Replicate: H199 favourites at 5/1+ in 14+ runner handicaps at GB courses only

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-601a8566 (Rival: second favourites at 5/1+ in big handicaps also beat their pric) met the sealed vault supportive: A/E 1.04 on 4207 bets (p 0.1988, threshold 0.0022). O-48574e29
   • 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 live H199 system's lift also carried by GB alone, with Irish big-handicap favourites at 5/1+ sitting at par, as the second-favourite rival showed (GB 1.40 vs IRE 0.99)? testing
3. MechanismGB big-field handicaps are priced by a deep off-course book with generous each-way terms. Money spreads across the field, and the bookmakers put their margin on the front of the market less heavily than on the long tail. The horse the market settles on at 5/1+ ends up a little too long. Irish big handicaps form their prices differently (smaller ring, more informed stable money) and come out at par.
4. Why the market misses itSP at 5/1+ is set against same-priced horses that are fourth or fifth choice in smaller fields. Pooled GB+IRE figures hide the GB edge, so neither the market nor the form book singles it out.

Mechanism and rationale

GB big-field handicaps are priced by a deep off-course book with generous each-way terms. Money spreads across the field, and the bookmakers put their margin on the front of the market less heavily than on the long tail. The horse the market settles on at 5/1+ ends up a little too long. Irish big handicaps form their prices differently (smaller ring, more informed stable money) and come out at par.

The rule

is_handicap = 1 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 14 AND course NOT IN ('LEOPARDSTOWN','CURRAGH','THE CURRAGH','PUNCHESTOWN','FAIRYHOUSE','GALWAY','NAAS','GOWRAN PARK','DUNDALK','CORK','LIMERICK','NAVAN','LISTOWEL','TRAMORE','KILLARNEY','BALLINROBE','BELLEWSTOWN','CLONMEL','DOWN ROYAL','DOWNPATRICK','KILBEGGAN','LAYTOWN','ROSCOMMON','SLIGO','THURLES','TIPPERARY','WEXFORD') AND course NOT LIKE '%(IRE)%' AND course NOT LIKE 'DUNDALK%' AND course NOT LIKE '%CURRAGH%'

Reads: is_handicap, is_favourite, sp_decimal, field_size, course

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

56393ab226bb9d1350040407995a5ecb1f2e3efb0ca6befff5464e7f07ec9976

Frozen

not frozen

The definition can still change.

Confounders

• The GB/IRE split was found after the fact in overlapping pools (H199, H235, H-601a8566, H-6da42014 all share runners), so a GB-only discovery lift is partly built in. The full-history vault is the real test.
• Course-name exclusion may leak Irish meetings with variant spellings (e.g. 'DUNDALK (AW)', '(IRE)' suffixes). Before reading the result, the lab should check that the Irish complement control holds every IRE race in the H199 pool and the main rule holds none.
• GB festival handicaps (Cheltenham, Aintree) may carry the lift. Refinement 2 tests this.
• The edge may exist at BSP but not at SP: a 5/1+ favourite in a big field may drift on the exchange relative to SP.
• Period concentration: sister rules lost their lift in the latest period and over the long history (second favourites went from 1.158 to 1.04 in the vault). A discovery window that ends in October 2025 may overstate it.
• Small winner counts: at 5/1+ around 15% of runners win, so 250 bets is about 40 winners and A/E is noisy by roughly ±0.15.

Control

Discovery result

fail broad rule · 16 Oct, 15:54

Bets

48

5 winners

A/E

0.83

95%: 0.27 to 1.93

SP return

-31.2%

level stakes

Betfair SP return

-7.5%

46 bets with a BSP

Strike rate 10.4%. p = 0.72. Same-race control: the other runners in these races had A/E 1.06, so the selection did not 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 · 51s · 16 Oct, 15:53

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 14 AND course NOT IN ('LEOPARDSTOWN','CURRAGH','THE CURRAGH','PUNCHESTOWN','FAIRYHOUSE','GALWAY','NAAS','GOWRAN PARK','DUNDALK','CORK','LIMERICK','NAVAN','LISTOWEL','TRAMORE','KILLARNEY','BALLINROBE','BELLEWSTOWN','CLONMEL','DOWN ROYAL','DOWNPATRICK','KILBEGGAN','LAYTOWN','ROSCOMMON','SLIGO','THURLES','TIPPERARY','WEXFORD') AND course NOT LIKE '%(IRE)%' AND course NOT LIKE 'DUNDALK%' AND course NOT LIKE '%CURRAGH%'",
    "refinements": [
      {
        "label": "16+ runners (four each-way places)",
        "rule": "is_handicap = 1 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 16 AND course NOT IN ('LEOPARDSTOWN','CURRAGH','THE CURRAGH','PUNCHESTOWN','FAIRYHOUSE','GALWAY','NAAS','GOWRAN PARK','DUNDALK','CORK','LIMERICK','NAVAN','LISTOWEL','TRAMORE','KILLARNEY','BALLINROBE','BELLEWSTOWN','CLONMEL','DOWN ROYAL','DOWNPATRICK','KILBEGGAN','LAYTOWN','ROSCOMMON','SLIGO','THURLES','TIPPERARY','WEXFORD') AND course NOT LIKE '%(IRE)%' AND course NOT LIKE 'DUNDALK%' AND course NOT LIKE '%CURRAGH%'"
      },
      {
        "label": "Exclude Cheltenham and Aintree (festival-handicap confounder)",
        "rule": "is_handicap = 1 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 14 AND course NOT IN ('CHELTENHAM','AINTREE','LEOPARDSTOWN','CURRAGH','THE CURRAGH','PUNCHESTOWN','FAIRYHOUSE','GALWAY','NAAS','GOWRAN PARK','DUNDALK','CORK','LIMERICK','NAVAN','LISTOWEL','TRAMORE','KILLARNEY','BALLINROBE','BELLEWSTOWN','CLONMEL','DOWN ROYAL','DOWNPATRICK','KILBEGGAN','LAYTOWN','ROSCOMMON','SLIGO','THURLES','TIPPERARY','WEXFORD') AND course NOT LIKE '%(IRE)%' AND course NOT LIKE 'DUNDALK%' AND course NOT LIKE '%CURRAGH%'"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) The lab's standard control: the other runners in the same GB races. These should come out near par or below (around 0.95 to 1.00), because the mechanism says the margin sits on the long tail. If they also show 1.10+, the lift is a property of the races (for example SP/BSP overround or a data artefact), not of the 5/1+ favourite. (2) The Irish complement: the same favourite/price/field filter at the excluded Irish courses. The mechanism predicts par there (about 0.95 to 1.03). If Ireland matches GB, the country split is noise and this rule adds nothing over H199.",
      "rule": "is_handicap = 1 AND is_favourite = 1 AND sp_decimal >= 6 AND field_size >= 14 AND (course IN ('LEOPARDSTOWN','CURRAGH','THE CURRAGH','PUNCHESTOWN','FAIRYHOUSE','GALWAY','NAAS','GOWRAN PARK','DUNDALK','CORK','LIMERICK','NAVAN','LISTOWEL','TRAMORE','KILLARNEY','BALLINROBE','BELLEWSTOWN','CLONMEL','DOWN ROYAL','DOWNPATRICK','KILBEGGAN','LAYTOWN','ROSCOMMON','SLIGO','THURLES','TIPPERARY','WEXFORD') OR course LIKE '%(IRE)%' OR course LIKE 'DUNDALK%' OR course LIKE '%CURRAGH%')"
    },
    "pass_criteria": {
      "min_bets": 200,
      "min_ae": 1.12,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "The GB/IRE split was found after the fact in overlapping pools (H199, H235, H-601a8566, H-6da42014 all share runners), so a GB-only discovery lift is partly built in. The full-history vault is the real test.",
      "Course-name exclusion may leak Irish meetings with variant spellings (e.g. 'DUNDALK (AW)', '(IRE)' suffixes). Before reading the result, the lab should check that the Irish complement control holds every IRE race in the H199 pool and the main rule holds none.",
      "GB festival handicaps (Cheltenham, Aintree) may carry the lift. Refinement 2 tests this.",
      "The edge may exist at BSP but not at SP: a 5/1+ favourite in a big field may drift on the exchange relative to SP.",
      "Period concentration: sister rules lost their lift in the latest period and over the long history (second favourites went from 1.158 to 1.04 in the vault). A discovery window that ends in October 2025 may overstate it.",
      "Small winner counts: at 5/1+ around 15% of runners win, so 250 bets is about 40 winners and A/E is noisy by roughly ±0.15."
    ],
    "mechanism": "GB big-field handicaps are priced by a deep off-course book on generous each-way terms (four places at 16+ runners). Each-way and multiple money spreads across the field, so bookmakers put their overround mostly on the long tail and less on the front of the market. A horse that the market makes favourite even at 5/1+ ends up slightly too long. Irish big handicaps form their prices in a smaller ring with more informed stable money, and they come out at par. Pooling the two dilutes the GB edge.",
    "market_miss": "Odds of 5/1+ usually go with a fourth or fifth choice in a smaller field, so bookmakers and punters treat a 5/1+ favourite like any other 5/1+ horse. That ignores the fact that, in a 14+ runner GB handicap, it is the market's single best estimate. Pooled GB+IRE figures (H199 at 1.14) hide the stronger GB pocket, so neither the market nor published systems pick it out.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.15 to 1.35 on 250+ GB-only discovery bets; 1.06+ on the full-history vault",
      "falsified_if": "Full-history vault A/E of 1.03 or below, or a GB-only discovery A/E no higher than the pooled H199 figure (1.11 to 1.14), or the Irish complement control scoring as high as the GB rule"
    },
    "notes": "I kept the investigator's course list and added three LIKE guards ('%(IRE)%', 'DUNDALK%', '%CURRAGH%') against spelling variants, because the catalogue has no country column. The course column is upper case. DOWN ROYAL and DOWNPATRICK are Northern Irish but race under the Irish authority, so I left them excluded. Pass min_ae is 1.12, above the pooled H199 figure, because the hypothesis says GB-only must beat the pooled result. The refinements come from the mechanism (four each-way places at 16+ runners) and from the festival confounder. Read the forward record by period as well as in total."
  }

Timeline

1. Job 1934 finished

   16 Oct, 15:54 · engine:compute-1

2. H-57a9737c Discovery → KILLED fails: 48 bets, needs 200; A/E 0.827, needs 1.12; p 0.7213, needs 0.05 or lower; periods [], each needs 1.0; A/E 0.827 against same-race control 1.062, needs 0.03 above (not a niche pass: A/E 0.827 below 1.12; p 0.7213170291834408 above 0.5; does not beat the same-race control 1.062)

   16 Oct, 15:54 · engine:evaluator

3. discovery result for H-57a9737c (broad rule): fail

   16 Oct, 15:54 · engine:evaluator

4. H-57a9737c Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 15:54 · engine:orchestrator

5. compute-1 took job 1934 (evaluate, attempt 1)

   16 Oct, 15:54 · compute-1

6. Job 1932 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-57a9737c Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 15:54 · agent:designer@v1

9. ai-1 took job 1932 (design, attempt 1)

   16 Oct, 15:53 · ai-1

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

    16 Oct, 15:53 · engine:scheduler

11. New idea H-57a9737c: Replicate: H199 favourites at 5/1+ in 14+ runner handicaps at GB courses only (replicate)

    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.