Rival: close-beaten drifters pay without the wide tag

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-c869bcd0 (Wide trip and the price drifted since last time: the market marked the) passed discovery: A/E 1.47 on 260 bets (p 0.009), same-race control 0.94. O-8e95e697
   • H-c869bcd0 (Wide trip and the price drifted since last time: the market marked the) survived falsification: 6 of 6 attacks left it standing; it is frozen for its one vault query. O-a6496079
2. AskedIs the drift lift specific to the wide tag, or do all close-beaten runners whose price drifted since last time beat their SP? testing
3. MechanismPunters and layers may overreact to any price drift after a close defeat, reading it as stable confidence gone or a negative signal, whatever the run comment says. Then the wide tag only adds runners to a general drift effect.
4. Why the market misses itA drift reads as negative information, but a horse beaten under five lengths after a clean start has shown ability the drift ignores.
5. Predictedpositive, A/E 1.10 to 1.20 on 1500+ bets at 15.0 SP or shorter
6. Falsified ifA/E 1.03 or lower. The drift edge would then need the wide tag, which supports the reading-the-wide-run mechanism.
7. Rule sketch

Mechanism and rationale

Punters and layers may overreact to any price drift after a close defeat, reading it as stable confidence gone or a negative signal, whatever the run comment says. Then the wide tag only adds runners to a general drift effect.

The rule

last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal * 1.2

Reads: last_raced_wide, last_position, last_beaten_lengths, last_slow_start, sp_decimal, last_sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

7b2e84e37e8512a24afbe84d32e27372cd1480ee0ef84b27048f7400c2a52621

Frozen

not frozen

The definition can still change.

Confounders

• Going change drives many drifts: the wide-tag drift lift fell to 1.245 where going was unchanged (refinement 1 checks this)
• Class rises lengthen prices and may hide better form than the price suggests
• Bigger fields today lengthen prices on their own (refinement 3 checks this)
• Favourite-longshot bias near the 15.0 cap: drifters cluster at longer prices
• Post-April 2026 period lift seen in several overlay subsets; check stability across periods
• Close-beaten form alone may be underpriced whether or not the price drifted (the control checks this)
• last_sp_decimal comes from a different race and field, so a drift partly reflects field strength and not market opinion

Control

{
  "description": "Same-race other runners should be near A/E 1.0 (about 0.95 to 1.0). The key comparison is the same close-beaten, clean-start, not-wide runners at 15.0 or shorter whose price did not drift (shortened or steady). If drifters beat this group, drift itself carries the edge; if both groups score alike, the result just reflects close-beaten form. Compare also against the recorded wide-tag drift subset (1.47 on 260): a non-wide drift A/E well below that means the wide tag is needed.",
  "rule": "last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal <= last_sp_decimal"
}

Discovery result

fail broad rule · 16 Oct, 13:07

Bets

6,259

733 winners

A/E

1.01

95%: 0.94 to 1.08

SP return

-16.8%

level stakes

Betfair SP return

+1.2%

6,222 bets with a BSP

Strike rate 11.7%. p = 0.41. Same-race control: the other runners in these races had A/E 1.00, 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 · 30s · 16 Oct, 12:59

  Output (schema-checked)

  {
    "rule": "last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal * 1.2",
    "refinements": [
      {
        "label": "same going as last time",
        "rule": "last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal * 1.2 AND going_group = last_going_group"
      },
      {
        "label": "drift over 40%",
        "rule": "last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal * 1.4"
      },
      {
        "label": "field no bigger than last time",
        "rule": "last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal > last_sp_decimal * 1.2 AND field_size <= last_field_size"
      }
    ],
    "control": {
      "description": "Same-race other runners should be near A/E 1.0 (about 0.95 to 1.0). The key comparison is the same close-beaten, clean-start, not-wide runners at 15.0 or shorter whose price did not drift (shortened or steady). If drifters beat this group, drift itself carries the edge; if both groups score alike, the result just reflects close-beaten form. Compare also against the recorded wide-tag drift subset (1.47 on 260): a non-wide drift A/E well below that means the wide tag is needed.",
      "rule": "last_raced_wide = 0 AND last_position > 1 AND last_beaten_lengths <= 5 AND last_slow_start = 0 AND sp_decimal <= 15 AND last_sp_decimal IS NOT NULL AND sp_decimal <= last_sp_decimal"
    },
    "pass_criteria": {
      "min_bets": 300,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 150,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Going change drives many drifts: the wide-tag drift lift fell to 1.245 where going was unchanged (refinement 1 checks this)",
      "Class rises lengthen prices and may hide better form than the price suggests",
      "Bigger fields today lengthen prices on their own (refinement 3 checks this)",
      "Favourite-longshot bias near the 15.0 cap: drifters cluster at longer prices",
      "Post-April 2026 period lift seen in several overlay subsets; check stability across periods",
      "Close-beaten form alone may be underpriced whether or not the price drifted (the control checks this)",
      "last_sp_decimal comes from a different race and field, so a drift partly reflects field strength and not market opinion"
    ],
    "mechanism": "After a close defeat (beaten 5 lengths or less, clean start), a price that lengthens by more than 20% may be read by punters and layers as a negative signal: lost stable confidence or bad news. The horse's shown ability has not gone away. If this overreaction applies to any drift, close-beaten drifters with no wide tag should also beat their SP, and the wide tag only adds runners to a general drift effect.",
    "market_miss": "The market reads a drift as information and prices it in again on top of the form. A horse beaten under five lengths after a clean start has recent evidence of ability that the drift ignores, especially when the drift comes from context (going, field, class) and not from the horse.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.10 to 1.20 on 1500+ bets at 15.0 SP or shorter",
      "falsified_if": "A/E 1.03 or lower on the broad rule, or no better than the non-drifting control. Either result means the drift edge needs the wide tag, which supports the reading-the-wide-run mechanism."
    },
    "notes": "The rule sketch is kept, with last_sp_decimal IS NOT NULL made explicit. There is no class-rise refinement because race_class for IRE only starts in 2026-07, so it would quietly drop most Irish runners. The minimum bet count is raised to 300 because the predicted sample is large and the predicted effect is modest."
  }

Timeline

1. Job 1522 finished

   16 Oct, 13:07 · engine:compute-1

2. H-15a788a1 Discovery → KILLED fails: A/E 1.009, needs 1.05; p 0.4107, needs 0.05 or lower; periods [0.97, 1.03], each needs 1.0; A/E 1.009 against same-race control 0.999, needs 0.03 above (not a niche pass: A/E 1.009 below 1.05; does not beat the same-race control 0.999)

   16 Oct, 13:07 · engine:evaluator

3. discovery result for H-15a788a1 (broad rule): fail

   16 Oct, 13:07 · engine:evaluator

4. H-15a788a1 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 13:07 · engine:orchestrator

5. compute-1 took job 1522 (evaluate, attempt 1)

   16 Oct, 13:07 · compute-1

6. Job 1519 finished

   16 Oct, 12:59 · engine:ai-1

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

   16 Oct, 12:59 · engine:scheduler

8. H-15a788a1 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 12:59 · agent:designer@v1

9. ai-1 took job 1519 (design, attempt 1)

   16 Oct, 12:59 · ai-1

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

    16 Oct, 12:59 · engine:scheduler

11. New idea H-15a788a1: Rival: close-beaten drifters pay without the wide tag (rival-explanation)

    16 Oct, 12:59 · 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.