Generalise: Snowden bumper runners beat their prices

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
   • H17-jamie-snowden-novices (H17: jamie snowden novices) was GREEN in the earlier research: A/E 1.11 on 488 bets (periods 1.11 / 1.07 / 1.20); now PROMISING. O-71eb5f8a
2. AskedDo Snowden's bumper runners (the same young stock before they jump) also beat their prices? testing
3. MechanismIf the lift comes from how the yard develops young stock, the same horses should beat their prices in bumpers before they ever jump.
4. Why the market misses itBumper runners carry almost no public form, so prices lean on pedigree and the yard's reputation. A yard whose horses outrun their profile would be underrated there.
5. Predictedpositive, A/E 1.05 to 1.20 on 100+ bets
6. Falsified ifA/E at or below 1.00 on 100+ bets, which would mean the edge is specific to jumping or to how the yard places novices
7. Rule sketchtrainer LIKE 'Jamie Snowden%' AND race_type = 'Bumper'
8. question idQ-55b058cb

Mechanism and rationale

If the lift comes from how the yard develops young stock, the same horses should beat their prices in bumpers before they ever jump.

The rule

trainer LIKE 'Jamie Snowden%' AND race_type = 'Bumper'

Reads: trainer, race_type

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

e6cb26b255aecc43b9869948da12a906b21c109793b08071c0308a309963bf47

Frozen

not frozen

The definition can still change.

Confounders

• Small sample: one yard's bumper runners over about six seasons may only just reach 150 bets, so one or two big-priced winners can swing the A/E
• The market may price bumper yards on a different basis (pedigree, the sale price of point-to-point purchases), separate from how it prices the same yard's novice jumpers
• Irish point purchases arrive with form the market can see; homebred or store horses arrive without any. The mix between the two can change over time
• A short-price cluster: other families show lifts carried entirely by the up-to-7/1 band. Check whether the result holds outside that band
• The yard's known novice edge only shows at exchange prices (+8.3% at BSP, -6.7% at SP). A bumper lift may also be uncapturable at SP
• Bumper races are a rare type: rows may be mislabelled, for example NH Flat races coded differently by region

Control

{
  "description": "The other runners in the same Snowden bumpers should come out at or slightly under par (about 0.95 to 1.00), as is normal for bumper fields once the overround is removed. If they also show a lift, the effect belongs to those races, not to the yard. A second comparison is all GB bumper runners. If that whole pool shows a lift close to Snowden's, then the market misprices bumpers in general, not this trainer.",
  "rule": "race_type = 'Bumper' AND region = 'GB' AND trainer NOT LIKE 'Jamie Snowden%'"
}

Discovery result

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

Bets

35

11 winners

A/E

1.11

95%: 0.56 to 1.99

SP return

+25.7%

level stakes

Betfair SP return

+47.9%

34 bets with a BSP

Strike rate 31.4%. p = 0.40. Same-race control: the other runners in these races had A/E 0.94, 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. 1 of 5 survived.

• partition: Debutants only (where the mechanism says the lift should peak)info

  18 bets · A/E 1.07 (0.39 to 2.33) · control 0.95 · 16 Oct, 15:18

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-2713–

  after 2026-04-275–

• price_band: Remove the 8/1-14/1 band (2 winners give it 5.03 on 5)info

  30 bets · A/E 0.95 (0.43 to 1.80) · control 1.00 · 16 Oct, 15:18

  Breakdown

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: drop: trainer LIKE 'Jamie Snowden%'fail

  3,749 bets · A/E 0.99 (0.89 to 1.09) · 16 Oct, 15:16

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272,4230.99

  after 2026-04-271,3260.99

• neighbour: drop: race_type = 'Bumper'pass

  359 bets · A/E 1.08 (0.87 to 1.33) · control 0.97 · 16 Oct, 15:16

  Breakdown

Lineage

Agent runs

• designer v1ok · 28s · 16 Oct, 15:16

  Output (schema-checked)

  {
    "rule": "trainer LIKE 'Jamie Snowden%' AND race_type = 'Bumper'",
    "refinements": [
      {
        "label": "debutants only",
        "rule": "trainer LIKE 'Jamie Snowden%' AND race_type = 'Bumper' AND prior_runs = 0"
      },
      {
        "label": "14/1 or shorter",
        "rule": "trainer LIKE 'Jamie Snowden%' AND race_type = 'Bumper' AND sp_decimal <= 15"
      }
    ],
    "control": {
      "description": "The other runners in the same Snowden bumpers should come out at or slightly under par (about 0.95 to 1.00), as is normal for bumper fields once the overround is removed. If they also show a lift, the effect belongs to those races, not to the yard. A second comparison is all GB bumper runners. If that whole pool shows a lift close to Snowden's, then the market misprices bumpers in general, not this trainer.",
      "rule": "race_type = 'Bumper' AND region = 'GB' AND trainer NOT LIKE 'Jamie Snowden%'"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Small sample: one yard's bumper runners over about six seasons may only just reach 150 bets, so one or two big-priced winners can swing the A/E",
      "The market may price bumper yards on a different basis (pedigree, the sale price of point-to-point purchases), separate from how it prices the same yard's novice jumpers",
      "Irish point purchases arrive with form the market can see; homebred or store horses arrive without any. The mix between the two can change over time",
      "A short-price cluster: other families show lifts carried entirely by the up-to-7/1 band. Check whether the result holds outside that band",
      "The yard's known novice edge only shows at exchange prices (+8.3% at BSP, -6.7% at SP). A bumper lift may also be uncapturable at SP",
      "Bumper races are a rare type: rows may be mislabelled, for example NH Flat races coded differently by region"
    ],
    "mechanism": "Snowden's lift with novices (1.11 on 488, 1.17 in non-handicaps) may come from how the yard develops and readies young stock, not from how it places horses over jumps. If so, the same horses should already beat their prices in bumpers, before jumping or race placement can play any part.",
    "market_miss": "Bumper runners, and debutants above all, have little or no public form. Their prices rest on pedigree, sale price and the yard's general reputation. A mid-profile yard whose young horses run above that profile would be priced below what they actually achieve. Little form also means the market corrects slowly, because every bumper horse is a fresh unknown.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.05 to 1.20 on 150+ bets (the lift should be largest among debutants)",
      "falsified_if": "A/E at or below 1.00 on 100+ bets. That would mean the edge is specific to jumping or to how the yard places novices, not to developing young stock. The rule is also undermined if the GB-bumper control shows a lift as large as Snowden's."
    },
    "notes": "The rule is kept from the sketch. Every bumper is a non-handicap for unexposed horses, so no extra novice or handicap filter is needed. The debutant refinement follows directly from the market-miss argument: no form at all means price leans hardest on reputation. The 14/1 cap tests the 'edges die at big prices' pattern and guards against a result driven by one tail winner. Earlier work showed the Snowden novice edge only at BSP, so report bumper ROI at both SP and BSP. Forward tracking of this rule overlaps the live SYS-jamie-snowden-novices system only at the yard level; the runners themselves do not overlap."
  }

Timeline

1. Job 1900 finished

   16 Oct, 15:18 · engine:compute-1

2. H-610ab1e7 Falsification → KILLED falsified: driven by one price band (up to 14/1): the rest score 0.95; fewer than two periods with enough bets to compare (thin: the forward record will judge); IRE: too few bets to judge

   16 Oct, 15:18 · engine:falsification

3. falsification result for H-610ab1e7 (perturbation: Perturb the population: Snowden young horses in bumpers and novice/maiden hurdles): pass

   16 Oct, 15:18 · engine:evaluator

4. compute-1 took job 1900 (attack, attempt 1)

   16 Oct, 15:18 · compute-1

5. Job 1899 finished

   16 Oct, 15:18 · engine:compute-1

6. falsification result for H-610ab1e7 (control: Control: Snowden bumper runners that have already run): info

   16 Oct, 15:18 · engine:evaluator

7. compute-1 took job 1899 (attack, attempt 1)

   16 Oct, 15:18 · compute-1

8. Job 1898 finished

   16 Oct, 15:18 · engine:compute-1

9. falsification result for H-610ab1e7 (partition: Fancied runners only, 7/1 or shorter): info

   16 Oct, 15:18 · engine:evaluator

10. compute-1 took job 1898 (attack, attempt 1)

    16 Oct, 15:18 · compute-1

11. Job 1897 finished

    16 Oct, 15:18 · engine:compute-1

12. falsification result for H-610ab1e7 (price_band: Remove the 8/1-14/1 band (2 winners give it 5.03 on 5)): info

    16 Oct, 15:18 · engine:evaluator

13. compute-1 took job 1897 (attack, attempt 1)

    16 Oct, 15:18 · compute-1

14. Job 1896 finished

    16 Oct, 15:18 · engine:compute-1

15. falsification result for H-610ab1e7 (partition: Debutants only (where the mechanism says the lift should peak)): info

    16 Oct, 15:18 · engine:evaluator

16. compute-1 took job 1896 (attack, attempt 1)

    16 Oct, 15:18 · compute-1

17. Job 1891 finished

    16 Oct, 15:18 · engine:ai-1

18. Queued attack on the compute lane: attack: Perturb the population: Snowden young horses in bumpers and novice/maiden hurdles

    16 Oct, 15:18 · engine:scheduler