Rival: turf-to-AW switchers that are not dropping in class are fairly priced, so a class drop drives the lift

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-e2d31d3d (Rival: turf-to-AW switchers after a short gap carry the same lift, so ) was predicted negative but came out the other way: A/E 1.07 on 4045 bets (p 0.089 for the opposite direction). O-c26a3db5
   • H-98ae1db9 (Turf-to-all-weather handicap returners after 150+ days beat their pric) passed discovery as a niche: A/E 1.23 on 654 bets (p 0.06), same-race control 0.97. O-e4c2a982
2. AskedIs the turf-to-AW lift really a class drop, with turf horses dropping into weaker AW handicaps, rather than a surface discount? testing
3. MechanismAW handicaps are mostly lower class than turf handicaps. A horse switching from turf often drops a class or more, and the drop, not the surface, may make it competitive. If the surface discount is the cause, switchers at the same or a higher class should still beat their prices. If the class drop is the cause, they should be fairly priced.
4. Why the market misses itUnder the surface story, punters mark down turf form on the AW even when the class is unchanged. Under the rival, the class drop is the under-weighted signal, and same-class switchers carry no edge.
5. Predictednegative, A/E 0.92 to 1.03 on 1000+ bets
6. Falsified ifA/E of 1.07 or higher on 1000+ bets. That would rule out the class drop and leave the surface discount standing.

Mechanism and rationale

AW handicaps are mostly lower class than turf handicaps. A horse switching from turf often drops a class or more, and the drop, not the surface, may make it competitive. If the surface discount is the cause, switchers at the same or a higher class should still beat their prices. If the class drop is the cause, they should be fairly priced.

The rule

race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND region = 'GB' AND race_class <= last_race_class

Reads: race_type, is_handicap, surface, last_surface, region, race_class, last_race_class

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

d837861e9f68b0f90b6b08a9bbfe23130ae05328d5990baa57bfb71dc091153d

Frozen

not frozen

The definition can still change.

Confounders

• race_class covers Ireland only from 2026-07, so the rule is GB only. The Irish under-performance seen in this thread cannot be checked here.
• Class bands differ between the turf and AW programmes, so the same number may mean a different strength of race. A 'same class' move may really be a soft drop.
• Same-class switchers may be chasing ground, a draw or a winter campaign rather than an easier race. That changes their intent, not their class.
• The layoff mix differs between the class-drop and same-class groups. The original 1.23 lift was for 150+ day returners, so refinement 1 checks the gap separately.
• Field size and overround differ by class, which affects A/E across the groups.
• The original long-gap lift leaned on 16/1+ runners (1.496 on 11 wins). A price-mix difference between the groups could mimic a class effect.
• A missing last_race_class drops a runner from both the rule and the control. If it is missing for some kinds of runners and not others, the samples are biased.

Control

{
  "description": "Two comparisons. (1) The other runners in the same GB AW handicaps should be close to 1.0 (about 0.95 to 1.00). That shows the races are not mispriced as a whole. (2) The key comparison is GB turf-to-AW switchers that do drop in class (race_class > last_race_class). Under the rival they should carry the lift, at A/E 1.07 or higher, while the same-or-higher-class group sits at about 1.0. If both groups score about the same and above 1.05, the class drop is not the driver and the surface discount stands. If the class-drop group is also about 1.0, the GB-only, all-gap switcher lift is weaker than the pooled findings suggested.",
  "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND region = 'GB' AND race_class > last_race_class"
}

Discovery result

fail broad rule · 16 Oct, 14:13

Bets

3,797

387 winners

A/E

1.08

95%: 0.98 to 1.20

SP return

-20.9%

level stakes

Betfair SP return

-0.9%

3,504 bets with a BSP

Strike rate 10.2%. p = 0.065. 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

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

Agent runs

• designer v1ok · 41s · 16 Oct, 14:07

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND region = 'GB' AND race_class <= last_race_class",
    "refinements": [
      {
        "label": "Long-gap same-or-higher-class switchers (the setting of the original 1.23 lift)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND region = 'GB' AND race_class <= last_race_class AND days_since_last_run >= 150"
      },
      {
        "label": "Strictly up in class: the cleanest test of the surface discount with no class help at all",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND region = 'GB' AND race_class < last_race_class"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) The other runners in the same GB AW handicaps should be close to 1.0 (about 0.95 to 1.00). That shows the races are not mispriced as a whole. (2) The key comparison is GB turf-to-AW switchers that do drop in class (race_class > last_race_class). Under the rival they should carry the lift, at A/E 1.07 or higher, while the same-or-higher-class group sits at about 1.0. If both groups score about the same and above 1.05, the class drop is not the driver and the surface discount stands. If the class-drop group is also about 1.0, the GB-only, all-gap switcher lift is weaker than the pooled findings suggested.",
      "rule": "race_type = 'Flat' AND is_handicap = 1 AND surface = 'AW' AND last_surface = 'Turf' AND region = 'GB' AND race_class > last_race_class"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "race_class covers Ireland only from 2026-07, so the rule is GB only. The Irish under-performance seen in this thread cannot be checked here.",
      "Class bands differ between the turf and AW programmes, so the same number may mean a different strength of race. A 'same class' move may really be a soft drop.",
      "Same-class switchers may be chasing ground, a draw or a winter campaign rather than an easier race. That changes their intent, not their class.",
      "The layoff mix differs between the class-drop and same-class groups. The original 1.23 lift was for 150+ day returners, so refinement 1 checks the gap separately.",
      "Field size and overround differ by class, which affects A/E across the groups.",
      "The original long-gap lift leaned on 16/1+ runners (1.496 on 11 wins). A price-mix difference between the groups could mimic a class effect.",
      "A missing last_race_class drops a runner from both the rule and the control. If it is missing for some kinds of runners and not others, the samples are biased."
    ],
    "mechanism": "AW handicaps are mostly lower class than turf handicaps, so a horse switching from turf often drops a class or more. The rival says the drop, not the surface, makes the switcher competitive. If the market under-weights class drops, the turf-to-AW lift (1.23 on 654 after 150+ days; 1.07 on 4045 after short gaps) should sit in the class-drop group. GB switchers running at the same or a higher class should then be fairly priced. If the market instead marks down turf form on the AW, same-or-higher-class switchers should still beat their prices.",
    "market_miss": "Under the rival, punters price the AW switcher on its turf form figures and under-weight how much easier an AW class is. Same-class switchers get no such help and carry no edge. Under the surface story, punters discount turf form on the AW even when the class is unchanged, so same-class switchers would keep the edge.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.92 to 1.03 on 1000+ bets: fairly priced or slightly over-bet, with the class-drop control at 1.07 or higher",
      "falsified_if": "A/E of 1.07 or higher on 1000+ bets for same-or-higher-class GB switchers. That rules out the class drop as the driver and leaves the surface discount standing. It is also falsified if the class-drop control is no higher than this group."
    },
    "notes": "I checked the columns: surface, last_surface and last_race_class are pre_off (GB from 2020-07); race_class is GB from 2020-07. Class 1 is the top, so race_class <= last_race_class means the same class or a higher one, i.e. no class drop. The rule keeps all gaps, as the rule sketch does, because the question covers the turf-to-AW lift as a whole; the 150+ day setting is refinement 1. This is a rival-explanation test with a negative direction, so the result that matters is the contrast with the class-drop control. Min_bets is raised to 300 because the claim needs a well-powered null."
  }

Timeline

1. Job 1752 finished

   16 Oct, 14:13 · engine:compute-1

2. H-f370f2f4 Discovery → KILLED fails: A/E 1.082, needs 0.95 or lower; p 0.941 (lower tail), needs 0.05; periods [1.22, 0.99], each needs 0.99 or lower; A/E 1.082 against same-race control 0.972, needs 0.03 below

   16 Oct, 14:13 · engine:evaluator

3. discovery result for H-f370f2f4 (broad rule): fail

   16 Oct, 14:13 · engine:evaluator

4. H-f370f2f4 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 14:13 · engine:orchestrator

5. compute-1 took job 1752 (evaluate, attempt 1)

   16 Oct, 14:13 · compute-1

6. Job 1750 finished

   16 Oct, 14:08 · engine:ai-1

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

   16 Oct, 14:08 · engine:scheduler

8. H-f370f2f4 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 14:08 · agent:designer@v1

9. ai-1 took job 1750 (design, attempt 1)

   16 Oct, 14:07 · ai-1

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

    16 Oct, 14:07 · engine:scheduler

11. New idea H-f370f2f4: Rival: turf-to-AW switchers that are not dropping in class are fairly priced, so a class drop drives the lift (rival-explanation)

    16 Oct, 14:07 · agent:investigator@v1

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