Rival: the jumps winner rebound is mark relief, strongest when the handicapper has dropped the horse

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-8cff5453 (Replicate: recent winners beaten 20+ lengths in non-handicap jumps rac) was predicted positive but came out the other way: A/E 0.88 on 211 bets (p 0.268 for the opposite direction). O-6896f0b0
   • H-40a72e24 (Sharpen: recent jumps winners whose 20+ length beating came at a short) passed discovery as a niche: A/E 1.18 on 226 bets (p 0.199), same-race control 0.99. O-def9716f
2. AskedIs the jumps recent-winner rebound handicap-only because the handicapper drops the mark after a heavy beating, so the lift comes from mark relief rather than from the market overreacting to form figures? answered
3. MechanismA recent winner beaten 20+ lengths is often dropped a few pounds by the handicapper even though its ability has not changed. It runs back off a mark close to its winning one, which helps it in a handicap. A non-handicap has no mark to drop, and the replicate failed there (0.88).
4. Why the market misses itPunters read the big beating in the form figures and the small mark drop as minor, so they underrate a winner running off a mark close to the one it won off.
5. Predictedpositive, A/E 1.30 to 1.50 on 300+ bets, above the rating_change >= 0 complement by 0.10+
6. Falsified ifA/E at or below 1.15, or no higher than the same rule with rating_change >= 0. Either would mean the mark drop does not carry the lift and market overreaction to the figures stays the explanation.
7. Rule sketchis_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change < 0

The question's answer so far: No: mark relief does not carry it. Dropped winners scored 1.249 on 380, against 1.412 on 269 for those whose mark was held or raised. The handicap-only pattern needs another explanation, such as the handicapper's no-drop working as an excuse signal.

Mechanism and rationale

A recent winner beaten 20+ lengths is often dropped a few pounds by the handicapper even though its ability has not changed. It runs back off a mark close to its winning one, which helps it in a handicap. A non-handicap has no mark to drop, and the replicate failed there (0.88).

The rule

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

Reads: is_handicap, race_type, last_beaten_lengths, best_last3_beaten, rating_change

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

b75cf09f7859bd304ef190f9f67b5418dcd7364017fb3eef7b838909f79faebe

Frozen

not frozen

The definition can still change.

Confounders

• Field size: big-field beatings are the lift niche in the parent family
• Class drop coinciding with the mark drop (easier company, not mark relief); refinement 3 checks this, but IRE race_class only starts in 2026-07, so that check leans on GB
• Ground or trip change since the beating
• Longshot band: earlier rebound lifts rest on mid prices and a few 16/1+ winners
• IRE vs GB handicapping practice: IRE marks may be cut faster, and earlier cuts in this family lean on IRE
• Days since last run: a quick return may come before the mark drop is applied, so it falls in the complement
• Short window: rating_change and best_last3_beaten exist only from 2025-11-01, so the sample is small and seasonal (mostly one jumps season)
• The handicap-only selection already excludes non-handicaps, so beating the complement is the only evidence that the mark drop itself matters

Control

{
  "description": "The main comparison is the complement: the same recent winners beaten 20+ lengths whose mark was not dropped (rating_change >= 0). If mark relief carries the rebound, the dropped group should beat this complement by 0.10 A/E or more. Other runners in the same races should score close to 1.0, which shows the lift is not a property of these races.",
  "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0"
}

Discovery result

pass broad rule (niche) · 17 Oct, 04:41

Bets

380

46 winners

A/E

1.25

95%: 0.91 to 1.67

SP return

+7.6%

level stakes

Betfair SP return

+30.6%

379 bets with a BSP

Strike rate 12.1%. p = 0.080. 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.

• perturbation: Perturb beating threshold down to 16 lengthspass

  495 bets · A/E 1.18 (0.90 to 1.51) · control 0.99 · 17 Oct, 04:47

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272831.24

  after 2026-04-272121.09

• partition: Partition: smaller fields only (13 or fewer runners)pass

  289 bets · A/E 1.25 (0.89 to 1.70) · control 0.98 · 17 Oct, 04:47

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: >= 20 -> 16 (-20%)pass

  495 bets · A/E 1.18 (0.90 to 1.51) · control 0.99 · 17 Oct, 04:48

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272831.24

  after 2026-04-272121.09

• neighbour: >= 20 -> 24 (+20%)pass

  306 bets · A/E 1.22 (0.84 to 1.70) · control 0.99 · 17 Oct, 04:48

  Breakdown

Lineage

Agent runs

• designer v1ok · 44s · 17 Oct, 04:37

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change < 0",
    "refinements": [
      {
        "label": "Dose-response: dropped 4 lb or more",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change <= -4"
      },
      {
        "label": "Price cap 14/1 or shorter (earlier work puts the rebound lift at mid prices, not longshots)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change < 0 AND sp_decimal <= 15"
      },
      {
        "label": "Mark drop without a class drop (separates mark relief from easier company)",
        "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change < 0 AND race_class IS NOT NULL AND last_race_class IS NOT NULL AND race_class <= last_race_class"
      }
    ],
    "control": {
      "description": "The main comparison is the complement: the same recent winners beaten 20+ lengths whose mark was not dropped (rating_change >= 0). If mark relief carries the rebound, the dropped group should beat this complement by 0.10 A/E or more. Other runners in the same races should score close to 1.0, which shows the lift is not a property of these races.",
      "rule": "is_handicap = 1 AND race_type IN ('Hurdle', 'Chase') AND last_beaten_lengths >= 20 AND best_last3_beaten = 0 AND rating_change IS NOT NULL AND rating_change >= 0"
    },
    "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": [
      "Field size: big-field beatings are the lift niche in the parent family",
      "Class drop coinciding with the mark drop (easier company, not mark relief); refinement 3 checks this, but IRE race_class only starts in 2026-07, so that check leans on GB",
      "Ground or trip change since the beating",
      "Longshot band: earlier rebound lifts rest on mid prices and a few 16/1+ winners",
      "IRE vs GB handicapping practice: IRE marks may be cut faster, and earlier cuts in this family lean on IRE",
      "Days since last run: a quick return may come before the mark drop is applied, so it falls in the complement",
      "Short window: rating_change and best_last3_beaten exist only from 2025-11-01, so the sample is small and seasonal (mostly one jumps season)",
      "The handicap-only selection already excludes non-handicaps, so beating the complement is the only evidence that the mark drop itself matters"
    ],
    "mechanism": "A horse that won within its last three runs and was then beaten 20+ lengths is often dropped a few pounds by the handicapper, though one bad run rarely means its ability has fallen. It comes back in a handicap off a mark at or below the one it won off, so it is better treated by the weights than its recent form suggests. A non-handicap has no mark to drop, which would explain why the rebound failed there (0.88 on 211).",
    "market_miss": "Punters weight the big beating in the form figures and treat a small mark drop as minor. So they underrate a recent winner now running off a mark close to or below its winning one. The relief does not show in the visible form line.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.30 to 1.50 (300+ bets hoped for, but the window is short), at least 0.10 above the rating_change >= 0 complement",
      "falsified_if": "Falsified if the broad rule's A/E is at or below 1.15, or no more than 0.05 above the rating_change >= 0 complement. Either would mean the mark drop does not carry the lift, and market overreaction to the form figures stays the explanation."
    },
    "notes": "I added rating_change IS NOT NULL so the rule matches its complement cleanly. rating_change, best_last3_beaten and last_official_rating all start on 2025-11-01, so the discovery sample covers only one jumps season. That is why min_bets stays at the floor while min_ae is raised to 1.10. If the complement scores as high as the dropped group, this rival explanation is refuted even if the broad rule passes."
  }

Timeline

1. Job 3012 finished

   17 Oct, 04:49 · engine:compute-1

2. hypothesis note H-4392b730

   17 Oct, 04:49 · engine:perturb

3. perturbation result for H-4392b730 (neighbour: drop: race_type IN ('Hurdle', 'Chase')): pass

   17 Oct, 04:49 · engine:perturb

4. perturbation result for H-4392b730 (neighbour: drop: is_handicap = 1): pass

   17 Oct, 04:49 · engine:perturb

5. perturbation result for H-4392b730 (neighbour: < 0 -> 1 (+20%)): pass

   17 Oct, 04:49 · engine:perturb

6. perturbation result for H-4392b730 (neighbour: < 0 -> -1 (-20%)): pass

   17 Oct, 04:49 · engine:perturb

7. perturbation result for H-4392b730 (neighbour: >= 20 -> 24 (+20%)): pass

   17 Oct, 04:48 · engine:perturb

8. perturbation result for H-4392b730 (neighbour: >= 20 -> 16 (-20%)): pass

   17 Oct, 04:48 · engine:perturb

9. compute-1 took job 3012 (mutate, attempt 1)

   17 Oct, 04:48 · compute-1

10. Job 3019 finished

    17 Oct, 04:47 · engine:compute-1

11. H-4392b730 Falsification → KILLED falsified: IRE A/E 0.96

    17 Oct, 04:47 · engine:falsification

12. falsification result for H-4392b730 (price_band: Cap the price at under 16/1 to strip the longshot band): pass

    17 Oct, 04:47 · engine:evaluator

13. compute-1 took job 3019 (attack, attempt 1)

    17 Oct, 04:47 · compute-1

14. Job 3018 finished

    17 Oct, 04:47 · engine:compute-1

15. falsification result for H-4392b730 (control: Control: mark dropped after a heavy beating, but no recent win): pass

    17 Oct, 04:47 · engine:evaluator

16. compute-1 took job 3018 (attack, attempt 1)

    17 Oct, 04:47 · compute-1