Sharpen: Snowden novices with at most one prior run in the code

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. AskedIs the lift strongest on a novice's first one or two runs over obstacles, where the form book holds the least about the horse? testing
3. MechanismSnowden's horses improve with experience and schooling at home that the public cannot see. The gap between the market's view and the horse's real level is widest on the first or second start over hurdles or fences, before the form book catches up.
4. Why the market misses itEarly starts over obstacles are priced from bumper or point form, or from one green run. That form understates a horse being brought along patiently.
5. Predictedpositive, A/E 1.15 to 1.30 on 120+ bets
6. Falsified ifA/E below the parent's 1.17 or at par on 120+ bets, which would mean the lift does not depend on how little form exists
7. Rule sketchtrainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND hurdle_runs <= 1) OR (race_type = 'Chase' AND chase_runs <= 1))
8. question idQ-269e9fad
9. operator

Mechanism and rationale

Snowden's horses improve with experience and schooling at home that the public cannot see. The gap between the market's view and the horse's real level is widest on the first or second start over hurdles or fences, before the form book catches up.

The rule

trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) <= 1) OR (race_type = 'Chase' AND COALESCE(chase_runs, 0) <= 1))

Reads: trainer, is_novice, is_handicap, race_type, hurdle_runs, chase_runs

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

676670bec958097f69311146fe2db773bae37f3ba1e9b2166a3b218e813adce3

Frozen

not frozen

The definition can still change.

Confounders

• First-time-out fitness: code debutants are often coming off a long layoff, and that has its own pricing pattern, separate from the trainer
• Debutants in small novice fields at short prices; this family's short-price cluster could carry the lift alone (check the up-to-7/1 band against the rest)
• Period concentration in the latest year: the yard's profile and strength have risen, so check the lift by season
• Selection overlap: this is a subset of the parent (1.17 on 282), so part of any lift is inherited and is not new evidence
• The run counts start in July/August 2020, so in early data a horse with older hurdle or chase runs can be miscounted as unexposed (mostly a chase issue)
• Edge exists at BSP only; it does not survive at SP (family finding)

Control

{
  "description": "Two comparisons. (1) The lab's standard check: other runners in the same races should sit at about par or below (around 0.95 to 1.0). That would show the lift belongs to the Snowden horse and is not a property of early-season novice non-handicaps. (2) The complement inside the parent: Snowden non-handicap novices with 2 or more prior runs in the code. If the mechanism holds, this group should score clearly below the selected group, near par, roughly 1.0 to 1.10. If it matches or beats the selected group, the lift does not depend on how little form exists.",
  "rule": "trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) >= 2) OR (race_type = 'Chase' AND COALESCE(chase_runs, 0) >= 2))"
}

Discovery result

fail broad rule · 16 Oct, 15:16

Bets

16

10 winners

A/E

1.88

95%: 0.90 to 3.46

SP return

+85.0%

level stakes

Betfair SP return

–

15 bets with a BSP

Strike rate 62.5%. p = 0.044. Same-race control: the other runners in these races had A/E 0.54, 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 · 34s · 16 Oct, 15:15

  Output (schema-checked)

  {
    "rule": "trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) <= 1) OR (race_type = 'Chase' AND COALESCE(chase_runs, 0) <= 1))",
    "refinements": [
      {
        "label": "code debut only",
        "rule": "trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) = 0) OR (race_type = 'Chase' AND COALESCE(chase_runs, 0) = 0))"
      },
      {
        "label": "hurdles only (bumper/point form is the only guide)",
        "rule": "trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) <= 1"
      },
      {
        "label": "14/1 or shorter",
        "rule": "trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) <= 1) OR (race_type = 'Chase' AND COALESCE(chase_runs, 0) <= 1)) AND sp_decimal <= 15.0"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) The lab's standard check: other runners in the same races should sit at about par or below (around 0.95 to 1.0). That would show the lift belongs to the Snowden horse and is not a property of early-season novice non-handicaps. (2) The complement inside the parent: Snowden non-handicap novices with 2 or more prior runs in the code. If the mechanism holds, this group should score clearly below the selected group, near par, roughly 1.0 to 1.10. If it matches or beats the selected group, the lift does not depend on how little form exists.",
      "rule": "trainer LIKE 'Jamie Snowden%' AND is_novice = 1 AND is_handicap = 0 AND ((race_type = 'Hurdle' AND COALESCE(hurdle_runs, 0) >= 2) OR (race_type = 'Chase' AND COALESCE(chase_runs, 0) >= 2))"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "First-time-out fitness: code debutants are often coming off a long layoff, and that has its own pricing pattern, separate from the trainer",
      "Debutants in small novice fields at short prices; this family's short-price cluster could carry the lift alone (check the up-to-7/1 band against the rest)",
      "Period concentration in the latest year: the yard's profile and strength have risen, so check the lift by season",
      "Selection overlap: this is a subset of the parent (1.17 on 282), so part of any lift is inherited and is not new evidence",
      "The run counts start in July/August 2020, so in early data a horse with older hurdle or chase runs can be miscounted as unexposed (mostly a chase issue)",
      "Edge exists at BSP only; it does not survive at SP (family finding)"
    ],
    "mechanism": "Snowden brings horses along patiently through schooling at home that the public cannot see. On their first or second start over hurdles or fences, they are better than any visible form shows. The gap closes as obstacle form builds up, so the lift should be largest when there is 0 or 1 prior run in the code.",
    "market_miss": "Early starts over obstacles are priced from bumper, point or flat form, or from one green run over the obstacles. That form understates a horse being brought along patiently by a yard whose debut improvement is not captured in public ratings. Non-handicaps have no official mark to anchor the price, so the market leans on thin visible form.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.15 to 1.30 at BSP on roughly 120 to 180 bets",
      "falsified_if": "A/E at or below the parent's 1.17, or at par, on 120+ bets. Also falsified if the 2+-runs complement scores the same as or higher than the selected group, or if the lift sits entirely in the up-to-7/1 band."
    },
    "notes": "The rule keeps the sketch but wraps the code-run counts in COALESCE so that true debutants with NULL counts are not dropped silently. The sample will probably fall below 150 bets, so a thin result pointing the right way should go to falsification and forward tracking rather than be killed. The discovery bar is set to A/E 1.10, above the 1.05 floor, because this is a sharpened subset of an already-positive parent."
  }

Timeline

1. Job 1889 finished

   16 Oct, 15:16 · engine:compute-1

2. H-963ba672 Discovery → KILLED fails: 16 bets, needs 150; periods [], each needs 1.0 (not a niche pass: only 16 bets)

   16 Oct, 15:16 · engine:evaluator

3. discovery result for H-963ba672 (broad rule): fail

   16 Oct, 15:16 · engine:evaluator

4. H-963ba672 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 15:16 · engine:orchestrator

5. compute-1 took job 1889 (evaluate, attempt 1)

   16 Oct, 15:16 · compute-1

6. Job 1886 finished

   16 Oct, 15:16 · engine:ai-1

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

   16 Oct, 15:16 · engine:scheduler

8. H-963ba672 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 15:16 · agent:designer@v1

9. ai-1 took job 1886 (design, attempt 1)

   16 Oct, 15:15 · ai-1

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

    16 Oct, 15:15 · engine:scheduler

11. New idea H-963ba672: Sharpen: Snowden novices with at most one prior run in the code (sharpen)

    16 Oct, 15:15 · 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.