Rival: Irish par is a big-yard effect, not a country effect

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-6da42014 (Sharpen: top-two in the betting at 5/1+ in handicaps of 18+ runners): the country 'GB' carries the effect (A/E 1.57) while the rest score IRE 0.95. O-875c02ce
   • H-6da42014 (Sharpen: top-two in the betting at 5/1+ in handicaps of 18+ runners) passed discovery but was falsified: IRE A/E 0.951. O-c4fd3752
   • H-601a8566 (Rival: second favourites at 5/1+ in big handicaps also beat their pric): the country 'GB' carries the effect (A/E 1.4) while the rest score IRE 0.99. O-d291de7f
2. AskedIs the GB/Ireland gap in big-handicap short-end runners a country effect, or a big-yard effect? Irish big handicaps are dominated by a few huge yards whose runners may be over-bet on reputation. answered
3. MechanismIn Irish big handicaps a few huge yards run several horses each, and their best-backed one draws money on the yard's name. That pushes its price below fair and cancels the front-of-market lift. Smaller yards' runners at the top of the market got there on the money alone, without that reputation push.
4. Why the market misses itPunters treat big-yard runners as the informed choice, and the yard's record makes them look stronger than the individual horse's handicap mark warrants.
5. Predictedpositive, A/E 1.08 to 1.18 on 500+ bets, with the IRE subset above 1.03
6. Falsified if

Mechanism and rationale

In Irish big handicaps a few huge yards run several horses each, and their best-backed one draws money on the yard's name. That pushes its price below fair and cancels the front-of-market lift. Smaller yards' runners at the top of the market got there on the money alone, without that reputation push.

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)

df38e3056c2b9a0c7a0837467af56e687d4e7a19eb25a8b720787542e5338d44

Frozen

not frozen

The definition can still change.

Confounders

• trainer_runs is cumulative from July 2020, so it grows over time: a yard below 1500 early may be above it later. The cut sorts yards differently by period, so results must be read by period.
• The 1500 cut is a guess at what counts as a big yard. Refinement 1 (<750) tests sensitivity, and a result that only holds at one cut is fragile.
• Yard size and country are correlated: Irish big handicaps are dominated by a few mega-yards (e.g. Mullins, Elliott). The small-yard IRE subset may be thin, and the overall lift may again come from GB.
• This pool overlaps the earlier H199, second-favourite and top-two pools, so it is not independent of the GB-heavy lift already seen. It can only add information through the IRE small-yard subset and the big-yard complement.
• The second-favourite lift shrank sharply over the long history (1.158 to 1.04 in the vault), so the discovery window may overstate any front-of-market effect, whatever the yard split.
• Small-yard runners at the top of a big-handicap market may differ in other ways (lower class, weaker jockeys, different courses), so a yard split could stand in for race quality.

Discovery result

fail broad rule · 16 Oct, 15:54

Bets

468

55 winners

A/E

1.03

95%: 0.78 to 1.34

SP return

-21.3%

level stakes

Betfair SP return

-3.5%

451 bets with a BSP

Strike rate 11.8%. p = 0.43. 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 · 41s · 16 Oct, 15:54

  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": "smaller yards, tighter cut (sensitivity to the guessed 1500 threshold)",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND trainer_runs < 750"
      },
      {
        "label": "smaller yards at 14/1 or shorter (edges die at big prices; keeps the bet on genuine front-of-market runners)",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND sp_decimal <= 15 AND field_size >= 14 AND trainer_runs < 1500"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) Same-race other runners: expected near par (~1.00), as in earlier big-handicap pools, so any lift is the selection, not the race. (2) The decisive control is the big-yard complement: top-two 5/1+ runners in 14+ runner handicaps from yards with trainer_runs >= 1500. The mechanism predicts these are at or below par (A/E <= 1.02), especially in Ireland. Both the rule and the complement must be read by country (GB vs IRE, via course) and by period, because trainer_runs is cumulative and grows over time. If the complement beats its price as much as the rule in GB, and IRE is at par in both, yard size is not the lever.",
      "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": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "trainer_runs is cumulative from July 2020, so it grows over time: a yard below 1500 early may be above it later. The cut sorts yards differently by period, so results must be read by period.",
      "The 1500 cut is a guess at what counts as a big yard. Refinement 1 (<750) tests sensitivity, and a result that only holds at one cut is fragile.",
      "Yard size and country are correlated: Irish big handicaps are dominated by a few mega-yards (e.g. Mullins, Elliott). The small-yard IRE subset may be thin, and the overall lift may again come from GB.",
      "This pool overlaps the earlier H199, second-favourite and top-two pools, so it is not independent of the GB-heavy lift already seen. It can only add information through the IRE small-yard subset and the big-yard complement.",
      "The second-favourite lift shrank sharply over the long history (1.158 to 1.04 in the vault), so the discovery window may overstate any front-of-market effect, whatever the yard split.",
      "Small-yard runners at the top of a big-handicap market may differ in other ways (lower class, weaker jockeys, different courses), so a yard split could stand in for race quality."
    ],
    "mechanism": "In big Irish handicaps a few mega-yards run several horses each, and the yard's best-backed runner draws money on the trainer's name rather than on the horse's handicap mark. That pushes its price below fair and cancels the front-of-market lift seen in GB. A top-two 5/1+ runner from a smaller yard reached the front of the market on form and informed money alone, without that reputation premium, so it should keep the lift in both countries.",
    "market_miss": "Punters treat a mega-yard's runner as the informed choice in a big handicap, and the yard's overall record makes the horse look stronger than its own mark warrants. That over-bets big-yard runners and leaves smaller-yard runners at the front of the market under-bet relative to it, so their price stays too long.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.08 to 1.18 on 500+ bets, with the IRE small-yard subset above 1.03 and the big-yard complement (trainer_runs >= 1500) at or below 1.02.",
      "falsified_if": "The IRE small-yard subset is at 1.00 or below while GB stays high, which would point to country or market structure rather than yards. Also falsified if the big-yard complement beats its price as much as the rule, or if the overall rule is under 1.05."
    },
    "notes": "I checked the catalogue: trainer_runs is a pre-off column (the trainer's record before today), available in both countries from 2020-07-01. The rule sketch is kept as written. The hypothesis is a rival explanation for the Irish par, so the deciding results are the IRE split of the rule and the big-yard complement control, not the headline A/E alone. The lab should report country (via course) and period splits for both the rule and the complement. min_bets is raised to 300 because the prediction is for 500+ bets and the pool is broad. Refinement 1 tests sensitivity to the guessed yard cut, and refinement 2 caps the price at 15.0 (14/1) for genuine front-of-market runners."
  }

Timeline

1. Job 1935 finished

   16 Oct, 15:54 · engine:compute-1

2. H-caaeaf8e Discovery → KILLED fails: A/E 1.03, needs 1.05; p 0.4305, needs 0.05 or lower; periods [1.22, 0.88], each needs 1.0; A/E 1.03 against same-race control 1.011, needs 0.03 above (not a niche pass: A/E 1.03 below 1.05; does not beat the same-race control 1.011; a period points the other way [1.22, 0.88])

   16 Oct, 15:54 · engine:evaluator

3. discovery result for H-caaeaf8e (broad rule): fail

   16 Oct, 15:54 · engine:evaluator

4. H-caaeaf8e Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 15:54 · engine:orchestrator

5. compute-1 took job 1935 (evaluate, attempt 1)

   16 Oct, 15:54 · compute-1

6. Job 1933 finished

   16 Oct, 15:54 · engine:ai-1

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

   16 Oct, 15:54 · engine:scheduler

8. H-caaeaf8e Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 15:54 · agent:designer@v1

9. ai-1 took job 1933 (design, attempt 1)

   16 Oct, 15:54 · ai-1

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

    16 Oct, 15:53 · engine:scheduler

11. New idea H-caaeaf8e: Rival: Irish par is a big-yard effect, not a country effect (rival-explanation)

    16 Oct, 15:53 · 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.