Sharpen: easy handicap winners back within 14 days after a small rise (1 to 5 lb)

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-528222df (Raised easy handicap winners turned out within 14 days beat their pric) passed discovery: A/E 1.33 on 231 bets (p 0.013), same-race control 0.91. O-991dc8ad
   • H-a4582701 (Rival: any raised handicap winner turned out within 14 days beats its ) pointed the predicted way but failed the bar: A/E 1.03 on 1050 bets (p 0.323), control 0.99. O-ec0a35a0
2. AskedWithin raised easy handicap winners back inside 14 days, is the lift largest when the rise is small (1 to 5 lb), as 'the handicapper can only react to the margin' predicts? answered
3. MechanismThe handicapper sizes the rise to the beaten margin. An easy win by a short margin gets a small rise that leaves the horse well-in against its real ability. A big rise means the handicapper has already looked past the margin.
4. Why the market misses itThe market reads 'won easily' and 'raised' as cancelling out. It does not tell a token rise from a full reassessment. Raised non-easy winners at 1.034 show the market prices the rise itself about right.
5. Predictedpositive, A/E 1.40 to 1.60 on 120+ bets; 6 lb+ rises near 1.1 or below
6. Falsified ifA/E at or below 1.334, or below 1 in either period. That would mean the size of the rise does not matter and the lift is not about the mark lagging the horse.
7. Rule sketchlast_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change BETWEEN 1 AND 5

The question's answer so far: No. Small rises of 1 to 5 lb scored 1.12 on 100, below the raised parent's 1.334 on 231, and faded to 0.91 after April 2026. On an approximate split, rises of 6 lb or more carry the lift at roughly 1.5 on about 131 bets.

Mechanism and rationale

The handicapper sizes the rise to the beaten margin. An easy win by a short margin gets a small rise that leaves the horse well-in against its real ability. A big rise means the handicapper has already looked past the margin.

The rule

last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change BETWEEN 1 AND 5

Reads: last_position, last_won_easily, days_since_last_run, is_handicap, last_is_handicap, rating_change

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

47fdcb8e8df0b862dde28ecb619094332bd4e78b8db967d65e1aa1962aed5036

Frozen

not frozen

The definition can still change.

Confounders

• Race class and rating band: smaller rises go with lower-rated, lower-class handicaps, which may have their own pricing biases. Check class mix against the 6 lb+ control.
• Flat and jumps rating scales differ (a 5 lb rise means different things), so the 1 to 5 lb band selects a different code mix from the 6 lb+ group.
• Penalty against new mark: horses whose new mark is not yet in force carry a penalty and show rating_change 0, so they fall outside this rule. The rule therefore tests only horses running off a new mark already in force, and that is not the same population as the parent.
• Sample size and window: rating_change and last_won_easily only start in Nov 2025, so the slice will be thin (about 120 expected, under the 150 bar), and a parent-period effect may simply carry over.
• Selection from the parent: this is a sharpening of a 1.334 result found on the same data, so some of any lift is the parent's luck rather than a new effect.
• GB dominance: almost all the parent's bets are GB, so any result says little about IRE.

Control

Discovery result

fail broad rule · 16 Oct, 16:26

Bets

100

22 winners

A/E

1.12

95%: 0.70 to 1.69

SP return

-4.1%

level stakes

Betfair SP return

+14.1%

99 bets with a BSP

Strike rate 22.0%. p = 0.33. Same-race control: the other runners in these races had A/E 0.97, 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 · 45s · 16 Oct, 16:24

  Output (schema-checked)

  {
    "rule": "last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change BETWEEN 1 AND 5",
    "refinements": [
      {
        "label": "14/1 or shorter",
        "rule": "last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change BETWEEN 1 AND 5 AND sp_decimal IS NOT NULL AND sp_decimal <= 15.0"
      },
      {
        "label": "7/1 or shorter, where the parent lift sat",
        "rule": "last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change BETWEEN 1 AND 5 AND sp_decimal IS NOT NULL AND sp_decimal <= 8.0"
      }
    ],
    "control": {
      "description": "Two controls. (1) The other runners in the same races should come out near the parent's same-race control (about 0.90 to 0.95), which shows the lift is not a property of these races. (2) The mechanism needs a dose comparison: the same easy handicap winners back within 14 days but raised 6 lb or more. If the lift comes from the mark lagging the horse, this group should score about 1.1 or below and clearly under the 1 to 5 lb group. If it scores as well as or better than the small-rise group, the size of the rise does not matter and the mechanism fails.",
      "rule": "last_position = 1 AND last_won_easily = 1 AND days_since_last_run IS NOT NULL AND days_since_last_run <= 14 AND is_handicap = 1 AND last_is_handicap = 1 AND rating_change IS NOT NULL AND rating_change >= 6"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.05,
      "max_p": 0.2
    },
    "confounders": [
      "Race class and rating band: smaller rises go with lower-rated, lower-class handicaps, which may have their own pricing biases. Check class mix against the 6 lb+ control.",
      "Flat and jumps rating scales differ (a 5 lb rise means different things), so the 1 to 5 lb band selects a different code mix from the 6 lb+ group.",
      "Penalty against new mark: horses whose new mark is not yet in force carry a penalty and show rating_change 0, so they fall outside this rule. The rule therefore tests only horses running off a new mark already in force, and that is not the same population as the parent.",
      "Sample size and window: rating_change and last_won_easily only start in Nov 2025, so the slice will be thin (about 120 expected, under the 150 bar), and a parent-period effect may simply carry over.",
      "Selection from the parent: this is a sharpening of a 1.334 result found on the same data, so some of any lift is the parent's luck rather than a new effect.",
      "GB dominance: almost all the parent's bets are GB, so any result says little about IRE."
    ],
    "mechanism": "The handicapper sets the rise from the visible winning margin. An easy win by a short distance gets a small rise of 1 to 5 lb, which leaves the horse well-in against what it could have shown under pressure. Back within 14 days, before the form is exposed or a further reassessment, it runs off a mark that still lags its ability. A rise of 6 lb or more means the handicapper has already looked past the margin and closed the gap.",
    "market_miss": "Punters and the market read 'won easily' and 'raised' as roughly cancelling out, and price the rise by the headline (raised, so less well-in) without telling a token rise from a full reassessment. Raised non-easy winners at 1.034 on 1,050 show the market prices a rise correctly when there is no hidden margin. The mispricing is the hidden margin that a small rise leaves in place.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.40 to 1.60 on about 120+ bets; the 6 lb+ control near 1.1 or below",
      "falsified_if": "A/E at or below 1.334 (no better than the unsliced raised parent), or below 1.0 in either period, or the 6 lb+ control scoring at or above the 1 to 5 lb group. Any of these means the size of the rise does not matter."
    },
    "notes": "The rule sketch is kept unchanged. All columns are pre_off. rating_change, last_won_easily and last_official_rating are only available from 2025-11-01 (GB and IRE), so the testable window is short and the slice will probably fall under the 150-bet discovery floor. In that case it should go on to falsification and forward tracking, not be killed for size. The forward bar is raised to A/E 1.05 because the claim is a sharpening that must beat its parent, not merely break even. Refinements follow the finding that long-priced raised easy winners do not convert (0 from 17 at 16/1+) and that the parent's lift sat at up to 7/1."
  }

Timeline

1. Job 1988 finished

   16 Oct, 16:26 · engine:compute-1

2. H-a81fea8a Discovery → KILLED fails: 100 bets, needs 150; p 0.329, needs 0.05 or lower; periods [1.33, 0.91], each needs 1.0 (not a niche pass: a period points the other way [1.33, 0.91])

   16 Oct, 16:26 · engine:evaluator

3. discovery result for H-a81fea8a (broad rule): fail

   16 Oct, 16:26 · engine:evaluator

4. H-a81fea8a Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 16:26 · engine:orchestrator

5. compute-1 took job 1988 (evaluate, attempt 1)

   16 Oct, 16:26 · compute-1

6. Job 1986 finished

   16 Oct, 16:25 · engine:ai-1

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

   16 Oct, 16:25 · engine:scheduler

8. H-a81fea8a Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 16:25 · agent:designer@v1

9. ai-1 took job 1986 (design, attempt 1)

   16 Oct, 16:24 · ai-1

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

    16 Oct, 16:23 · engine:scheduler

11. New idea H-a81fea8a: Sharpen: easy handicap winners back within 14 days after a small rise (1 to 5 lb) (sharpen)

    16 Oct, 16:23 · 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.