Rival: higher-weighted runners of all ages beat their price in marathon handicap chases

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-a4349569 (Age boundary: 8 and 9 year olds in marathon handicap chases sit near p) pointed the predicted way but failed the bar: A/E 0.97 on 229 bets (p 0.588), control 1.05. O-089a750b
   • H-ebbea470 (Rival: light weight, not youth, carries the marathon handicap chase li) was predicted positive but came out the other way: A/E 0.48 on 124 bets (p 0.08 for the opposite direction). O-ccf039fe
2. AskedIn 3m4f+ handicap chases, do the higher-weighted (better-class) runners of every age beat their price, so the young-stayer lift is partly a class effect? answered
3. MechanismOver a marathon trip with a tiring finish, class and a proper chasing engine count for more than a few pounds of weight. The better-rated horses higher in the weights keep their jumping and their cruising speed when the low-rated ones are spent.
4. Why the market misses itPunters and pundits repeat 'weight stops trains' and mark down horses with big weights over 3m4f+. The low-weighted horses look 'well in' on paper, so money moves from the top of the weights to the bottom.
5. Predictedpositive, A/E 1.08 to 1.15 on 600+ bets
6. Falsified ifA/E at or below 1.0 on 500+ bets, or older runners in this group scoring no better than the field (the young-stayer lift would then stand alone)

Mechanism and rationale

Over a marathon trip with a tiring finish, class and a proper chasing engine count for more than a few pounds of weight. The better-rated horses higher in the weights keep their jumping and their cruising speed when the low-rated ones are spent.

The rule

race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs > bottom_weight_lbs + 7

Reads: race_type, is_handicap, distance_yards, weight_lbs, bottom_weight_lbs

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

171b2dcbe30aa0ee2bb18e0c1ab4d0786380e7d6376aeb09ec5684aa4fd25e55

Frozen

16 Oct, 14:12

Nothing in the definition can change after this.

Confounders

• Young horses (age <= 7) inside the group carry the existing young-stayer lift (1.16 on 618); the age >= 8 refinement must hold for a class effect to stand on its own
• Price mix: higher-weighted runners are shorter-priced, so favourite-longshot bias could produce the lift by itself; check the 14/1-or-shorter band and the lift by price band
• Out-of-the-handicap runners all sit at bottom weight and are long-priced poor horses, which drags the comparison group down and flatters the heavier group against the field
• Festival Nationals (Grand National, Scottish/Irish/Welsh Nationals) have compressed weights and big fields; a few big races could dominate
• Thin sample, with one season or a few big-priced winners able to drive the result; check by period and with the top winners removed
• Overlap with the observation that motivated this: heavier older runners already scored 1.154 on 352 in discovery, so the older refinement is partly seen data and must be confirmed in the vault

Control

{
  "description": "Same-race control: the other runners in these races are mostly the runners within 7 lb of bottom weight (including out-of-the-handicap horses). If class beats weight, they should come out below par (A/E around 0.85–0.95), and the heavier group should beat them by a clear margin. If the same-race control is also at or above 1.0, the lift is a property of the races (e.g. a pool-wide price effect), not of weight position. The second control rule isolates youth from class: older horses near the bottom of the weights in the same race type. They have already scored 0.48 on 124, so the comparison that matters is the age >= 8 refinement measured against the same-race field.",
  "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs <= bottom_weight_lbs + 7"
}

Discovery result

pass broad rule (niche) · 16 Oct, 14:05

Bets

338

35 winners

A/E

1.21

95%: 0.84 to 1.68

SP return

-21.0%

level stakes

Betfair SP return

+1.0%

338 bets with a BSP

Strike rate 10.4%. p = 0.16. Same-race control: the other runners in these races had A/E 0.60, 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: Weight cut-off raised from 7 lb to 9 lb above bottom weightpass

  294 bets · A/E 1.35 (0.94 to 1.88) · control 0.48 · 16 Oct, 14:12

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272691.37

  after 2026-04-2725–

• partition: Older horses only (age 8+): strip out the young-stayer liftpass

  253 bets · A/E 1.18 (0.75 to 1.77) · control 0.87 · 16 Oct, 14:12

  Breakdown

Sealed holdout (vault)

Bets

1,378

16 Oct, 14:12

A/E

0.98

p = 0.59

BH threshold then

0.0024

over every vault test so far

Outcome

Failed

Forward record

No forward record. Shadow tracking starts after the vault and counts only races run after the freeze.

Other experiments

• neighbour: >= 6160 -> 4928 (-20%)pass

  3,006 bets · A/E 1.02 (0.92 to 1.14) · control 0.98 · 16 Oct, 14:17

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-272,1681.03

  after 2026-04-278381.00

• neighbour: >= 6160 -> 7392 (+20%)pass

  33 bets · A/E 1.33 (0.16 to 4.82) · control 0.00 · 16 Oct, 14:17

  Breakdown

Lineage

Agent runs

• designer v1ok · 39s · 16 Oct, 14:00

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs > bottom_weight_lbs + 7",
    "refinements": [
      {
        "label": "older horses only (strip the young-stayer lift)",
        "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs > bottom_weight_lbs + 7 AND age >= 8"
      },
      {
        "label": "14/1 or shorter (price-mix / favourite-longshot check)",
        "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs > bottom_weight_lbs + 7 AND sp_decimal <= 15"
      },
      {
        "label": "top of the weights: 14 lb+ above bottom weight (dose-response)",
        "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs > bottom_weight_lbs + 14"
      }
    ],
    "control": {
      "description": "Same-race control: the other runners in these races are mostly the runners within 7 lb of bottom weight (including out-of-the-handicap horses). If class beats weight, they should come out below par (A/E around 0.85–0.95), and the heavier group should beat them by a clear margin. If the same-race control is also at or above 1.0, the lift is a property of the races (e.g. a pool-wide price effect), not of weight position. The second control rule isolates youth from class: older horses near the bottom of the weights in the same race type. They have already scored 0.48 on 124, so the comparison that matters is the age >= 8 refinement measured against the same-race field.",
      "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs <= bottom_weight_lbs + 7"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.07,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Young horses (age <= 7) inside the group carry the existing young-stayer lift (1.16 on 618); the age >= 8 refinement must hold for a class effect to stand on its own",
      "Price mix: higher-weighted runners are shorter-priced, so favourite-longshot bias could produce the lift by itself; check the 14/1-or-shorter band and the lift by price band",
      "Out-of-the-handicap runners all sit at bottom weight and are long-priced poor horses, which drags the comparison group down and flatters the heavier group against the field",
      "Festival Nationals (Grand National, Scottish/Irish/Welsh Nationals) have compressed weights and big fields; a few big races could dominate",
      "Thin sample, with one season or a few big-priced winners able to drive the result; check by period and with the top winners removed",
      "Overlap with the observation that motivated this: heavier older runners already scored 1.154 on 352 in discovery, so the older refinement is partly seen data and must be confirmed in the vault"
    ],
    "mechanism": "In 3m4f+ handicap chases with a tiring finish, class and a proper chasing engine (jumping, cruising speed) count for more than the few pounds of weight they cost. Higher-rated horses carrying more than 7 lb above bottom weight keep going when the low-rated horses are spent, so the handicap's weight compensation over-corrects at marathon trips. If so, part of the young-stayer lift is a class effect shared by better horses of every age.",
    "market_miss": "'Weight stops trains' is repeated most loudly over marathon trips, so punters mark down the horses high in the weights and back the low-weighted ones that look 'well in' on paper. Money moves from the top of the weights to the bottom, and the prices of better-class horses drift past their true chance.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.08 to 1.15 on 600+ bets; the age >= 8 subset around 1.10, with the same-race lower-weighted runners below par",
      "falsified_if": "A/E at or below 1.0 on 500+ bets, or the age >= 8 refinement scoring no better than its same-race field (the young-stayer lift then stands alone), or the lift existing only beyond 14/1 (a few longshot winners, not class)"
    },
    "notes": "I kept the rule sketch unchanged: it uses only pre-off columns, and weight_lbs and bottom_weight_lbs were already used in H-ebbea470. I made the pass criteria stricter (300 bets, A/E 1.07) because the population should be large, so a thin pass would be suspicious. The age >= 8 refinement is the one that actually answers the rival question. The weight-position split mirrors H-ebbea470 (0.48 on 124 for light-weighted older horses), so the heavier and lighter groups should be read side by side."
  }

Timeline

1. Job 1745 finished

   16 Oct, 14:17 · engine:compute-1

2. hypothesis note H-d473ec0f

   16 Oct, 14:17 · engine:perturb

3. perturbation result for H-d473ec0f (neighbour: drop: weight_lbs > bottom_weight_lbs + 7): pass

   16 Oct, 14:17 · engine:perturb

4. perturbation result for H-d473ec0f (neighbour: drop: distance_yards >= 6160): fail

   16 Oct, 14:17 · engine:perturb

5. perturbation result for H-d473ec0f (neighbour: drop: is_handicap = 1): pass

   16 Oct, 14:17 · engine:perturb

6. perturbation result for H-d473ec0f (neighbour: drop: race_type = 'Chase'): pass

   16 Oct, 14:17 · engine:perturb

7. perturbation result for H-d473ec0f (neighbour: >= 6160 -> 7392 (+20%)): pass

   16 Oct, 14:17 · engine:perturb

8. perturbation result for H-d473ec0f (neighbour: >= 6160 -> 4928 (-20%)): pass

   16 Oct, 14:17 · engine:perturb

9. compute-1 took job 1745 (mutate, attempt 1)

   16 Oct, 14:17 · compute-1

10. Job 1769 finished

    16 Oct, 14:12 · engine:compute-1

11. H-d473ec0f Vault → KILLEDfailed the vault vault CONTRADICTED: A/E 0.981 on 1378 bets, p 0.5948 against BH threshold 0.002439 over 41 vault tests

    16 Oct, 14:12 · engine:orchestrator

12. vault result for H-d473ec0f (vault query (contradicted)): fail

    16 Oct, 14:12 · engine:orchestrator

13. H-d473ec0f Frozen → FROZENvault vault query started

    16 Oct, 14:12 · engine:orchestrator

14. compute-1 took job 1769 (vault, attempt 1)

    16 Oct, 14:12 · compute-1

15. Job 1763 finished

    16 Oct, 14:12 · engine:compute-1

16. Queued vault on the compute lane: frozen: one sealed vault query

    16 Oct, 14:12 · engine:scheduler