Raised handicap winners sent off at 9/1+ are over-bet, at any gap since the win

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-7dbfdd08 (Rival: big-rise quick-returning handicap winners stepping up in class ) was predicted negative but came out the other way: A/E 1.16 on 138 bets (p 0.225 for the opposite direction). O-23b5bd4e
   • H-21f9e6e7 (Long-priced raised handicap winners turned out within 14 days are over) pointed the predicted way but failed the bar: A/E 0.57 on 255 bets (p 0.967), control 1.03. O-7c9eecf8
2. AskedIs the near-miss negative on long-priced raised handicap winners (0.57 on 255, killed only on count and p 0.065) a real feature of raised winners the market lets drift, and not something tied to a 14-day turnout?
3. MechanismA handicap winner raised for the win has to run off a higher mark. When the price has still drifted to 9/1+ despite a last-time win, the horse usually won a weak or messy race and the rise has found it out. The win was a one-off on the day, not a step forward. The 14-day limit in H-21f9e6e7 was there only because of the story; the physical cause (a higher mark on a horse that is not improving) does not depend on the gap.
4. Why the market misses itThe form figure '1' and the 'last-time winner' tag still pull in money from the public and from tipsters' systems, so even a drifting winner is backed shorter than its true chance. The rival result (big-rise class-risers at 1.16) shows the market does over-fade the obvious class-up case. That leaves this quieter, long-price case under-faded.
5. Predictednegative, A/E 0.75 to 0.90 on 500+ bets
6. Falsified ifA/E 0.95 or higher on 300+ bets, or the drop exists only within 14 days (over-15-day bets at par).
7. Rule sketchlast_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 1 AND sp_decimal > 9

Mechanism and rationale

A handicap winner raised for the win has to run off a higher mark. When the price has still drifted to 9/1+ despite a last-time win, the horse usually won a weak or messy race and the rise has found it out. The win was a one-off on the day, not a step forward. The 14-day limit in H-21f9e6e7 was there only because of the story; the physical cause (a higher mark on a horse that is not improving) does not depend on the gap.

The rule

last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 1 AND sp_decimal > 9

Reads: last_position, is_handicap, last_is_handicap, rating_change, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

bb42581b2228ff27f3198046d7f06420a88c9389c76f9649b9e164d16876f1e1

Frozen

not frozen

The definition can still change.

Confounders

• Favourite-longshot bias: every 9/1+ runner tends to be over-bet, so the result has to beat the same-race 9/1+ runners and not just par
• Field size: big handicaps produce more long prices and lower strike rates; compare within the same races
• Data window: rating_change only exists from Nov 2025, so the sample is short and seasonal (mostly winter jumps plus all-weather flat)
• Quick turnouts overlap H-21f9e6e7's earlier 14-day pool (0.57 on 255), so the lift may come from bets already seen; the over-14-day refinement checks this
• Mix of flat and jumps: the size of a rise and how much it matters differ by code
• Unraised winners (penalty races, or ones not yet reassessed) are left out by rating_change >= 1; the control group covers them

Control

{
  "description": "Same-race control: the other runners in the same races, priced 9/1+, should sit near the usual longshot level (around 0.85-0.95). The hypothesis needs the raised last-time winners to sit clearly below them, not just at the general longshot level. Second comparison group: last-time handicap winners at 9/1+ that were NOT raised (rating unchanged or lowered). If these score as poorly as the raised group, the effect is about a drifting last-time winner and not about the higher mark, and the mechanism fails.",
  "rule": "last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change <= 0 AND sp_decimal > 9"
}

Discovery result

fail broad rule · 16 Oct, 16:47

Bets

1,561

69 winners

A/E

0.84

95%: 0.66 to 1.07

SP return

-37.3%

level stakes

Betfair SP return

-18.1%

1,552 bets with a BSP

Strike rate 4.4%. p = 0.93. Same-race control: the other runners in these races had A/E 1.02, 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 · 37s · 16 Oct, 16:45

  Output (schema-checked)

  {
    "rule": "last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 1 AND sp_decimal > 9",
    "refinements": [
      {
        "label": "gap over 14 days (tests the falsifier: is the drop not tied to quick turnouts?)",
        "rule": "last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 1 AND sp_decimal > 9 AND days_since_last_run > 14"
      },
      {
        "label": "rise of 3lb or more (a bigger rise should find out a non-improver harder)",
        "rule": "last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 3 AND sp_decimal > 9"
      },
      {
        "label": "9/1 to 20/1 only (removes the extreme-longshot tail where favourite-longshot bias alone could make the negative)",
        "rule": "last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 1 AND sp_decimal > 9 AND sp_decimal <= 21"
      }
    ],
    "control": {
      "description": "Same-race control: the other runners in the same races, priced 9/1+, should sit near the usual longshot level (around 0.85-0.95). The hypothesis needs the raised last-time winners to sit clearly below them, not just at the general longshot level. Second comparison group: last-time handicap winners at 9/1+ that were NOT raised (rating unchanged or lowered). If these score as poorly as the raised group, the effect is about a drifting last-time winner and not about the higher mark, and the mechanism fails.",
      "rule": "last_position = 1 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change <= 0 AND sp_decimal > 9"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Favourite-longshot bias: every 9/1+ runner tends to be over-bet, so the result has to beat the same-race 9/1+ runners and not just par",
      "Field size: big handicaps produce more long prices and lower strike rates; compare within the same races",
      "Data window: rating_change only exists from Nov 2025, so the sample is short and seasonal (mostly winter jumps plus all-weather flat)",
      "Quick turnouts overlap H-21f9e6e7's earlier 14-day pool (0.57 on 255), so the lift may come from bets already seen; the over-14-day refinement checks this",
      "Mix of flat and jumps: the size of a rise and how much it matters differ by code",
      "Unraised winners (penalty races, or ones not yet reassessed) are left out by rating_change >= 1; the control group covers them"
    ],
    "mechanism": "A handicap winner raised for the win now runs off a higher mark. If it still drifts to 9/1+ despite winning last time, the market already rates the win as weak or messy. The rise then catches out a horse that is not improving, so the win was a one-off and not a step forward. The cause is the higher mark on a horse that has not improved, and that does not depend on how long ago it won.",
    "market_miss": "The form figure '1' and the 'last-time winner' tag still pull money from the public and from tipsters' systems. So even a drifting winner is backed shorter than its true chance. The rival result (big-rise class-risers at 1.16) suggests the market over-fades the obvious class-up case and under-fades this quieter long-price case.",
    "prediction": {
      "direction": "negative",
      "size": "A/E 0.75 to 0.90 on 300-500+ bets, clearly below the same-race 9/1+ runners",
      "falsified_if": "A/E 0.95 or higher on 300+ bets; or the drop is only within 14 days (the over-14-day refinement at 0.95+); or unraised 9/1+ handicap winners (the control rule) score as low as the raised group."
    },
    "notes": "The rule sketch is kept: every column is pre-off and none reads the result. rating_change starts in Nov 2025, so discovery is limited to that window and may come in thin. If it does, it goes to falsification and forward tracking, as lab policy says. The direction is negative, so min_ae is read as the lab's mirrored threshold for over-bet rules (A/E at or below 1/1.05)."
  }

Timeline

1. Job 2052 finished

   16 Oct, 16:47 · engine:compute-1

2. H-220ad315 Discovery → KILLED fails: p 0.0837 (lower tail), needs 0.05

   16 Oct, 16:47 · engine:evaluator

3. discovery result for H-220ad315 (broad rule): fail

   16 Oct, 16:47 · engine:evaluator

4. H-220ad315 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 16:47 · engine:orchestrator

5. compute-1 took job 2052 (evaluate, attempt 1)

   16 Oct, 16:47 · compute-1

6. Job 2049 finished

   16 Oct, 16:46 · engine:ai-1

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

   16 Oct, 16:46 · engine:scheduler

8. H-220ad315 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 16:46 · agent:designer@v1

9. ai-1 took job 2049 (design, attempt 1)

   16 Oct, 16:45 · ai-1

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

    16 Oct, 16:45 · engine:scheduler

11. New idea H-220ad315: Raised handicap winners sent off at 9/1+ are over-bet, at any gap since the win (scout)

    16 Oct, 16:45 · agent:scout@v1

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Killed.