Big-yard runners at the front of the market in big handicaps beat their price

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-caaeaf8e (Rival: Irish par is a big-yard effect, not a country effect): the country 'GB' carries the effect (A/E 1.37) while the rest score IRE 0.88. O-35669ef1
   • H-caaeaf8e (Rival: Irish par is a big-yard effect, not a country effect) pointed the predicted way but failed the bar: A/E 1.03 on 468 bets (p 0.43), control 1.01. O-a921f96e
2. AskedDo big-yard top-two runners at 5/1+ in 14+ runner handicaps (trainer_runs >= 1500) beat their price, as the arithmetic implies, and is that true in Ireland as well as GB? answered
3. MechanismBig yards often aim their handicappers at the big-field targets and often run several horses in the same race. Stable money and the market's attention are split across those entries, so the one that ends up near the top of the betting is still backed less than its chance deserves.
4. Why the market misses itWith several stable entries, the form book cannot show which one the yard is targeting. The price is set against a horse's own exposed form rather than how the yard plans its season, and the 5/1+ price makes it look like 'just another' entry from a big string.
5. Predictedpositive, A/E 1.10 to 1.25 on 700+ bets, with IRE above 1.0
6. Falsified ifA/E 1.05 or below on 500+ bets, or IRE at 1.00 or below while GB stays high (which would make it the same country effect again)

Mechanism and rationale

Big yards often aim their handicappers at the big-field targets and often run several horses in the same race. Stable money and the market's attention are split across those entries, so the one that ends up near the top of the betting is still backed less than its chance deserves.

The rule

is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 1500

Reads: is_handicap, sp_rank, sp_decimal, field_size, trainer_runs

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

19972ef189b94e55eabb311b32f4bcbd1fc1f4dd43bfa6f0ac66a8d32bca887c

Frozen

16 Oct, 16:00

Nothing in the definition can change after this.

Confounders

• GB/Ireland country split: every prior big-handicap short-end pool lifted in GB (~1.4) and sat at par in IRE (~0.95-0.99); a big-yard result carried by GB alone is the same country effect, not a yard effect
• trainer_runs is cumulative from the 2020-07-01 data start, so the >=1500 filter is time-varying: early on only the very largest yards qualify and the qualifying set grows later. The pool is therefore weighted to the later period, where the parent lift faded (0.97); read by period and by trainer
• Irish festival handicaps (Galway, Punchestown, Leopardstown, Fairyhouse) where big Irish yards run many entries and the market is heavily informed
• Concentration: a handful of the largest yards (e.g. one or two Irish/GB mega-stables) may supply most bets; check per-trainer contribution
• Pool overlaps heavily with the second-favourite rival (H-601a8566) and favourite rule (H199); not an independent replication
• Price basis: A/E measured at BSP may not survive at SP

Control

{
  "description": "Same-race other runners (lab default) should sit near par (~1.00-1.02, as in earlier big-handicap controls at 1.008-1.011); if they are also lifted, the effect is a race/overround property, not a big-yard one. The decisive comparison is the complementary small-yard pool with the identical market filter (already seen at 1.03 on 468): the mechanism requires the big-yard pool to clearly beat it. Every result must be read split by GB vs IRE and early vs late period.",
  "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs < 1500"
}

Discovery result

pass broad rule · 16 Oct, 15:56

Bets

386

58 winners

A/E

1.33

95%: 1.01 to 1.71

SP return

+1.6%

level stakes

Betfair SP return

+27.4%

371 bets with a BSP

Strike rate 15.0%. p = 0.022. 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. 5 of 6 survived.

• perturbation: Yard-size cut raised to 1800 runspass

  345 bets · A/E 1.38 (1.04 to 1.80) · control 0.96 · 16 Oct, 16:00

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271571.65

  after 2026-04-271881.17

• perturbation: Yard-size cut lowered to 1200 runspass

  451 bets · A/E 1.30 (1.01 to 1.66) · control 0.97 · 16 Oct, 16:00

Sealed holdout (vault)

Bets

1,003

16 Oct, 16:00

A/E

1.14

p = 0.075

BH threshold then

0.0022

over every vault test so far

Outcome

Failed

Forward record

• forward vs bar {'min_bets': 150, 'min_ae': 1.0, 'max_p': 0.2}info

  0 bets · · 16 Oct, 16:00

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: <= 2 -> 1 (-20%)pass

  54 bets · A/E 1.06 (0.43 to 2.19) · control 1.01 · 16 Oct, 15:56

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-2722–

  after 2026-04-2732–

• neighbour: <= 2 -> 3 (+20%)pass

  887 bets · A/E 1.15 (0.94 to 1.39) · control 0.99 · 16 Oct, 15:56

  Breakdown

  bets

Lineage

Agent runs

• designer v1ok · 1m 9s · 16 Oct, 15:55

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 1500",
    "refinements": [
      {
        "label": "multiple stable entries (direct test of the split-attention mechanism)",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 1500 AND n_stable >= 2"
      },
      {
        "label": "yard's market choice among multiple entries",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs >= 1500 AND n_stable >= 2 AND stable_rank = 1"
      }
    ],
    "control": {
      "description": "Same-race other runners (lab default) should sit near par (~1.00-1.02, as in earlier big-handicap controls at 1.008-1.011); if they are also lifted, the effect is a race/overround property, not a big-yard one. The decisive comparison is the complementary small-yard pool with the identical market filter (already seen at 1.03 on 468): the mechanism requires the big-yard pool to clearly beat it. Every result must be read split by GB vs IRE and early vs late period.",
      "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs < 1500"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.08,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "GB/Ireland country split: every prior big-handicap short-end pool lifted in GB (~1.4) and sat at par in IRE (~0.95-0.99); a big-yard result carried by GB alone is the same country effect, not a yard effect",
      "trainer_runs is cumulative from the 2020-07-01 data start, so the >=1500 filter is time-varying: early on only the very largest yards qualify and the qualifying set grows later. The pool is therefore weighted to the later period, where the parent lift faded (0.97); read by period and by trainer",
      "Irish festival handicaps (Galway, Punchestown, Leopardstown, Fairyhouse) where big Irish yards run many entries and the market is heavily informed",
      "Concentration: a handful of the largest yards (e.g. one or two Irish/GB mega-stables) may supply most bets; check per-trainer contribution",
      "Pool overlaps heavily with the second-favourite rival (H-601a8566) and favourite rule (H199); not an independent replication",
      "Price basis: A/E measured at BSP may not survive at SP"
    ],
    "mechanism": "Big yards target their handicappers at big-field handicaps and often run several horses in the same race (n_stable >= 2). Stable money and market attention are split across those entries, so the runner that still ends up first or second in the betting at 5/1+ is backed less than its true chance. If this is the mechanism, the lift should be in big-yard runners (not the small-yard complement), strongest where the yard has multiple runners, and present in Ireland as well as GB.",
    "market_miss": "With several stable entries the form book cannot show which one the yard is targeting; prices are set off each horse's own exposed form rather than the yard's season plan, and at 5/1+ a big-string runner reads as 'just another' entry, so the targeted horse is under-bet.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.25 on 700+ bets, above the small-yard control (1.03), with IRE above 1.00",
      "falsified_if": "A/E 1.05 or below on 500+ bets; or A/E not clearly above the small-yard control; or IRE at 1.00 or below while GB stays high (the same country effect again); or the lift confined to the early period"
    },
    "notes": "I checked the columns: trainer_runs, sp_decimal, sp_rank, n_stable and stable_rank are all pre_off from 2020-07-01 for GB and IRE. I kept the broad rule exactly as sketched. Refinements use n_stable and stable_rank because they test the multiple-entry part of the mechanism directly. Since the small-yard complement (trainer_runs < 1500) scored 1.03 on 468, the parent pool would have to be roughly 1.25+ on the big-yard side for the blend to match the parent figures (around 1.2). So a sizeable broad-rule A/E is expected even without a yard effect. That makes the GB/IRE split, the period split and the per-trainer split the real test. I set the pass bar stricter than the lab floor (300 bets, A/E 1.08) because the overlap with already-measured pools leaves little new information."
  }

Timeline

1. Job 1953 finished

   16 Oct, 16:00 · engine:compute-1

2. forward result for H-659f3a37 (forward vs bar {'min_bets': 150, 'min_ae': 1.0, 'max_p': 0.2}): info

   16 Oct, 16:00 · engine:evaluator

3. compute-1 took job 1953 (forward_check, attempt 1)

   16 Oct, 16:00 · compute-1

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

   16 Oct, 16:00 · engine:scheduler

5. Job 1952 finished

   16 Oct, 16:00 · engine:compute-1

6. H-659f3a37 Vault → SHADOW vault SUPPORTIVE: A/E 1.14 on 1003 bets, p 0.07478 against BH threshold 0.002174 over 46 vault tests

   16 Oct, 16:00 · engine:orchestrator

7. vault result for H-659f3a37 (vault query (supportive)): pass

   16 Oct, 16:00 · engine:orchestrator

8. H-659f3a37 Frozen → FROZENvault vault query started

   16 Oct, 16:00 · engine:orchestrator

9. compute-1 took job 1952 (vault, attempt 1)

   16 Oct, 16:00 · compute-1

10. Job 1951 finished

    16 Oct, 16:00 · engine:compute-1

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

    16 Oct, 16:00 · engine:scheduler

12. H-659f3a37 Falsification → FROZEN survived falsification

    16 Oct, 16:00 · engine:falsification

13. falsification result for H-659f3a37 (partition: Jumps-only handicaps (Hurdle/Chase)): fail

    16 Oct, 16:00 · engine:evaluator

14. compute-1 took job 1951 (attack, attempt 1)

    16 Oct, 16:00 · compute-1

15. Job 1950 finished

    16 Oct, 16:00 · engine:compute-1

16. falsification result for H-659f3a37 (control: Big-yard second favourites in 14+ runner handicaps vs the yard-blind rival): pass

    16 Oct, 16:00 · engine:evaluator

17. compute-1 took job 1950 (attack, attempt 1)

    16 Oct, 16:00 · compute-1

18. Job 1949 finished

    16 Oct, 16:00 · engine:compute-1

19. falsification result for H-659f3a37 (partition: Big-yard favourites only (strip out second favourites)): pass

    16 Oct, 16:00 · engine:evaluator