Sharpen: recent jumps winners whose 20+ length beating came at a short last price

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-32b815ff (Test-mechanism: recent winners beaten only 8 to 19 lengths should show) passed discovery as a niche: A/E 1.12 on 828 bets (p 0.117), same-race control 0.99. O-2ecb16ca
   • H-32b815ff (Test-mechanism: recent winners beaten only 8 to 19 lengths should show) survived falsification: 6 of 6 attacks left it standing; it is frozen for its one vault query. O-4cd42568
2. AskedWithin the recent-winner jumps rebound, is the lift largest when the heavy beating came at a short last price, i.e. a fancied winner that flopped? answered
3. MechanismA recent winner sent off fancied and then beaten 20+ lengths most likely had a one-off problem (ground, a mistake, a bleed, being eased once beaten). Its ability is unchanged, and the yard's confidence before that run was real information.
4. Why the market misses itThe form line shows a big margin, and punters drift off a horse that 'flopped when fancied'. The dose-response (1.12 at 8 to 19 lengths, 1.34 at 20+) suggests the size of the margin drives the overreaction, and a fancied flop makes it look worse.
5. Predictedpositive, A/E 1.35 to 1.60 on 250+ bets
6. Falsified ifA/E below 1.15, or no higher than the same rule with last_sp_decimal > 6
7. Rule sketch

Mechanism and rationale

A recent winner sent off fancied and then beaten 20+ lengths most likely had a one-off problem (ground, a mistake, a bleed, being eased once beaten). Its ability is unchanged, and the yard's confidence before that run was real information.

The rule

is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND last_sp_decimal IS NOT NULL AND last_sp_decimal <= 6

Reads: is_handicap, race_type, last_beaten_lengths, best_last3_beaten, last_sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

19270914c17350cc49c2661d59fee62caa35ca9b025de0aa96cc91484b69058d

Frozen

17 Oct, 04:41

Nothing in the definition can change after this.

Confounders

• Heavy overlap with the parent winner-rebound rule (1.34 on 669). This is a subset, not independent evidence, so it must beat the last_sp_decimal > 6 complement and not just the floor.
• Price band today: horses fancied last time may still be short today. A short price narrows the possible A/E, and the parent's lift sat below 16/1, so both rules should be read by price band.
• Known injury or bleed behind the flop (pulled up, eased or lame). Those that return quickly may differ from those given long breaks, so check days_since_last_run.
• Thin window: best_last3_beaten exists only from 2025-11-01, so the sample is small and seasonal (one jumps season).
• A class or rating drop after the flop (handicapper relief) rather than market overreaction.
• Region mix: GB and IRE behave differently in earlier rebound findings.

Control

{
  "description": "Same-race other runners should be near 0.95-1.0, so the lift belongs to the selected horses and not to the races. The comparison that decides the question is the same winner-rebound rule with an unfancied last run (last_sp_decimal > 6). The hypothesis holds only if the broad rule's A/E is clearly above this control's. If the two are equal, last-run price adds nothing to the parent rebound.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND last_sp_decimal IS NOT NULL AND last_sp_decimal > 6"
}

Discovery result

pass broad rule (niche) · 17 Oct, 04:36

Bets

226

32 winners

A/E

1.18

95%: 0.81 to 1.66

SP return

+62.2%

level stakes

Betfair SP return

+168.2%

225 bets with a BSP

Strike rate 14.2%. p = 0.20. Same-race control: the other runners in these races had A/E 0.99, 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. 5 of 5 survived.

• price_band: Cap today's price at 15/1 or shorterpass

  182 bets · A/E 1.00 (0.66 to 1.47) · control 1.01 · 17 Oct, 04:41

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27970.90

  after 2026-04-27851.14

• perturbation: Perturb looser: last SP 7.0 or shorter, beaten 17+ lengthspass

  357 bets · A/E 1.31 (0.98 to 1.70) · control 0.97 · 17 Oct, 04:41

Sealed holdout (vault)

Not queried. Only a frozen idea can spend its one query against the 2020-25 holdout.

Forward record

• forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}info

  0 bets · · 17 Oct, 04:41

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: >= 20 -> 16 (-20%)pass

  303 bets · A/E 1.21 (0.88 to 1.61) · control 0.98 · 17 Oct, 04:41

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271661.08

  after 2026-04-271371.36

• neighbour: >= 20 -> 24 (+20%)pass

  165 bets · A/E 1.20 (0.75 to 1.81) · control 0.99 · 17 Oct, 04:41

  Breakdown

Lineage

Agent runs

• designer v1ok · 30s · 17 Oct, 04:34

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND last_sp_decimal IS NOT NULL AND last_sp_decimal <= 6",
    "refinements": [
      {
        "label": "today 15/1 or shorter",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND last_sp_decimal IS NOT NULL AND last_sp_decimal <= 6 AND sp_decimal < 16"
      },
      {
        "label": "heavily fancied last time (9/4 or shorter)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND last_sp_decimal IS NOT NULL AND last_sp_decimal <= 3.25"
      }
    ],
    "control": {
      "description": "Same-race other runners should be near 0.95-1.0, so the lift belongs to the selected horses and not to the races. The comparison that decides the question is the same winner-rebound rule with an unfancied last run (last_sp_decimal > 6). The hypothesis holds only if the broad rule's A/E is clearly above this control's. If the two are equal, last-run price adds nothing to the parent rebound.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND last_sp_decimal IS NOT NULL AND last_sp_decimal > 6"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.15,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Heavy overlap with the parent winner-rebound rule (1.34 on 669). This is a subset, not independent evidence, so it must beat the last_sp_decimal > 6 complement and not just the floor.",
      "Price band today: horses fancied last time may still be short today. A short price narrows the possible A/E, and the parent's lift sat below 16/1, so both rules should be read by price band.",
      "Known injury or bleed behind the flop (pulled up, eased or lame). Those that return quickly may differ from those given long breaks, so check days_since_last_run.",
      "Thin window: best_last3_beaten exists only from 2025-11-01, so the sample is small and seasonal (one jumps season).",
      "A class or rating drop after the flop (handicapper relief) rather than market overreaction.",
      "Region mix: GB and IRE behave differently in earlier rebound findings."
    ],
    "mechanism": "A jumps handicapper that won within its last three runs, was then sent off fancied (6.0 or shorter) and beaten 20+ lengths most likely had a one-off problem: ground, a jumping mistake, a bleed, or being eased once beaten. Its ability is unchanged. The short last price shows the yard and market rated it highly on information that the single bad run does not cancel.",
    "market_miss": "The form line shows a big losing margin, and the narrative 'flopped when fancied' pushes punters away. The parent dose-response (1.12 at 8 to 19 lengths, 1.34 at 20+) suggests the size of the margin drives the overreaction, and the disappointment of a fancied flop adds to it. Meanwhile the recent win and the prior market confidence are under-weighted.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.35 to 1.60 on roughly 150 to 300 bets",
      "falsified_if": "Falsified if the broad rule's A/E is below 1.15, or no higher than the control rule (same rule with last_sp_decimal > 6)."
    },
    "notes": "I added `last_sp_decimal IS NOT NULL` for explicit NULL handling; otherwise the sketch rule is unchanged. `best_last3_beaten = 0` means a win in the last three runs, and because the last run was a 20+ length beating, that win came two or three runs back. Pass min_ae is raised to 1.15 to match the stated falsifier. The column window from 2025-11 may leave fewer than 150 bets. If so, a thin positive result goes to falsification and forward tracking as the lab's policy allows."
  }

Timeline

1. Job 3011 finished

   17 Oct, 04:41 · engine:compute-1

2. forward result for H-40a72e24 (forward vs bar {'min_bets': 100, 'min_ae': 1.05, 'max_p': 0.2}): info

   17 Oct, 04:41 · engine:evaluator

3. compute-1 took job 3011 (forward_check, attempt 1)

   17 Oct, 04:41 · compute-1

4. Queued forward_check on the compute lane: daily forward check of a shadow system

   17 Oct, 04:41 · engine:scheduler

5. Job 2977 finished

   17 Oct, 04:41 · engine:compute-1

6. hypothesis note H-40a72e24

   17 Oct, 04:41 · engine:perturb

7. perturbation result for H-40a72e24 (neighbour: drop: race_type IN ('Hurdle', 'Chase')): pass

   17 Oct, 04:41 · engine:perturb

8. perturbation result for H-40a72e24 (neighbour: drop: is_handicap = 1): pass

   17 Oct, 04:41 · engine:perturb

9. perturbation result for H-40a72e24 (neighbour: <= 6 -> 7 (+20%)): pass

   17 Oct, 04:41 · engine:perturb

10. perturbation result for H-40a72e24 (neighbour: <= 6 -> 5 (-20%)): pass

    17 Oct, 04:41 · engine:perturb

11. perturbation result for H-40a72e24 (neighbour: >= 20 -> 24 (+20%)): pass

    17 Oct, 04:41 · engine:perturb

12. perturbation result for H-40a72e24 (neighbour: >= 20 -> 16 (-20%)): pass

    17 Oct, 04:41 · engine:perturb

13. compute-1 took job 2977 (mutate, attempt 1)

    17 Oct, 04:41 · compute-1

14. Job 3006 finished

    17 Oct, 04:41 · engine:compute-1

15. H-40a72e24 Frozen → SHADOW vault skipped: last_beaten_lengths, best_last3_beaten do not exist before the discovery zone; track it forward instead

    17 Oct, 04:41 · engine:orchestrator

16. compute-1 took job 3006 (vault, attempt 1)

    17 Oct, 04:41 · compute-1