Long-priced early-spring handicap returners are over-bet: the 16/1+ tail needs the run

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-5167b372 (All early-spring Flat handicap returners: a calendar effect rival to t) met the sealed vault supportive: A/E 1.03 on 6934 bets (p 0.2352, threshold 0.0026). O-352505b3
   • H-5167b372 (All early-spring Flat handicap returners: a calendar effect rival to t): the country 'GB' carries the effect (A/E 1.11) while the rest score IRE 1.0. O-972b7090
   • H-5167b372 (All early-spring Flat handicap returners: a calendar effect rival to t) passed discovery as a niche: A/E 1.09 on 1424 bets (p 0.173), same-race control 0.98. O-c14824ce
2. AskedAre long-priced (16/1+) early-spring Flat handicap returners over-bet beyond the general longshot shortfall, while 16/1+ returners later in the year are not? answered
3. MechanismEarly in the season, trainers use a first run back for two purposes. Returners they have primed are fancied and go off short. Returners at long prices are mostly getting a pipe-opener before their real targets, short of peak fitness and not aimed to win. The same trainer-intent story that would lift the short end of the returner pool predicts a shortfall at the long end.
4. Why the market misses itBig-field spring handicaps draw heavy each-way and novelty money. A fresh horse at 20/1 with a decent old mark looks like value, because the form book shows only the mark and the layoff, not whether the trainer intends this run as preparation. Long-price SPs are therefore shorter than these horses' real chances.
5. Predictednegative, A/E 0.72 to 0.88 on 500+ bets
6. Falsified ifA/E of 0.95 or higher on 500+ bets, or a same-race control of 16/1+ non-returners scoring as low (within 0.05), which would make it plain favourite-longshot bias

The question's answer so far: Not established. The 16/1+ early-spring returners scored 0.742 on 563 (p 0.86, re-tested on the same data), and AW 16/1+ returners year-round scored 1.28 on 626, so there is no general longshot-returner over-bet.

Mechanism and rationale

Early in the season, trainers use a first run back for two purposes. Returners they have primed are fancied and go off short. Returners at long prices are mostly getting a pipe-opener before their real targets, short of peak fitness and not aimed to win. The same trainer-intent story that would lift the short end of the returner pool predicts a shortfall at the long end.

The rule

race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4) AND days_since_last_run >= 150 AND sp_decimal > 15

Reads: race_type, is_handicap, month, days_since_last_run, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

7b0fe34066da75c12fef82e6b7a7dbc553e7cca059444b0987e7461c7bcec1f9

Frozen

not frozen

The definition can still change.

Confounders

• General favourite-longshot bias: all March-April 16/1+ Flat handicap runners scored 0.91 on 2211, so the returner shortfall has to beat that baseline, not 1.0
• Few wins (10 of 563) mean two or three results can swing A/E widely, and p-values on a negative edge rest on very few winners
• Discovery data is essentially one spring (2026). A single wet or dry spring, or one trainer's habits, could produce the whole effect
• Big-field heritage handicaps (Lincoln, spring cups) may dominate the sample and bring their own each-way pricing
• Year-round 16/1+ returners scored 1.118 on 1821, so the effect must be specific to early season, and a calendar fluke is plausible
• The overround (book_pct) is higher in big fields and inflates the longshot shortfall in every group, not only returners
• Irish and GB pricing differ; the thread has repeatedly found region-dependent results

Control

{
  "description": "The main control is the 16/1+ runners in the same March-April Flat handicaps that are NOT returners (last ran under 150 days ago). They carry the same field size, race type and favourite-longshot bias. The mechanism predicts they score near the general 16/1+ spring baseline (about 0.91), and the returners well below that (at least 0.05 lower). The lab's same-race measure of all other runners should come out near 1.0. If 16/1+ non-returners score as low as the returners (within 0.05), the effect is plain longshot bias and the hypothesis is falsified. A second check is year-round 16/1+ returners (earlier 1.118 on 1821): the shortfall has to be specific to early spring.",
  "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4) AND days_since_last_run < 150 AND sp_decimal > 15"
}

Discovery result

fail broad rule · 16 Oct, 13:58

Bets

563

10 winners

A/E

0.74

95%: 0.36 to 1.36

SP return

-57.5%

level stakes

Betfair SP return

-35.8%

563 bets with a BSP

Strike rate 1.8%. p = 0.86. Same-race control: the other runners in these races had A/E 1.01, 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, 13:53

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4) AND days_since_last_run >= 150 AND sp_decimal > 15",
    "refinements": [
      {
        "label": "Big fields (each-way and novelty money)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4) AND days_since_last_run >= 150 AND sp_decimal > 15 AND field_size >= 12"
      },
      {
        "label": "True winter layoff (180+ days, clearer pipe-opener)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4) AND days_since_last_run >= 180 AND sp_decimal > 15"
      }
    ],
    "control": {
      "description": "The main control is the 16/1+ runners in the same March-April Flat handicaps that are NOT returners (last ran under 150 days ago). They carry the same field size, race type and favourite-longshot bias. The mechanism predicts they score near the general 16/1+ spring baseline (about 0.91), and the returners well below that (at least 0.05 lower). The lab's same-race measure of all other runners should come out near 1.0. If 16/1+ non-returners score as low as the returners (within 0.05), the effect is plain longshot bias and the hypothesis is falsified. A second check is year-round 16/1+ returners (earlier 1.118 on 1821): the shortfall has to be specific to early spring.",
      "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4) AND days_since_last_run < 150 AND sp_decimal > 15"
    },
    "pass_criteria": {
      "min_bets": 400,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "General favourite-longshot bias: all March-April 16/1+ Flat handicap runners scored 0.91 on 2211, so the returner shortfall has to beat that baseline, not 1.0",
      "Few wins (10 of 563) mean two or three results can swing A/E widely, and p-values on a negative edge rest on very few winners",
      "Discovery data is essentially one spring (2026). A single wet or dry spring, or one trainer's habits, could produce the whole effect",
      "Big-field heritage handicaps (Lincoln, spring cups) may dominate the sample and bring their own each-way pricing",
      "Year-round 16/1+ returners scored 1.118 on 1821, so the effect must be specific to early season, and a calendar fluke is plausible",
      "The overround (book_pct) is higher in big fields and inflates the longshot shortfall in every group, not only returners",
      "Irish and GB pricing differ; the thread has repeatedly found region-dependent results"
    ],
    "mechanism": "At the start of the Flat turf season, trainers use first runs back for two purposes. Returners they have primed to win are backed and go off short. Returners sent off at 16/1+ are mostly having a pipe-opener before later targets: short of peak fitness and not aimed to win. So the same trainer-intent story that lifts the short end of the returner pool predicts that the long end falls below its prices. It also predicts a bigger shortfall than for other 16/1+ runners in the same races, who are race-fit.",
    "market_miss": "Big-field spring handicaps draw heavy each-way and novelty money. A fresh horse at 20/1 on a reasonable old mark looks like value, because the form book shows the mark and the layoff but not whether the trainer means this run as preparation. Recreational money pushes these SPs shorter than their real chances. Sharp money does not correct this, because laying big-priced outsiders at SP is not worth the effort.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.72 to 0.88 on 500+ bets, at least 0.05 below same-race 16/1+ non-returners",
      "falsified_if": "A/E of 0.95 or higher on 400+ bets, or 16/1+ non-returners in the same March-April handicaps scoring within 0.05 of the returners (which would make it plain favourite-longshot bias)"
    },
    "notes": "The hypothesis predicts a negative direction, so I assume the lab reads min_ae 1.05 in that direction: passing means A/E at or below about 0.952 (1/1.05), with p ≤ 0.05. I raised min_bets to 400 because few wins make thin samples unreadable. The month column is used as in earlier family rules. The refinements come from the market_miss (big each-way fields) and the mechanism (a true winter layoff means a pipe-opener). I added no region or trainer slicing."
  }

Timeline

1. Job 1719 finished

   16 Oct, 13:58 · engine:compute-1

2. H-5ad83d6a Discovery → KILLED fails: p 0.2134 (lower tail), needs 0.05; periods [0.64], each needs 0.99 or lower

   16 Oct, 13:58 · engine:evaluator

3. discovery result for H-5ad83d6a (broad rule): fail

   16 Oct, 13:58 · engine:evaluator

4. H-5ad83d6a Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 13:58 · engine:orchestrator

5. compute-1 took job 1719 (evaluate, attempt 1)

   16 Oct, 13:58 · compute-1

6. Job 1717 finished

   16 Oct, 13:53 · engine:ai-1

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

   16 Oct, 13:53 · engine:scheduler

8. H-5ad83d6a Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 13:53 · agent:designer@v1

9. ai-1 took job 1717 (design, attempt 1)

   16 Oct, 13:53 · ai-1

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

    16 Oct, 13:53 · engine:scheduler

11. New idea H-5ad83d6a: Long-priced early-spring handicap returners are over-bet: the 16/1+ tail needs the run (test-mechanism)

    16 Oct, 13:53 · 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.