Rival: a change of handicap status, not of code, carries the switcher lift

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-070b2026 (Rival: held-up PU lift comes from a switch of race code since the pull) passed discovery as a niche: A/E 1.36 on 304 bets (p 0.092), same-race control 0.98. O-9a6df59c
2. AskedWithin the 226 jumps-to-jumps switchers, is the lift in hurdle-to-chase (a first try over fences) or chase-to-hurdle (fences taken away), and does a jumping error in the pull-up matter? answered
3. MechanismAfter a pull-up, connections often change the type of race: into a novice or maiden race, or into a handicap off a dropped mark. Any change of conditions makes the last run less relevant, which matches the code-switch story.
4. Why the market misses itIf the generic explanation is right, the market over-weights a P whenever conditions change, not only when the code changes.
5. Predictedpositive, A/E 1.20+ on 250+ bets if the rival holds
6. Falsified ifA/E below 1.10 with the late-period skew of other same-code cuts, which would leave the code switch itself as the carrier
7. Rule sketchlast_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap IS NOT NULL AND is_handicap <> last_is_handicap

Mechanism and rationale

After a pull-up, connections often change the type of race: into a novice or maiden race, or into a handicap off a dropped mark. Any change of conditions makes the last run less relevant, which matches the code-switch story.

The rule

last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap IS NOT NULL AND is_handicap IS NOT NULL AND is_handicap <> last_is_handicap

Reads: last_pulled_up, last_run_style, race_type, last_race_type, last_is_handicap, is_handicap

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

57913a588e2746c954228341bf0bab733ca6fc9596c02b98a3100b40f55d2152

Frozen

not frozen

The definition can still change.

Confounders

• Late-period skew in same-code pull-ups (0.99 before 27 April 2026, 1.45 after): result must be read split by period against the no-status-change control
• Handicap debut marks: first-time handicappers off an untested mark may carry their own lift unrelated to the pull-up
• Price band 16/1 or bigger: outsiders never beat par in this thread, so the mix of long prices can mask or dilute the effect
• Summer jumping race mix: off-season races with small fields and weak opposition skew both handicap/non-handicap transitions and the late period
• Novice/maiden entries after a pull-up may be class drops (weaker opposition), a class effect rather than a conditions-relevance effect
• last_run_style only available from 2025-11-01, so the discovery sample is short and calendar-concentrated

Control

{
  "description": "Same-race other runners are measured by the lab and should sit near par (about 1.0); a lift there would mean the races, not the switchers, carry it. The key comparison is same-code held-up/midfield pull-ups whose handicap status did NOT change: they are known to score 0.99 before 27 April 2026 and 1.45 after. The rival holds only if the status-change group beats that control in BOTH periods (split at 27 April 2026); if it shows the same early-par/late-lift shape, the lift is the calendar skew and the code switch remains the carrier.",
  "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap IS NOT NULL AND is_handicap IS NOT NULL AND is_handicap = last_is_handicap"
}

Discovery result

fail broad rule · 17 Oct, 02:09

Bets

112

12 winners

A/E

1.41

95%: 0.73 to 2.47

SP return

+9.1%

level stakes

Betfair SP return

+73.6%

112 bets with a BSP

Strike rate 10.7%. p = 0.15. Same-race control: the other runners in these races had A/E 0.96, so the selection 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 · 47s · 17 Oct, 02:05

  Output (schema-checked)

  {
    "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap IS NOT NULL AND is_handicap IS NOT NULL AND is_handicap <> last_is_handicap",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap IS NOT NULL AND is_handicap IS NOT NULL AND is_handicap <> last_is_handicap AND sp_decimal < 15"
      },
      {
        "label": "Into a handicap from a non-handicap (debut or dropped mark)",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap = 0 AND is_handicap = 1"
      },
      {
        "label": "Out of a handicap into a non-handicap (novice/maiden/conditions)",
        "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap = 1 AND is_handicap = 0"
      }
    ],
    "control": {
      "description": "Same-race other runners are measured by the lab and should sit near par (about 1.0); a lift there would mean the races, not the switchers, carry it. The key comparison is same-code held-up/midfield pull-ups whose handicap status did NOT change: they are known to score 0.99 before 27 April 2026 and 1.45 after. The rival holds only if the status-change group beats that control in BOTH periods (split at 27 April 2026); if it shows the same early-par/late-lift shape, the lift is the calendar skew and the code switch remains the carrier.",
      "rule": "last_pulled_up = 1 AND last_run_style IN ('held_up', 'midfield') AND race_type = last_race_type AND last_is_handicap IS NOT NULL AND is_handicap IS NOT NULL AND is_handicap = last_is_handicap"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.1,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Late-period skew in same-code pull-ups (0.99 before 27 April 2026, 1.45 after): result must be read split by period against the no-status-change control",
      "Handicap debut marks: first-time handicappers off an untested mark may carry their own lift unrelated to the pull-up",
      "Price band 16/1 or bigger: outsiders never beat par in this thread, so the mix of long prices can mask or dilute the effect",
      "Summer jumping race mix: off-season races with small fields and weak opposition skew both handicap/non-handicap transitions and the late period",
      "Novice/maiden entries after a pull-up may be class drops (weaker opposition), a class effect rather than a conditions-relevance effect",
      "last_run_style only available from 2025-11-01, so the discovery sample is short and calendar-concentrated"
    ],
    "mechanism": "After a held-up or midfield pull-up, connections often switch race conditions within the same code: into a handicap (debut or off a dropped mark) or out of one into novice/maiden/conditions company. If the code-switch lift is really 'new conditions make the P irrelevant', any such change should reproduce it; if only the code switch carries it, a handicap-status change alone should not.",
    "market_miss": "Punters and the market read a recent P as a form negative and mark the horse out, without discounting it for the change of conditions; if the generic explanation is right this over-weighting of the P appears whenever conditions change, not only when the code changes.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.20+ on 250+ bets if the rival holds, with the lift present in both periods around 27 April 2026",
      "falsified_if": "A/E below 1.10, or a lift confined to the period after 27 April 2026 matching the no-status-change control (about 0.99 then 1.45), which would leave the code switch itself as the carrier"
    },
    "notes": "Kept the rule sketch; added is_handicap IS NOT NULL so the inequality cannot silently drop or mishandle nulls. Refinements are the 14/1 cap (outsiders never beat par in this thread) and the two directions of status change the mechanism names. Pass min_ae set to 1.10 to match the falsification threshold. Key read is period-split against the same-code no-change control."
  }

Timeline

1. Job 2412 finished

   17 Oct, 02:09 · engine:compute-1

2. H-d105f080 Discovery → KILLED fails: 112 bets, needs 150; p 0.1511, needs 0.05 or lower; periods [0.82, 2.92], each needs 1.0 (not a niche pass: a period points the other way [0.82, 2.92])

   17 Oct, 02:09 · engine:evaluator

3. discovery result for H-d105f080 (broad rule): fail

   17 Oct, 02:09 · engine:evaluator

4. H-d105f080 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 02:09 · engine:orchestrator

5. compute-1 took job 2412 (evaluate, attempt 1)

   17 Oct, 02:09 · compute-1

6. Job 2409 finished

   17 Oct, 02:05 · engine:ai-1

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

   17 Oct, 02:05 · engine:scheduler

8. H-d105f080 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 02:05 · agent:designer@v1

9. ai-1 took job 2409 (design, attempt 1)

   17 Oct, 02:05 · ai-1

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

    17 Oct, 02:03 · engine:scheduler

11. New idea H-d105f080: Rival: a change of handicap status, not of code, carries the switcher lift (rival-explanation)

    17 Oct, 02:03 · 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.