 'Chase' AND region = 'IRE' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)"
    },
    "pass_criteria": {
      "min_bets": 200,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Small class-dropper sample (about 40 to 60). The full-pool vs no-drop gap may be noise, since 40 runners move the pooled A/E only a little.",
      "Heavy overlap with the five or six frozen chase-error subsets. This is the same evidence re-cut, not an independent confirmation.",
      "IRE race_class only starts in July 2026, so class and region cannot be separated inside Ireland in the discovery window.",
      "Irish error remarks may be coded or written differently, so remarked_last3 and jumping_errors_last3 may not mean the same thing across regions.",
      "Earlier partitions found the close-run lift concentrated in recent winners and GB placed losers at 0.52 on 62. Sub-pool composition can drive the result more than region does."
    ],
    "mechanism": "race_class is null for IRE before July 2026, so any no-drop condition also quietly removes all Irish runners. If Irish one-off-error chasers are priced efficiently, or their error remarks are less reliable, removing them would raise A/E on its own. The class condition would then be getting credit for a country split. Under this rival, the GB one-off-error pool with class droppers left in should score close to the no-drop subset (about 1.30 to 1.36), and IRE runners should score near 1.0.",
    "market_miss": "GB books under-react to a single jumping-error remark in a well-beaten chase run in general. They treat the error as lost form when it hid the horse's real ability. The trainer's class choice adds nothing beyond that. Irish markets, or Irish remark conventions, do not show the same mispricing.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.30 to 1.36 on roughly 280 to 320 GB bets, with the IRE control near 1.0",
      "falsified_if": "The GB full pool scores A/E 1.25 or lower while the no-drop subset holds near 1.36. That would rule the region rival out and leave the class choice standing. It is also falsified if the IRE control scores as high as the GB pool, because then region is not what separates them."
    },
    "notes": "I kept the rule sketch unchanged. It uses only pre-off columns and no race_class, so it is the GB pool with droppers included. Pass criteria are tighter than the lab floor (200 bets, A/E 1.10) because this pool overlaps heavily with subsets that already passed, so a marginal pass would add little. Reading: if the full pool is about 1.3 or higher and IRE is about 1.0, region explains the lift. If the full pool is 1.25 or lower and the no-drop subset holds, class choice stands. The dropper refinement checks whether droppers stay near 1.0 (earlier: 0.997 on 40)."
  }

Timeline

1. Job 2240 finished

   17 Oct, 01:26 · engine:compute-1

2. forward result for H-db7f42b8 (forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}): info

   17 Oct, 01:26 · engine:evaluator

3. compute-1 took job 2240 (forward_check, attempt 1)

   17 Oct, 01:26 · compute-1

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

   17 Oct, 01:26 · engine:scheduler

5. Job 2238 finished

   17 Oct, 01:26 · engine:compute-1

6. H-db7f42b8 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, 01:26 · engine:orchestrator

7. compute-1 took job 2238 (vault, attempt 1)

   17 Oct, 01:26 · compute-1

8. Job 2237 finished

   17 Oct, 01:26 · engine:compute-1

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

   17 Oct, 01:26 · engine:scheduler

10. H-db7f42b8 Falsification → FROZEN survived falsification

    17 Oct, 01:26 · engine:falsification

11. falsification result for H-db7f42b8 (partition: Handicap chases at 14/1 or shorter only): pass

    17 Oct, 01:26 · engine:evaluator

12. compute-1 took job 2237 (attack, attempt 1)

    17 Oct, 01:26 · compute-1

13. Job 2236 finished

    17 Oct, 01:26 · engine:compute-1

14. falsification result for H-db7f42b8 (perturbation: Beaten-lengths threshold tightened to more than 12.5): pass

    17 Oct, 01:26 · engine:evaluator

15. compute-1 took job 2236 (attack, attempt 1)

    17 Oct, 01:26 · compute-1

16. Job 2235 finished

    17 Oct, 01:26 · engine:compute-1

17. falsification result for H-db7f42b8 (perturbation: Beaten-lengths threshold loosened to more than 8): pass

    17 Oct, 01:26 · engine:evaluator

18. compute-1 took job 2235 (attack, attempt 1)

    17 Oct, 01:26 · compute-1

19. Job 2234 finished

    17 Oct, 01:26 · engine:compute-1

20. falsification result for H-db7f42b8 (control: No jumping error: GB chasers well beaten last time with a clean record): pass

    17 Oct, 01:26 · engine:evaluator