Near-bottom-weight runners of all ages in marathon handicap chases are over-bet

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-d473ec0f (Rival: higher-weighted runners of all ages beat their price in maratho) passed discovery as a niche: A/E 1.21 on 338 bets (p 0.155), same-race control 0.6. O-4afbccc8
   • 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, are runners within 7 lb of bottom weight (all ages) over-bet, and is that the true source of the weight-position gap? testing
3. MechanismOver marathon trips the bottom of a handicap is filled with horses that are badly in. Some are out of the handicap and carry more than their mark. Others are exposed types the handicapper has dropped but who lack the class to win. The weight they save does not make up for the class they lack.
4. Why the market misses itPunters treat a low weight over a long trip as an advantage ('weight stops trains'). On the racecard, a light weight looks like a chance, while being out of the handicap is hidden in small print.
5. Predictednegative, A/E 0.60 to 0.85 on 160+ bets
6. Falsified ifA/E at or above 0.95 on 150+ bets, or no lower than the 1.21 scored by higher-weighted runners in the same races
7. Rule sketchrace_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs <= bottom_weight_lbs + 7

Mechanism and rationale

Over marathon trips the bottom of a handicap is filled with horses that are badly in. Some are out of the handicap and carry more than their mark. Others are exposed types the handicapper has dropped but who lack the class to win. The weight they save does not make up for the class they lack.

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)

781098ab08e46b8560f82305e6779174a772c1d3cbbe6159aee9507cb260da87

Frozen

not frozen

The definition can still change.

Confounders

• Longshot bias: light-weights are mostly 16/1+, and all big-priced runners are over-bet. Compare against the same-race control within price bands.
• Field size: light-weights cluster in big festival marathon fields (Grand National-type races), where every runner has a low strike rate
• Course concentration: a few marathon venues (Aintree, Cheltenham, Haydock, Uttoxeter, Fairyhouse, Punchestown) dominate the sample
• The 7 lb cut-off is arbitrary. A result that only appears at exactly +7 lb is fragile.
• Out-of-the-handicap status is not directly observable, so the rule only proxies it by weight position
• Overlap with the age result: older light-weights already scored 0.48 on 124, so an all-age result may be driven mostly by the 8+ subgroup
• Period drift: the parent young-stayer edge is weakening over time

Control

{
  "description": "The other runners in the same 3m4f+ handicap chases (those more than 7 lb above bottom weight). These are expected to sit at or above par: prior work puts them at A/E ~1.15-1.21. The over-bet is real only if the light-weights are clearly below both par and that same-race control. If both groups are low, the effect belongs to the race type and not to weight position. No separate control rule is needed.",
  "rule": null
}

Discovery result

fail broad rule · 16 Oct, 14:13

Bets

161

8 winners

A/E

0.60

95%: 0.26 to 1.18

SP return

-64.1%

level stakes

Betfair SP return

-54.8%

161 bets with a BSP

Strike rate 5.0%. p = 0.96. Same-race control: the other runners in these races had A/E 1.21, so the selection did not 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 · 35s · 16 Oct, 14:09

  Output (schema-checked)

  {
    "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs <= bottom_weight_lbs + 7",
    "refinements": [
      {
        "label": "8/1 or bigger: the price band where older light-weights went 0 from 92",
        "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs <= bottom_weight_lbs + 7 AND sp_decimal >= 9.0"
      },
      {
        "label": "Exposed chasers (10+ chase runs): the 'dropped but lacks the class' arm of the mechanism",
        "rule": "race_type = 'Chase' AND is_handicap = 1 AND distance_yards >= 6160 AND weight_lbs <= bottom_weight_lbs + 7 AND chase_runs >= 10"
      }
    ],
    "control": {
      "description": "The other runners in the same 3m4f+ handicap chases (those more than 7 lb above bottom weight). These are expected to sit at or above par: prior work puts them at A/E ~1.15-1.21. The over-bet is real only if the light-weights are clearly below both par and that same-race control. If both groups are low, the effect belongs to the race type and not to weight position. No separate control rule is needed.",
      "rule": null
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Longshot bias: light-weights are mostly 16/1+, and all big-priced runners are over-bet. Compare against the same-race control within price bands.",
      "Field size: light-weights cluster in big festival marathon fields (Grand National-type races), where every runner has a low strike rate",
      "Course concentration: a few marathon venues (Aintree, Cheltenham, Haydock, Uttoxeter, Fairyhouse, Punchestown) dominate the sample",
      "The 7 lb cut-off is arbitrary. A result that only appears at exactly +7 lb is fragile.",
      "Out-of-the-handicap status is not directly observable, so the rule only proxies it by weight position",
      "Overlap with the age result: older light-weights already scored 0.48 on 124, so an all-age result may be driven mostly by the 8+ subgroup",
      "Period drift: the parent young-stayer edge is weakening over time"
    ],
    "mechanism": "In 3m4f+ handicap chases, the bottom of the weights fills up with horses that are badly in. Some are out of the handicap and carry more than their mark says. Others are exposed, declining types the handicapper has dropped but who lack the class to win. Over a marathon trip, saving weight does not make up for a class gap, so runners within 7 lb of bottom weight win less often than their prices imply.",
    "market_miss": "Punters lean on 'weight stops trains' and see a light weight over a long trip as an advantage. The racecard shows the low weight clearly, but out-of-the-handicap status and long-handicap marks are in the small print. Light-weights are also mostly longshots in big, high-profile fields that draw casual each-way money. The result is prices that are too short for their real chance.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.60 to 0.85 on 160+ bets, clearly below the same-race heavier runners (~1.15-1.21)",
      "falsified_if": "A/E at or above 0.95 on 150+ bets, or the light-weights' A/E is not below that of the heavier runners in the same races (the same-race control)"
    },
    "notes": "I kept the rule sketch unchanged. Every column is pre-off and available from 2020-07 (weight_lbs and bottom_weight_lbs as used in earlier family rules). Direction is negative, so the vault should read the pass criteria as their mirror: a lift of at least 1.05 here means an A/E of 0.95 or below, with p at most 0.05 on the short side. Refinements come only from the reasoning: the 8/1+ price band from the 0-from-92 finding, and exposure as the 'dropped but lacks class' arm. Expect the sample to sit near the 150 floor, since older light-weights alone were 124 and young ones 37."
  }

Timeline

1. Job 1756 finished

   16 Oct, 14:13 · engine:compute-1

2. H-66bfadbb Discovery → KILLED fails: 161 bets, needs 300; p 0.0827 (lower tail), needs 0.05; periods [0.56], each needs 0.99 or lower

   16 Oct, 14:13 · engine:evaluator

3. discovery result for H-66bfadbb (broad rule): fail

   16 Oct, 14:13 · engine:evaluator

4. H-66bfadbb Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 14:13 · engine:orchestrator

5. compute-1 took job 1756 (evaluate, attempt 1)

   16 Oct, 14:13 · compute-1

6. Job 1753 finished

   16 Oct, 14:10 · engine:ai-1

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

   16 Oct, 14:10 · engine:scheduler

8. H-66bfadbb Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 14:10 · agent:designer@v1

9. ai-1 took job 1753 (design, attempt 1)

   16 Oct, 14:09 · ai-1

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

    16 Oct, 14:09 · engine:scheduler

11. New idea H-66bfadbb: Near-bottom-weight runners of all ages in marathon handicap chases are over-bet (test-mechanism)

    16 Oct, 14:09 · 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.