Long-priced four-year-old handicap returners: is the 16/1+ tail priced right?

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-1a6f83c9 (Fancied lightly raced four-year-old returners: the market flags which ) met the sealed vault supportive: A/E 1.11 on 1567 bets (p 0.0484, threshold 0.0031). O-c4322e77
   • H-3c53e905 (Short-priced first-time-out four-year-olds: rival to the handicap-mark) passed discovery as a niche: A/E 1.18 on 251 bets (p 0.108), same-race control 0.93. O-da190682
2. AskedWithin the H126 handicap base, do returners priced 16/1 or bigger (sp_decimal > 15) score at or below 0.90, or as well as the capped group? testing
3. MechanismIf stable confidence moves the ready horses' prices, the drifters are the horses needing the run and should win about as often as their prices say, or less. If instead the cause is winter improvement that the mark misses, the improvement is spread across all returners and the long-priced ones should beat their prices too.
4. Why the market misses itUnder the readiness story, a long price correctly reflects an unfit horse. Under the improvement story, the long price rests on old three-year-old figures, and the form book shows weak runs that understate a stronger horse.
5. Predictednegative, A/E 0.80 to 0.95 on 500+ bets
6. Falsified ifA/E of 1.05 or higher on 500+ bets, which would mean the price cap is not where the lift lives and the whole base should be kept

Mechanism and rationale

If stable confidence moves the ready horses' prices, the drifters are the horses needing the run and should win about as often as their prices say, or less. If instead the cause is winter improvement that the mark misses, the improvement is spread across all returners and the long-priced ones should beat their prices too.

The rule

race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_decimal > 15

Reads: race_type, is_handicap, month, age, prior_runs, days_since_last_run, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

ebcea9c7839c624e18cb28457fb5cb1e0a9ea04181043ec3f5783f6608a2a52a

Frozen

not frozen

The definition can still change.

Confounders

• High variance at long prices: two or three winners at 33/1 or longer can move A/E by 0.2 or more on a few hundred bets. Refinement 2 caps the price at 33/1 and the price-matched control is read alongside.
• Favourite-longshot bias pulls 16/1+ A/E below 1.00 whether or not the horse is a returner, so the raw A/E has to be compared with same-price non-returners (the control rule) and the same-race 16/1+ runners, not with 1.00 alone.
• Surface mix: long-priced returners on the all-weather went 0 from 48, so a negative overall result could be an all-weather effect rather than a readiness effect. Refinement 1 checks turf only.
• Overlap with prior evidence: unfancied returners (sp_rank > 3) already scored 1.11 on 450, and 1.19 on 245 at 16/1+, which largely overlaps this rule. That leans against the negative prediction and must not be counted as independent confirmation either way.
• Field size and race class: long prices cluster in big, competitive handicaps, where every runner's SP is longer. That can mix 'unfancied' up with 'big field'.

Control

Discovery result

fail broad rule · 16 Oct, 13:07

Bets

182

4 winners

A/E

0.99

95%: 0.27 to 2.55

SP return

-61.5%

level stakes

Betfair SP return

-48.4%

182 bets with a BSP

Strike rate 2.2%. p = 0.57. Same-race control: the other runners in these races had A/E 1.00, 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 · 38s · 16 Oct, 13:02

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_decimal > 15",
    "refinements": [
      {
        "label": "Turf only (removes the all-weather surface-mix confounder, which went 0 from 48 at long prices)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_decimal > 15 AND surface = 'Turf'"
      },
      {
        "label": "16/1 to 33/1 only (drops the extreme tail, where one or two winners swing A/E and the favourite-longshot bias is strongest)",
        "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_decimal > 15 AND sp_decimal <= 34"
      }
    ],
    "control": {
      "description": "There are two comparisons. (1) Same-race runners: the other runners in these spring Flat handicaps should come out near 1.00 overall. Their 16/1+ runners should sit below 1.00 (about 0.85 to 0.95) because of the ordinary favourite-longshot bias. The test of the readiness story is whether long-priced returners match or fall below that same-price baseline. Beating it would support the improvement story. (2) The control rule below is a price-matched non-returner group: four-year-olds with 10 or fewer runs in spring handicaps at 16/1+ who have run within the last 150 days. If returners score clearly above this group, the break is what carries the edge even at long prices, and the negative prediction fails. If both groups score below 1.00 and close together, long-priced returners are simply priced like other long shots, which is what the readiness story predicts. The capped sibling (sp_decimal <= 15: 1.21 discovery, 1.11 vault) is the reference for 'matches the capped group'.",
      "rule": "race_type = 'Flat' AND is_handicap = 1 AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run < 150 AND sp_decimal > 15"
    },
    "pass_criteria": {
      "min_bets": 200,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "High variance at long prices: two or three winners at 33/1 or longer can move A/E by 0.2 or more on a few hundred bets. Refinement 2 caps the price at 33/1 and the price-matched control is read alongside.",
      "Favourite-longshot bias pulls 16/1+ A/E below 1.00 whether or not the horse is a returner, so the raw A/E has to be compared with same-price non-returners (the control rule) and the same-race 16/1+ runners, not with 1.00 alone.",
      "Surface mix: long-priced returners on the all-weather went 0 from 48, so a negative overall result could be an all-weather effect rather than a readiness effect. Refinement 1 checks turf only.",
      "Overlap with prior evidence: unfancied returners (sp_rank > 3) already scored 1.11 on 450, and 1.19 on 245 at 16/1+, which largely overlaps this rule. That leans against the negative prediction and must not be counted as independent confirmation either way.",
      "Field size and race class: long prices cluster in big, competitive handicaps, where every runner's SP is longer. That can mix 'unfancied' up with 'big field'."
    ],
    "mechanism": "This is a test between two rival stories. Readiness story: stable confidence and work-watcher money move the prices of returners who are fit to win, so the drifters (16/1+) are the ones needing the run. They should win no more often than their prices say, and probably less: A/E 0.80 to 0.95. Improvement story: four-year-olds with 10 or fewer runs improve physically over the winter, and the handicap mark (built on three-year-old form) misses this. That improvement is spread across all returners whatever their price, so the 16/1+ group should beat its prices about as much as the capped group does (about 1.1 or more).",
    "market_miss": "Under the readiness story there is no miss at long prices: the long SP correctly marks an unfit horse, and the lift is confined to the fancied returners. Under the improvement story, the market prices long-priced returners off stale three-year-old figures and weak form-book runs. Those understate a physically stronger horse, and the price signal does not correct for that because there is no recent public evidence of the improvement.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.80 to 0.95 on about 300 to 600 bets, close to or below price-matched non-returners at 16/1+",
      "falsified_if": "A/E of 1.05 or higher on 200+ bets, or clearly above the price-matched non-returner control. Either would mean the lift does not depend on price, the readiness story fails, and the whole uncapped base should be kept."
    },
    "notes": "The predicted direction is negative, so the vault should judge this as over-bet. The pass_criteria floors are written in the lab's standard form; for a negative rule, read min_ae 1.05 as A/E at or below 1/1.05 (about 0.95) with p < 0.05 in the lower tail. Prior evidence leans against the prediction: unfancied returners at 16/1+ scored 1.19 on 245. A positive result is therefore a live outcome and would itself be informative, because it falsifies the readiness story in favour of winter improvement. If that happens, record it as a falsification of this hypothesis, not as a new long-price system, unless it survives independent testing. min_bets was raised to 200 because of the long-price variance. The rule is unchanged from the sketch; all columns are pre-off."
  }

Timeline

1. Job 1528 finished

   16 Oct, 13:07 · engine:compute-1

2. H-0078ee33 Discovery → KILLED fails: 182 bets, needs 300; A/E 0.995, needs 0.95 or lower; p 0.6245 (lower tail), needs 0.05; periods [0.41, 1.86], each needs 0.99 or lower; A/E 0.995 against same-race control 1.002, needs 0.03 below

   16 Oct, 13:07 · engine:evaluator

3. discovery result for H-0078ee33 (broad rule): fail

   16 Oct, 13:07 · engine:evaluator

4. H-0078ee33 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 13:07 · engine:orchestrator

5. compute-1 took job 1528 (evaluate, attempt 1)

   16 Oct, 13:07 · compute-1

6. Job 1525 finished

   16 Oct, 13:03 · engine:ai-1

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

   16 Oct, 13:03 · engine:scheduler

8. H-0078ee33 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 13:03 · agent:designer@v1

9. ai-1 took job 1525 (design, attempt 1)

   16 Oct, 13:02 · ai-1

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

    16 Oct, 13:02 · engine:scheduler

11. New idea H-0078ee33: Long-priced four-year-old handicap returners: is the 16/1+ tail priced right? (rival-explanation)

    16 Oct, 13:02 · 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.