Sharpen: top-two in the betting at 5/1+ in handicaps of 18+ runners

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) passed discovery as a niche: A/E 1.16 on 721 bets (p 0.087), same-race control 0.99. O-59b08a04
   • 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 lift specific to the favourite, or do all short-end runners at 5/1+ in big-field handicaps beat their prices (an overround or price-band effect)? answered
3. MechanismThe mismatch between a horse's rank and its price grows with field size. In 18+ runner handicaps a 6/1 to 8/1 top-two horse is the market's clear choice among many, while the same price elsewhere marks a mid-ranked runner. Chosen in advance from the mechanism: the effect should rise with field size.
4. Why the market misses itPrice alone ignores how many rivals the market had to spread its money over. Huge-field handicaps also look chaotic, so punters undervalue even the leaders of the market.
5. Predictedpositive, A/E 1.12 to 1.25 on 400+ bets
6. Falsified ifA/E no higher than the 14+ runner pool (about 1.13), or 1.00 or below on 300+ bets
7. Rule sketchis_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 18

The question's answer so far: Not specific to the favourite. Second favourites matched it in discovery (1.158 on 721), but the long-history vault cut that to 1.04 on 4,207 (p 0.199), and third favourites sit at par. Whatever lift exists covers the top two and is small.

Mechanism and rationale

The mismatch between a horse's rank and its price grows with field size. In 18+ runner handicaps a 6/1 to 8/1 top-two horse is the market's clear choice among many, while the same price elsewhere marks a mid-ranked runner. Chosen in advance from the mechanism: the effect should rise with field size.

The rule

is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 18

Reads: is_handicap, sp_rank, sp_decimal, field_size

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

c8fcbd4e1197b5b4a579f6ec740d3ec5b717f3de96c73a3790238d41656c8a5e

Frozen

not frozen

The definition can still change.

Confounders

• Big festival and Saturday handicaps make up much of the 18+ fields (Cheltenham, Aintree, Galway, Royal Ascot, big Saturday heritage handicaps); those meetings may carry the lift, not field size as such
• GB is over-represented in the biggest fields, and the earlier second-favourite lift was GB-only (GB 1.40 vs IRE 0.99), so a pooled 18+ result may just reflect the GB mix
• Overround rival: book_pct rises with field size, so short-end runners may beat SP-implied odds simply because the longshot side carries more of the margin; the same-race control and book_pct need checking
• Draw bias in big Flat fields: top-two-in-the-betting horses may be helped by draw, which is real but a different mechanism
• Overlap with H199 and H-601a8566: this sample is a subset of the 14+ pools already measured, so it is not independent evidence; only the gap against the 14 to 17 band is new information
• Price basis: H199's edge existed at BSP but not SP, so an A/E against SP-implied odds may not mean profit at SP

Control

{
  "description": "Same-race other runners (lab default): expect A/E at or slightly below 1.00 (about 0.95 to 1.00), since the overround is paid by the rest of the field; a lift spread across all runners in these races would point to a race-level artefact, not rank-price mismatch. Second comparison: the identical top-two, 5/1+ handicap selection in 14 to 17 runner fields. The mechanism says A/E should rise with field size, so the 18+ rule must beat this band; if the 14 to 17 band scores the same or higher, the field-size mechanism fails even if the broad rule passes.",
  "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND field_size <= 17"
}

Discovery result

pass broad rule (niche) · 16 Oct, 15:49

Bets

293

39 winners

A/E

1.22

95%: 0.87 to 1.67

SP return

-8.4%

level stakes

Betfair SP return

+18.6%

287 bets with a BSP

Strike rate 13.3%. p = 0.12. Same-race control: the other runners in these races had A/E 0.98, 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. 6 of 6 survived.

• control: Control: same selection in 14-17 runner handicapspass

  561 bets · A/E 1.13 (0.89 to 1.42) · control 0.99 · 16 Oct, 15:51

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272461.35

  after 2026-04-273150.97

• perturbation: Dose-response: field size threshold raised to 20+pass

  174 bets · A/E 1.36 (0.88 to 2.00) · control 0.96 · 16 Oct, 15:51

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.

Other experiments

• neighbour: <= 2 -> 1 (-20%)pass

  61 bets · A/E 1.12 (0.48 to 2.20) · control 1.01 · 16 Oct, 15:49

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-2728–

  after 2026-04-2733–

• neighbour: <= 2 -> 3 (+20%)pass

  605 bets · A/E 1.13 (0.88 to 1.44) · control 0.98 · 16 Oct, 15:49

  Breakdown

  bets

Lineage

Agent runs

• designer v1ok · 42s · 16 Oct, 15:48

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 18",
    "refinements": [
      {
        "label": "GB only",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 18 AND region = 'GB'"
      },
      {
        "label": "dose-response: 22+ runners",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 22"
      },
      {
        "label": "5/1 to 9/1 only",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND sp_decimal <= 10 AND field_size >= 18"
      }
    ],
    "control": {
      "description": "Same-race other runners (lab default): expect A/E at or slightly below 1.00 (about 0.95 to 1.00), since the overround is paid by the rest of the field; a lift spread across all runners in these races would point to a race-level artefact, not rank-price mismatch. Second comparison: the identical top-two, 5/1+ handicap selection in 14 to 17 runner fields. The mechanism says A/E should rise with field size, so the 18+ rule must beat this band; if the 14 to 17 band scores the same or higher, the field-size mechanism fails even if the broad rule passes.",
      "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND field_size <= 17"
    },
    "pass_criteria": {
      "min_bets": 250,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Big festival and Saturday handicaps make up much of the 18+ fields (Cheltenham, Aintree, Galway, Royal Ascot, big Saturday heritage handicaps); those meetings may carry the lift, not field size as such",
      "GB is over-represented in the biggest fields, and the earlier second-favourite lift was GB-only (GB 1.40 vs IRE 0.99), so a pooled 18+ result may just reflect the GB mix",
      "Overround rival: book_pct rises with field size, so short-end runners may beat SP-implied odds simply because the longshot side carries more of the margin; the same-race control and book_pct need checking",
      "Draw bias in big Flat fields: top-two-in-the-betting horses may be helped by draw, which is real but a different mechanism",
      "Overlap with H199 and H-601a8566: this sample is a subset of the 14+ pools already measured, so it is not independent evidence; only the gap against the 14 to 17 band is new information",
      "Price basis: H199's edge existed at BSP but not SP, so an A/E against SP-implied odds may not mean profit at SP"
    ],
    "mechanism": "The rank-price mismatch grows with field size. In handicaps of 18+ runners the probability mass is spread over many rivals, so the market's clear first or second choice can still be 5/1 to 8/1. Bettors anchor on the price, which elsewhere marks a mid-ranked runner, and do not judge it against field size. So top-two horses at these prices are under-bet, and the more runners there are, the bigger the under-betting. Chosen in advance: the effect should be stronger at 18+ than at 14 to 17, and stronger again at 22+.",
    "market_miss": "Price alone ignores how many rivals the market spread its money over. Huge-field handicaps look like lotteries, so recreational money scatters onto outsiders and each-way prices, and even the market leaders end up underpriced. The bookmakers' margin also sits more heavily on the long end in big fields.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.12 to 1.25 on roughly 400 to 700 bets, above the 14 to 17 runner band",
      "falsified_if": "A/E no higher than the 14+ pool (about 1.13), or no higher than the 14 to 17 runner control band, or 1.00 or below on 300+ bets"
    },
    "notes": "The draft rule is kept unchanged: every column (is_handicap, sp_rank, sp_decimal, field_size) is pre_off and available from 2020-07-01 in GB and IRE. Each refinement comes from the reasoning or an earlier finding: GB-only from the finding that the second-favourite lift is GB-only; 22+ runners as the dose-response test the mechanism calls for; and a cap at 9/1 (sp_decimal <= 10) because the earlier sample sat almost entirely at 5/1 to 7/1 and edges fade at longer prices. The pass bar is stricter than the floor (A/E 1.10, 250 bets) because this is a sharpening of pools already measured at about 1.13 to 1.16. Passing at a lower A/E would not show that the effect grows with field size. Any live tracking should be staked and judged at BSP, given the H199 finding."
  }

Timeline

1. Job 1930 finished

   16 Oct, 15:51 · engine:compute-1

2. H-6da42014 Falsification → KILLED falsified: IRE A/E 0.951

   16 Oct, 15:51 · engine:falsification

3. falsification result for H-6da42014 (partition: Favourites only: is the second favourite doing all the work?): pass

   16 Oct, 15:51 · engine:evaluator

4. compute-1 took job 1930 (attack, attempt 1)

   16 Oct, 15:51 · compute-1

5. Job 1929 finished

   16 Oct, 15:51 · engine:compute-1

6. falsification result for H-6da42014 (partition: Jumps only: removes Flat draw bias): pass

   16 Oct, 15:51 · engine:evaluator

7. compute-1 took job 1929 (attack, attempt 1)

   16 Oct, 15:51 · compute-1

8. Job 1928 finished

   16 Oct, 15:51 · engine:compute-1

9. falsification result for H-6da42014 (perturbation: Price floor lowered to 4/1 (sp_decimal >= 5)): pass

   16 Oct, 15:51 · engine:evaluator

10. compute-1 took job 1928 (attack, attempt 1)

    16 Oct, 15:51 · compute-1

11. Job 1927 finished

    16 Oct, 15:51 · engine:compute-1

12. falsification result for H-6da42014 (perturbation: Field size threshold lowered to 16+): pass

    16 Oct, 15:51 · engine:evaluator

13. compute-1 took job 1927 (attack, attempt 1)

    16 Oct, 15:51 · compute-1

14. Job 1926 finished

    16 Oct, 15:51 · engine:compute-1

15. falsification result for H-6da42014 (perturbation: Dose-response: field size threshold raised to 20+): pass

    16 Oct, 15:51 · engine:evaluator

16. compute-1 took job 1926 (attack, attempt 1)

    16 Oct, 15:51 · compute-1

17. Job 1925 finished

    16 Oct, 15:51 · engine:compute-1

18. falsification result for H-6da42014 (control: Control: same selection in 14-17 runner handicaps): pass

    16 Oct, 15:51 · engine:evaluator

19. compute-1 took job 1925 (attack, attempt 1)

    16 Oct, 15:51 · compute-1

20. Job 1920 finished

    16 Oct, 15:51 · engine:ai-1

21. Queued attack on the compute lane: attack: Favourites only: is the second favourite doing all the work?

    16 Oct, 15:51 · engine:scheduler