Rival: the recent-winner rebound is a longshot artefact and vanishes at 14/1 or shorter

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-363890c3 (Sharpen: recent winners beaten 20+ lengths next time over jumps) passed discovery: A/E 1.34 on 669 bets (p 0.006), same-race control 0.97. O-c88bca5b
2. AskedDoes the recent-winner rebound hold below 16/1, or is the 1.34 carried by about 12 winners at 16/1 or bigger? answered
3. MechanismIf the rebound is real, a recent winner beaten 20+ lengths only once still has its ability. That should show at every price, including the shorter prices where most of the money goes. The rival is that 12 big-priced winners at 16/1 or bigger make the 1.34.
4. Why the market misses itThe 20+ length beating is the most recent and most visible line of form. Punters and pricing models weight it more heavily than the earlier win, so even well-fancied winners drift past their true chance.
5. Predictedpositive, A/E 1.12 to 1.25 on 400+ bets at 14/1 or shorter
6. Falsified ifA/E at or below 1.03 on 350+ bets at 14/1 or shorter, which would mean the winner lift is a longshot artefact
7. Rule sketchis_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal <= 15

Mechanism and rationale

If the rebound is real, a recent winner beaten 20+ lengths only once still has its ability. That should show at every price, including the shorter prices where most of the money goes. The rival is that 12 big-priced winners at 16/1 or bigger make the 1.34.

The rule

is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal <= 15

Reads: is_handicap, race_type, last_beaten_lengths, best_last3_beaten, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

482998582952183561d5be280a15b88b8b141543f855dc7e4776bd52cedc03e0

Frozen

17 Oct, 04:23

Nothing in the definition can change after this.

Confounders

• Few winners in the 16/1-or-bigger band: the 1.34 headline may rest on about 12 longshot wins, which is why the rule caps the price at 15.0 decimal
• Field size: bigger fields both lengthen prices and make 20+ length beatings easier, so the lift may follow field size and not the form story
• Recent winners carry fresh mark rises, which may cancel the rebound at short prices where the market already prices the win
• IRE share: the IRE part of the winner group was much stronger (1.50 against 1.03 for GB without winners), so check the region split
• best_last3_beaten is only populated from 2025-11-01, so the sample covers about one jumps season and may reflect that season's conditions
• Price band and odds-on/short-price favourite-longshot bias: A/E at short prices sits closer to 1.0 across the board, so a smaller lift there is expected even if the effect is real

Control

{
  "description": "The other runners in the same races should come out near A/E 1.0 (about 0.95 to 1.02), which shows the lift belongs to the horse and not to the race. The rival explanation also needs a direct comparison: the same profile at 16/1 or bigger. That band was 2.01 on 222 from only 12 wins. If the rebound is real, the band at 14/1 or shorter stays clearly above 1.0 while the longshot band is noisy. If it is an artefact, the band at 14/1 or shorter sits near 1.0 and all of the lift is at the long prices.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal > 15"
}

Discovery result

pass broad rule · 17 Oct, 04:22

Bets

447

72 winners

A/E

1.26

95%: 0.99 to 1.59

SP return

+7.5%

level stakes

Betfair SP return

+27.7%

445 bets with a BSP

Strike rate 16.1%. p = 0.030. Same-race control: the other runners in these races had A/E 0.97, 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.

• perturbation: Perturb: heavier beating 24+ lengths, cap keptpass

  340 bets · A/E 1.28 (0.96 to 1.67) · control 0.97 · 17 Oct, 04:23

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-271781.39

  after 2026-04-271621.15

• perturbation: Perturb: 16+ length beating with tighter 11/1 price cappass

  515 bets · A/E 1.22 (0.98 to 1.50) · control 0.97 · 17 Oct, 04:23

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': 1.03, 'max_p': 0.2}info

  0 bets · · 17 Oct, 04:24

  Breakdown

  betsA/E0.71.01.3

  forward0–

Other experiments

• neighbour: >= 20 -> 16 (-20%)pass

  613 bets · A/E 1.19 (0.96 to 1.45) · control 0.98 · 17 Oct, 04:24

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-273351.26

  after 2026-04-272781.10

• neighbour: >= 20 -> 24 (+20%)pass

  340 bets · A/E 1.28 (0.96 to 1.67) · control 0.97 · 17 Oct, 04:24

  Breakdown

Lineage

Agent runs

• designer v1ok · 48s · 17 Oct, 04:16

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal <= 15",
    "refinements": [
      {
        "label": "8/1 or shorter",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal <= 9"
      },
      {
        "label": "fields of 13 or fewer (field-size confounder)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal <= 15 AND field_size <= 13"
      }
    ],
    "control": {
      "description": "The other runners in the same races should come out near A/E 1.0 (about 0.95 to 1.02), which shows the lift belongs to the horse and not to the race. The rival explanation also needs a direct comparison: the same profile at 16/1 or bigger. That band was 2.01 on 222 from only 12 wins. If the rebound is real, the band at 14/1 or shorter stays clearly above 1.0 while the longshot band is noisy. If it is an artefact, the band at 14/1 or shorter sits near 1.0 and all of the lift is at the long prices.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND sp_decimal > 15"
    },
    "pass_criteria": {
      "min_bets": 200,
      "min_ae": 1.08,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Few winners in the 16/1-or-bigger band: the 1.34 headline may rest on about 12 longshot wins, which is why the rule caps the price at 15.0 decimal",
      "Field size: bigger fields both lengthen prices and make 20+ length beatings easier, so the lift may follow field size and not the form story",
      "Recent winners carry fresh mark rises, which may cancel the rebound at short prices where the market already prices the win",
      "IRE share: the IRE part of the winner group was much stronger (1.50 against 1.03 for GB without winners), so check the region split",
      "best_last3_beaten is only populated from 2025-11-01, so the sample covers about one jumps season and may reflect that season's conditions",
      "Price band and odds-on/short-price favourite-longshot bias: A/E at short prices sits closer to 1.0 across the board, so a smaller lift there is expected even if the effect is real"
    ],
    "mechanism": "A jumps handicapper that won within its last three runs and was then beaten 20+ lengths last time has probably not lost its ability. The bad run is more likely a one-off (trouble, wrong conditions, a bad day). If so, the rebound should show at every price, including 14/1 or shorter where most of the money goes. The rival says the 1.34 headline comes from about 12 longshot winners at 16/1 or bigger, and the short-price band carries no edge.",
    "market_miss": "The 20+ length beating is the most recent and most visible line of form. Punters and pricing models weight it more heavily than the earlier win, so even a well-fancied recent winner drifts a little past its true chance. At short prices the error should be smaller than at long prices but still present.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.12 to 1.25 on 400+ bets at 14/1 or shorter (sp_decimal <= 15)",
      "falsified_if": "A/E at or below 1.03 on 350+ bets at 14/1 or shorter, which would mean the recent-winner lift is a longshot artefact carried by the 16/1+ band"
    },
    "notes": "The draft rule is kept as written. It uses only pre_off columns, and sp_decimal <= 15 is exactly 14/1 or shorter. best_last3_beaten = 0 picks out horses with a win in their last three runs. Because the last run was beaten 20+ lengths, that win has to be one of the two runs before it. This is a rival-explanation test, so the useful result is the short-price band against the 16/1+ control: a clear pass here would weaken the longshot-artefact rival. Pass criteria are set a little above the floor (200 bets, A/E 1.08), because about 450 bets are expected and A/E at short prices tends to be compressed. The refinements check the shortest prices (8/1 or shorter) and the field-size confounder. They are read only if the broad rule shows evidence."
  }

Timeline

1. Job 2906 finished

   17 Oct, 04:24 · engine:compute-1

2. hypothesis note H-b0d006d8

   17 Oct, 04:24 · engine:perturb

3. perturbation result for H-b0d006d8 (neighbour: drop: race_type IN ('Hurdle', 'Chase')): pass

   17 Oct, 04:24 · engine:perturb

4. perturbation result for H-b0d006d8 (neighbour: drop: is_handicap = 1): pass

   17 Oct, 04:24 · engine:perturb

5. perturbation result for H-b0d006d8 (neighbour: <= 15 -> 18 (+20%)): pass

   17 Oct, 04:24 · engine:perturb

6. perturbation result for H-b0d006d8 (neighbour: <= 15 -> 12 (-20%)): pass

   17 Oct, 04:24 · engine:perturb

7. perturbation result for H-b0d006d8 (neighbour: >= 20 -> 24 (+20%)): pass

   17 Oct, 04:24 · engine:perturb

8. perturbation result for H-b0d006d8 (neighbour: >= 20 -> 16 (-20%)): pass

   17 Oct, 04:24 · engine:perturb

9. compute-1 took job 2906 (mutate, attempt 1)

   17 Oct, 04:24 · compute-1

10. Queued mutate on the compute lane: parameter stability: do neighbouring thresholds hold?

    17 Oct, 04:24 · engine:scheduler

11. Job 2901 finished

    17 Oct, 04:24 · engine:compute-1

12. forward result for H-b0d006d8 (forward vs bar {'min_bets': 100, 'min_ae': 1.03, 'max_p': 0.2}): info

    17 Oct, 04:24 · engine:evaluator

13. compute-1 took job 2901 (forward_check, attempt 1)

    17 Oct, 04:24 · compute-1

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

    17 Oct, 04:24 · engine:scheduler

15. Job 2883 finished

    17 Oct, 04:23 · engine:compute-1

16. H-b0d006d8 Frozen → SHADOW vault skipped: last_beaten_lengths, best_last3_beaten do not exist before the discovery zone; track it forward instead

    17 Oct, 04:23 · engine:orchestrator