Rival: yard quality, not yard size, drives the big-handicap top-two 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-659f3a37 (Big-yard runners at the front of the market in big handicaps beat thei) passed discovery: A/E 1.33 on 386 bets (p 0.022), same-race control 0.96. O-fe42f7ad
2. AskedIs the big-yard top-two lift in big handicaps carried by a few trainers (for example two Irish mega-yards), or spread across many big yards in both countries? testing
3. MechanismStrong yards get their handicappers into a big-field target well placed and fit. At 5/1+ the market prices the horse's own exposed form and gives too little weight to how reliably the yard produces a horse ready for that day. Size is only a proxy: big yards tend to be strong ones.
4. Why the market misses itThe form book shows the horse's own form figures and handicap mark, not the yard's record of placing horses to win big handicaps. At 5/1+ in a 14+ runner field the horse looks like one of many, so the yard's strength is under-weighted.
5. Predictedpositive, A/E 1.10 to 1.25 on 400+ bets, with IRE above 1.00
6. Falsified ifA/E 1.04 or below on 300+ bets, or IRE at 1.00 or below, which would mean yard size (or particular trainers), not strike rate, is the lever
7. Rule sketchis_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 300 AND trainer_wins >= 0.14 * trainer_runs

Mechanism and rationale

Strong yards get their handicappers into a big-field target well placed and fit. At 5/1+ the market prices the horse's own exposed form and gives too little weight to how reliably the yard produces a horse ready for that day. Size is only a proxy: big yards tend to be strong ones.

The rule

is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 300 AND trainer_wins >= 0.14 * trainer_runs

Reads: is_handicap, sp_rank, sp_decimal, field_size, trainer_runs, trainer_wins

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

0f563ee4af531f8aac6e43ce23cdddbb7dfa5811a1cce430ee2df2e06bd1324d

Frozen

not frozen

The definition can still change.

Confounders

• Overlap with the big-yard pool (H-659f3a37): most strong yards are big, so the broad rule alone cannot separate quality from size. The strong-small refinement and the big-weak control do that job.
• trainer_runs and trainer_wins are cumulative from 2020-07, so strike rates in the early months are noisy. trainer_runs >= 300 limits this, and the period split should be checked.
• GB/IRE mix: small Irish yards scored 0.88 and the Irish short end has been at par throughout. The IRE split (via course) must sit above 1.00 or the prediction fails.
• Concentration: a few high-strike yards could carry the result, so check the share of bets and profit from the top 3 trainers.
• Price basis: the big-yard pool showed a BSP lift against SP ROI near zero, so A/E against SP may overstate how much of the edge can actually be taken.
• Time fade: the related second-favourite and small-yard pools faded from the early period to the later one.
• Strike rate covers all race types, not handicaps alone, so a yard strong in maidens or novice races may count as strong here without being good at placing handicappers.

Discovery result

fail broad rule · 16 Oct, 16:00

Bets

162

20 winners

A/E

1.10

95%: 0.67 to 1.70

SP return

-17.6%

level stakes

Betfair SP return

-0.7%

156 bets with a BSP

Strike rate 12.3%. p = 0.37. Same-race control: the other runners in these races had A/E 1.01, 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 · 37s · 16 Oct, 15:58

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 300 AND trainer_wins >= 0.14 * trainer_runs",
    "refinements": [
      {
        "label": "Strong but not big yards (the discriminating test against yard size)",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 300 AND trainer_wins >= 0.14 * trainer_runs AND trainer_runs < 1500"
      },
      {
        "label": "Second favourites only, where the big-yard lift was found to sit",
        "rule": "is_handicap = 1 AND sp_rank = 2 AND is_favourite = 0 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 300 AND trainer_wins >= 0.14 * trainer_runs"
      },
      {
        "label": "Stricter strike rate (16%+), a dose-response check on yard quality",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 300 AND trainer_wins >= 0.16 * trainer_runs"
      }
    ],
    "control": {
      "description": "Same-race control: the other runners in these races should sit near par (about 0.95 to 1.01, as in the earlier big-yard control at 0.965), so any lift belongs to the selection and not to the race type. Rival-group control: big yards with a low strike rate, in the same price and field slots. If strike rate is the lever, these should be near par (1.00 or below) while the strong-yard rule is lifted. If they are lifted as much as the strong-yard pool, yard size is the lever and this rival fails.",
      "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 1500 AND trainer_wins < 0.14 * trainer_runs"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Overlap with the big-yard pool (H-659f3a37): most strong yards are big, so the broad rule alone cannot separate quality from size. The strong-small refinement and the big-weak control do that job.",
      "trainer_runs and trainer_wins are cumulative from 2020-07, so strike rates in the early months are noisy. trainer_runs >= 300 limits this, and the period split should be checked.",
      "GB/IRE mix: small Irish yards scored 0.88 and the Irish short end has been at par throughout. The IRE split (via course) must sit above 1.00 or the prediction fails.",
      "Concentration: a few high-strike yards could carry the result, so check the share of bets and profit from the top 3 trainers.",
      "Price basis: the big-yard pool showed a BSP lift against SP ROI near zero, so A/E against SP may overstate how much of the edge can actually be taken.",
      "Time fade: the related second-favourite and small-yard pools faded from the early period to the later one.",
      "Strike rate covers all race types, not handicaps alone, so a yard strong in maidens or novice races may count as strong here without being good at placing handicappers."
    ],
    "mechanism": "Strong yards (high strike rate before today) get their handicappers to big-field targets well placed and fit. At 5/1+ in a 14+ runner handicap, the market prices the horse's own exposed form and mark and gives too little weight to how reliably the yard has a horse ready for the day. Yard size, in the form of a cumulative run count, is only a proxy for this, because big yards tend to be strong ones.",
    "market_miss": "The form book shows the horse's own figures and handicap mark, not the yard's record of producing winners. In a 14+ runner handicap a 5/1+ top-two runner looks like one of many, so punters and layers under-weight the yard's strength. This should show most clearly in strong yards that are not big, which do not have a big yard's market profile.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.25 on 400+ bets, with IRE above 1.00 and the strong-but-not-big refinement also above 1.05",
      "falsified_if": "Any of these falsifies it: A/E 1.04 or below on 300+ bets; IRE at 1.00 or below; or the big-but-weak control lifted as much as the strong-yard rule while strong-but-small yards are at par. Each would mean yard size or particular trainers, not strike rate, is the lever."
    },
    "notes": "I kept the rule sketch as written. trainer_runs and trainer_wins are pre-off ('record before today') and available from 2020-07 in both GB and IRE. I raised min_bets to 300 to match the falsification threshold. The key evidence against the size explanation is refinement 1 (strong yards under 1500 runs) set against the control (big yards with a low strike rate). IRE should be split using the course column, as the lab has done before."
  }

Timeline

1. Job 1944 finished

   16 Oct, 16:00 · engine:compute-1

2. H-0939d0ac Discovery → KILLED fails: 162 bets, needs 300; p 0.3671, needs 0.05 or lower; periods [1.29, 0.9], each needs 1.0 (not a niche pass: a period points the other way [1.29, 0.9])

   16 Oct, 16:00 · engine:evaluator

3. discovery result for H-0939d0ac (broad rule): fail

   16 Oct, 16:00 · engine:evaluator

4. H-0939d0ac Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 16:00 · engine:orchestrator

5. compute-1 took job 1944 (evaluate, attempt 1)

   16 Oct, 16:00 · compute-1

6. Job 1942 finished

   16 Oct, 15:58 · engine:ai-1

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

   16 Oct, 15:58 · engine:scheduler

8. H-0939d0ac Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 15:58 · agent:designer@v1

9. ai-1 took job 1942 (design, attempt 1)

   16 Oct, 15:58 · ai-1

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

    16 Oct, 15:58 · engine:scheduler

11. New idea H-0939d0ac: Rival: yard quality, not yard size, drives the big-handicap top-two lift (rival-explanation)

    16 Oct, 15:58 · 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.