Short-priced first-time-out four-year-olds: rival to the handicap-mark story

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-52d3c28e (Irish lightly raced four-year-old returners in spring handicaps are fu) was predicted negative but came out the other way: A/E 1.35 on 135 bets (p 0.212 for the opposite direction). O-6b28e152
   • H-1a3e129f (Lightly raced four-year-old returners in Flat non-handicaps: progress ) passed discovery but was falsified: IRE A/E 0.861; driven by one price band (up to 3/1): the rest score 1.00. O-d5ec2505
2. AskedDo short-priced lightly raced four-year-old returners in non-handicaps (up to 3/1) beat their prices because the yard's confidence shows in the market, or is that band one of many segments? testing
3. MechanismIf market-signalled readiness is the cause, it needs no handicap mark. Fancied lightly raced four-year-olds back from a break should beat their prices in handicaps and non-handicaps alike.
4. Why the market misses itThe absence and the thin record make the price drift against strong stable confidence. The headline 'first run for 150 days' weighs heavier with the public than the yard's money does.
5. Predictedpositive, A/E 1.10 to 1.20 on 500+ bets across both race kinds
6. Falsified ifA/E below 1.05 on 500+ bets, or the non-handicap part below 1.0 while handicaps carry it all, which would put the mark back as the cause
7. Rule sketchrace_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank <= 3

Mechanism and rationale

If market-signalled readiness is the cause, it needs no handicap mark. Fancied lightly raced four-year-olds back from a break should beat their prices in handicaps and non-handicaps alike.

The rule

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

Reads: race_type, month, age, prior_runs, days_since_last_run, sp_rank

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

4fe928f9dec033eabd566392f361688f70383a44c5b96df1e3f39158d30af6f8

Frozen

not frozen

The definition can still change.

Confounders

• Favourite-longshot bias: the top three in the betting beat their SP-implied chance in general, so the lift must clear the base A/E of all sp_rank <= 3 Flat runners in March to May, not just 1.0
• Group, Listed and maiden fields with odds-on standouts: these add many short-priced bets with little information and can dilute or inflate A/E
• The non-handicap sample is small, so the per-kind split behind the falsification test may be noisy
• Yard concentration: a few big stables (e.g. Gosden, Appleby, Irish majors) may supply most fancied returners, so the result could be a trainer effect
• Overlap with the handicap returner base (H126) and the sibling fancied-returner spec H-1a6f83c9: the handicap part is not independent evidence
• Irish runners underperform in earlier cuts, so the country mix can move the headline

Control

{
  "description": "Same-race other runners should score about 1.0 or a little below, because the field absorbs the overround and the selections take win share from them. That shows the lift is not a property of these races (for example, weak spring fields). The main extra control is the same returner profile outside the first three in the betting (sp_rank > 3). If market-signalled readiness is the cause, the unfancied returners should score about 1.0 or below (earlier cuts went 0 wins from 48 to 86 at 16/1 or bigger), so the fancied group should clearly beat them. If the unfancied returners score as well as the fancied ones, the market signal is not what drives the lift.",
  "rule": "race_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank > 3"
}

Discovery result

pass broad rule (niche) · 16 Oct, 12:49

Bets

251

66 winners

A/E

1.18

95%: 0.91 to 1.50

SP return

+4.5%

level stakes

Betfair SP return

+19.6%

251 bets with a BSP

Strike rate 26.3%. p = 0.11. Same-race control: the other runners in these races had A/E 0.93, 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. 4 of 5 survived.

• partition: Non-handicap part only: readiness without a markpass

  83 bets · A/E 1.14 (0.72 to 1.72) · control 0.90 · 16 Oct, 12:51

  Breakdown

  betsA/E0.71.01.3

  2025-10-01 to 2026-04-27421.11

  after 2026-04-27411.19

• control: Control: fancied lightly raced 4yo in spring that are NOT returning from a breakfail

  463 bets · A/E 0.97 (0.79 to 1.19) · control 1.01 · 16 Oct, 12:51

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 · 33s · 16 Oct, 12:45

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank <= 3",
    "refinements": [
      {
        "label": "non-handicap part only (the rival test: readiness without a mark)",
        "rule": "race_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank <= 3 AND is_handicap = 0"
      },
      {
        "label": "handicap part only (comparison to the non-handicap part)",
        "rule": "race_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank <= 3 AND is_handicap = 1"
      },
      {
        "label": "GB only (Ireland has repeatedly blanked in the non-handicap and side cuts)",
        "rule": "race_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank <= 3 AND region = 'GB'"
      }
    ],
    "control": {
      "description": "Same-race other runners should score about 1.0 or a little below, because the field absorbs the overround and the selections take win share from them. That shows the lift is not a property of these races (for example, weak spring fields). The main extra control is the same returner profile outside the first three in the betting (sp_rank > 3). If market-signalled readiness is the cause, the unfancied returners should score about 1.0 or below (earlier cuts went 0 wins from 48 to 86 at 16/1 or bigger), so the fancied group should clearly beat them. If the unfancied returners score as well as the fancied ones, the market signal is not what drives the lift.",
      "rule": "race_type = 'Flat' AND month IN (3, 4, 5) AND age = 4 AND prior_runs <= 10 AND days_since_last_run >= 150 AND sp_rank > 3"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.02,
      "max_p": 0.2
    },
    "confounders": [
      "Favourite-longshot bias: the top three in the betting beat their SP-implied chance in general, so the lift must clear the base A/E of all sp_rank <= 3 Flat runners in March to May, not just 1.0",
      "Group, Listed and maiden fields with odds-on standouts: these add many short-priced bets with little information and can dilute or inflate A/E",
      "The non-handicap sample is small, so the per-kind split behind the falsification test may be noisy",
      "Yard concentration: a few big stables (e.g. Gosden, Appleby, Irish majors) may supply most fancied returners, so the result could be a trainer effect",
      "Overlap with the handicap returner base (H126) and the sibling fancied-returner spec H-1a6f83c9: the handicap part is not independent evidence",
      "Irish runners underperform in earlier cuts, so the country mix can move the headline"
    ],
    "mechanism": "The rival explanation: the returner lift comes from readiness that the market signals, not from a handicap mark that lags behind improvement. Lightly raced four-year-olds back after 150+ days that still trade in the first three in the betting are those whose stables have shown fitness and intent. If that is the cause, the edge needs no mark, so it should appear in non-handicaps as well as handicaps. Earlier observations support this: the non-handicap version held only in the up-to-3/1 band.",
    "market_miss": "'First run for 150 days' and a thin public record make the public lay these horses. Even when stable money keeps them near the top of the market, their price drifts longer than their true readiness justifies. The public overweights the absence; the market underweights the yard's confidence.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.20 on 500+ bets across handicaps and non-handicaps combined",
      "falsified_if": "Falsified if the broad rule scores A/E below 1.05 on 500+ bets, or if the non-handicap part scores below 1.0 while handicaps carry all the lift (which would put the mark back as the cause), or if unfancied returners (sp_rank > 3) do as well as the fancied ones."
    },
    "notes": "The draft rule is kept as it was. All columns are pre-off, and sp_rank is the betting rank at the off, which matches SP settlement. The rule is broad on purpose: no is_handicap condition, because the hypothesis is that the edge does not depend on the mark. The handicap and non-handicap refinements exist to run the falsification test. min_bets is raised to 300 because the prediction names 500+ bets and short-priced A/E needs volume. Judge the lift against the general sp_rank <= 3 baseline before crediting it to readiness."
  }

Timeline

1. Job 1459 finished

   16 Oct, 12:51 · engine:compute-1

2. H-3c53e905 Falsification → KILLED falsified: IRE A/E 0.887

   16 Oct, 12:51 · engine:falsification

3. falsification result for H-3c53e905 (price_band: Exclude odds-on and the 8/1+ tail: the 2/1 to 7/1 core): pass

   16 Oct, 12:51 · engine:evaluator

4. compute-1 took job 1459 (attack, attempt 1)

   16 Oct, 12:51 · compute-1

5. Job 1458 finished

   16 Oct, 12:51 · engine:compute-1

6. falsification result for H-3c53e905 (perturbation: Perturb prior_runs to 8 and favourite rank to top two): pass

   16 Oct, 12:51 · engine:evaluator

7. compute-1 took job 1458 (attack, attempt 1)

   16 Oct, 12:51 · compute-1

8. Job 1457 finished

   16 Oct, 12:51 · engine:compute-1

9. falsification result for H-3c53e905 (perturbation: Perturb layoff threshold to 120 and 180 days): pass

   16 Oct, 12:51 · engine:evaluator

10. compute-1 took job 1457 (attack, attempt 1)

    16 Oct, 12:51 · compute-1

11. Job 1456 finished

    16 Oct, 12:51 · engine:compute-1

12. falsification result for H-3c53e905 (control: Control: fancied lightly raced 4yo in spring that are NOT returning from a break): fail

    16 Oct, 12:51 · engine:evaluator

13. compute-1 took job 1456 (attack, attempt 1)

    16 Oct, 12:51 · compute-1

14. Job 1455 finished

    16 Oct, 12:51 · engine:compute-1

15. falsification result for H-3c53e905 (partition: Non-handicap part only: readiness without a mark): pass

    16 Oct, 12:51 · engine:evaluator

16. compute-1 took job 1455 (attack, attempt 1)

    16 Oct, 12:51 · compute-1

17. Job 1448 finished

    16 Oct, 12:50 · engine:ai-1

18. Queued attack on the compute lane: attack: Exclude odds-on and the 8/1+ tail: the 2/1 to 7/1 core

    16 Oct, 12:50 · engine:scheduler

19. Queued attack on the compute lane: attack: Perturb prior_runs to 8 and favourite rank to top two

    16 Oct, 12:50 · engine:scheduler