Rival: the big-handicap front-of-market lift is a jumps effect, and the period fade is the switch to the flat

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-2ae0cf0d (Replicate big-yard top-two lift with a fixed, time-stable definition o): the period '2025-10-01 to 2026-04-27' carries the effect (A/E 1.33) while the rest score after 2026-04-27 0.92. O-1bd420a0
   • H-0939d0ac (Rival: yard quality, not yard size, drives the big-handicap top-two li): the country 'GB' carries the effect (A/E 1.33) while the rest score IRE 0.64. O-5f33166a
   • H-2ae0cf0d (Replicate big-yard top-two lift with a fixed, time-stable definition o) pointed the predicted way but failed the bar: A/E 1.13 on 196 bets (p 0.303), control 1.0. O-91bc292c
   • H-659f3a37 (Big-yard runners at the front of the market in big handicaps beat thei) met the sealed vault supportive: A/E 1.14 on 1003 bets (p 0.0748, threshold 0.0022). O-c42ebd58
2. AskedIs the period fade in big-handicap top-two pools (1.33 Oct–Apr, 0.92 after) really a code effect: lifted in jumps handicaps and at par in flat handicaps? testing
3. MechanismBig-field jumps handicaps are full of lightly raced, plot-aimed horses whose real ability is hidden by form over different trips and obstacles. The horses the market settles on at the front are often the yard's well-handicapped targets. Big flat handicaps are run by exposed horses whose marks are well known, so the front of the market has less hidden information.
4. Why the market misses itAt 5/1 to 10/1 the front-of-market jumps runner is priced like any 6/1 shot. The form book shows unplaced runs over the wrong trip or ground, but not the yard's targeting, so the price stays too long. On the flat the same price is accurate because the form is exposed.

Mechanism and rationale

Big-field jumps handicaps are full of lightly raced, plot-aimed horses whose real ability is hidden by form over different trips and obstacles. The horses the market settles on at the front are often the yard's well-handicapped targets. Big flat handicaps are run by exposed horses whose marks are well known, so the front of the market has less hidden information.

The rule

is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND race_type IN ('Hurdle','Chase')

Reads: is_handicap, sp_rank, sp_decimal, field_size, race_type

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

f1b735ba08335c6a54af1e8c9dddde1e0a4e22fd6fd03612d4bd021910553cc5

Frozen

not frozen

The definition can still change.

Confounders

• Cheltenham, Aintree, Punchestown and Leopardstown festival handicaps fall in Dec to April, inside the early period. A jumps lift may really be a festival-week lift. Check the result with March excluded and by meeting.
• Season confound: jumps big handicaps are mostly winter and flat ones summer, so a code split and a period fade cannot be told apart within one season. Compare jumps across two winters, not jumps in winter with flat in summer.
• GB/IRE split: Irish top-two big-handicap runners were at par earlier (small yards 0.88). Report GB and IRE jumps separately; a lift that is GB only reproduces the old split rather than explaining it.
• Overround (book_pct) is higher in big jumps fields, which shifts A/E against SP. Compare with the same-race control and BSP ROI, since SP ROI was earlier far below BSP ROI (+1.6% vs +27.4%).
• prior_handicap_runs counts only from July 2020, so early-period runners look less exposed than they are. This affects the lightly-raced refinement.
• Short forward sample after April: jumps big handicaps are rare from May to October, so forward bets build slowly and depend on the season.

Discovery result

fail broad rule · 16 Oct, 16:02

Bets

365

39 winners

A/E

0.98

95%: 0.70 to 1.34

SP return

-28.6%

level stakes

Betfair SP return

-10.9%

359 bets with a BSP

Strike rate 10.7%. p = 0.57. Same-race control: the other runners in these races had A/E 1.03, 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 · 45s · 16 Oct, 16:01

  Output (schema-checked)

  {
    "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND race_type IN ('Hurdle','Chase')",
    "refinements": [
      {
        "label": "5/1 to 10/1 only",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND sp_decimal <= 11 AND field_size >= 14 AND race_type IN ('Hurdle','Chase')"
      },
      {
        "label": "lightly raced in handicaps (mark not yet exposed)",
        "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND race_type IN ('Hurdle','Chase') AND prior_handicap_runs <= 5"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) Same-race control, measured by the lab as usual: the other runners in these big jumps handicaps should score about 1.00 (0.95 to 1.02), so any lift is the front-of-market runner, not a property of the race or overround. (2) Code-split control, which is the point of this rival: the identical rule restricted to Flat big handicaps. The hypothesis needs flat to be at par (1.00 or below) and at least 0.05 below jumps. If flat scores the same as jumps, the lift is not a code effect and the period fade is not the switch from jumps to flat.",
      "rule": "is_handicap = 1 AND sp_rank <= 2 AND sp_decimal >= 6 AND field_size >= 14 AND race_type = 'Flat'"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.08,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Cheltenham, Aintree, Punchestown and Leopardstown festival handicaps fall in Dec to April, inside the early period. A jumps lift may really be a festival-week lift. Check the result with March excluded and by meeting.",
      "Season confound: jumps big handicaps are mostly winter and flat ones summer, so a code split and a period fade cannot be told apart within one season. Compare jumps across two winters, not jumps in winter with flat in summer.",
      "GB/IRE split: Irish top-two big-handicap runners were at par earlier (small yards 0.88). Report GB and IRE jumps separately; a lift that is GB only reproduces the old split rather than explaining it.",
      "Overround (book_pct) is higher in big jumps fields, which shifts A/E against SP. Compare with the same-race control and BSP ROI, since SP ROI was earlier far below BSP ROI (+1.6% vs +27.4%).",
      "prior_handicap_runs counts only from July 2020, so early-period runners look less exposed than they are. This affects the lightly-raced refinement.",
      "Short forward sample after April: jumps big handicaps are rare from May to October, so forward bets build slowly and depend on the season."
    ],
    "mechanism": "Big-field jumps handicaps are full of lightly raced, plot-aimed horses whose real ability is hidden by form over the wrong trip, ground or obstacle. When the market puts such a horse in the top two at 5/1+, that often reflects the yard's targeting of a well-handicapped horse, and the price does not fully account for it. Big flat handicaps are run by exposed horses with well-known marks, so the front of the market holds less hidden information and should be priced accurately. If this is right, the family's big-handicap front-of-market lift lives in jumps races, and its fade after April is the calendar moving to flat racing.",
    "market_miss": "At 5/1 to 10/1 a front-of-market jumps handicapper is priced like any 6/1 shot. Its visible form shows unplaced runs over the wrong trip, ground or obstacle. The yard's targeting, and how far the horse is under its mark, do not show, so the price stays too long. On the flat the form is exposed and the same price is accurate.",
    "prediction": {
      "direction": "positive",
      "size": "Jumps A/E 1.10 to 1.20 on 600+ bets, with the flat control at 1.00 or below",
      "falsified_if": "Jumps A/E is 1.03 or below on 400+ bets, or the flat control rule scores within 0.05 of jumps. Either means the lift and its period fade are not a jumps/flat code split."
    },
    "notes": "I kept the rule sketch unchanged: every column is pre-off, and race_type has no NH Flat value and bumpers are not handicaps, so IN ('Hurdle','Chase') covers all jumps handicaps. The pass bar is stricter than the floor (300 bets, A/E 1.08), because the family's earlier lifts shrank sharply in the vault (1.33 to 1.14). The flat control is central: a jumps pass with flat also lifted does not support this hypothesis. Read period splits within jumps across separate winters (for example 2024-25 against 2025-26) so the season is not confused with the code. Refinement 1 follows the hypothesis's own 5/1 to 10/1 framing. Refinement 2 tests the 'mark not yet exposed' part of the mechanism directly."
  }

Timeline

1. Job 1956 finished

   16 Oct, 16:02 · engine:compute-1

2. H-12f04c30 Discovery → KILLED fails: A/E 0.981, needs 1.08; p 0.5685, needs 0.05 or lower; periods [1.18, 0.71], each needs 1.0; A/E 0.981 against same-race control 1.03, needs 0.03 above (not a niche pass: A/E 0.981 below 1.08; p 0.568478005324285 above 0.5; does not beat the same-race control 1.03; a period points the other way [1.18, 0.71])

   16 Oct, 16:02 · engine:evaluator

3. discovery result for H-12f04c30 (broad rule): fail

   16 Oct, 16:02 · engine:evaluator

4. H-12f04c30 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 16:02 · engine:orchestrator

5. compute-1 took job 1956 (evaluate, attempt 1)

   16 Oct, 16:02 · compute-1

6. Job 1955 finished

   16 Oct, 16:02 · engine:ai-1

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

   16 Oct, 16:02 · engine:scheduler

8. H-12f04c30 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 16:02 · agent:designer@v1

9. ai-1 took job 1955 (design, attempt 1)

   16 Oct, 16:01 · ai-1

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

    16 Oct, 16:01 · engine:scheduler

11. New idea H-12f04c30: Rival: the big-handicap front-of-market lift is a jumps effect, and the period fade is the switch to the flat (rival-explanation)

    16 Oct, 16:01 · 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.