Dose-response: a big trip drop (660+ yards) into a stiff Flat sprint pays more

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-2b483ac5 (Generalise: last run over further (not just last win) pays in stiff Fl) met the sealed vault supportive: A/E 1.17 on 415 bets (p 0.1238, threshold 0.0028). O-7f1cba89
   • H-cbd60237 (Generalise: any proven stamina (last run or last win 330+ yards furthe) passed discovery as a niche: A/E 1.54 on 100 bets (p 0.047), same-race control 0.91. O-c404ad04
   • H-2b483ac5 (Generalise: last run over further (not just last win) pays in stiff Fl) survived falsification: 4 of 5 attacks left it standing; it is frozen for its one vault query. O-fdffc600
2. AskedDoes the stiff-sprint edge grow with the size of the trip drop (a 660+ yard drop beating a 330 to 659 yard drop), as the stamina-becomes-speed mechanism needs? testing
3. MechanismAn uphill finish turns a sprint into a stamina test over the last furlong. A horse that last ran over a mile or more brings more stamina than one that ran over seven furlongs, so a bigger drop should give a bigger late-race edge over the speed horses that tire up the hill.
4. Why the market misses itA big drop in trip reads as wrong-trip form: the last run was slow early and beaten over a trip the punter sees as irrelevant, so the horse is priced as lacking sprint speed. That ignores the fact that the stiff finish rewards what the horse showed.
5. Predictedpositive, A/E 1.20 to 1.40 on 150+ bets, above the 1.17 of the 330+ yard rule

Mechanism and rationale

An uphill finish turns a sprint into a stamina test over the last furlong. A horse that last ran over a mile or more brings more stamina than one that ran over seven furlongs, so a bigger drop should give a bigger late-race edge over the speed horses that tire up the hill.

The rule

race_type = 'Flat' AND distance_yards <= 1340 AND track_stiff_finish = 1 AND last_distance_yards >= distance_yards + 660 AND sp_decimal <= 15

Reads: race_type, distance_yards, track_stiff_finish, last_distance_yards, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

bb2746ad14f5a167ad97368a238385db7bef842a52abf2b17b4ace642316a497

Frozen

not frozen

The definition can still change.

Confounders

• Drop size goes with age and class: horses dropped from a mile to a sprint are often older, exposed handicappers, so a difference between drop bands may reflect horse type, not stamina
• Course mix: there are few stiff GB sprint courses, and big drops may cluster at particular courses (and draw positions), so a course quirk could pose as a dose effect
• Region: the parent edge is GB only (IRE about 0.74 to 0.85), and a different GB/IRE mix between bands would distort the comparison
• Thin sample: the 330+ parent had only 83 discovery bets at 14/1 or shorter, so the 660+ subset will probably miss 150 bets and only the vault and forward record can judge it
• Overlap with the parent: every 660+ bet is also in the 330+ rule, so the 330-659 control band, not the parent figure, is the clean dose comparison
• Going and season: big trip drops may cluster in the late season on soft ground, which adds a separate stamina demand

Control

{
  "description": "Two comparisons. (1) Same-race rivals: the other runners in these races should score about 0.95 to 1.00, so any lift is not a property of the races. (2) Dose-response control: the smaller-drop band (330 to 659 yards further last time) at the same stiff Flat sprints and the same 14/1 cap. The hypothesis requires the 660+ group to beat this band; if the 330-659 band scores the same or higher, drop size does not matter and the lift is the general trip-drop effect.",
  "rule": "race_type = 'Flat' AND distance_yards <= 1340 AND track_stiff_finish = 1 AND last_distance_yards >= distance_yards + 330 AND last_distance_yards < distance_yards + 660 AND sp_decimal <= 15"
}

Discovery result

fail broad rule · 16 Oct, 13:22

Bets

2

1 winners

A/E

5.05

95%: 0.13 to 28.14

SP return

+250.0%

level stakes

Betfair SP return

–

2 bets with a BSP

Strike rate 50.0%. p = 0.18. Same-race control: the other runners in these races had A/E 0.56, 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 · 31s · 16 Oct, 13:19

  Output (schema-checked)

  {
    "rule": "race_type = 'Flat' AND distance_yards <= 1340 AND track_stiff_finish = 1 AND last_distance_yards >= distance_yards + 660 AND sp_decimal <= 15",
    "refinements": [
      {
        "label": "GB only (Irish null seen twice in this family)",
        "rule": "race_type = 'Flat' AND distance_yards <= 1340 AND track_stiff_finish = 1 AND last_distance_yards >= distance_yards + 660 AND sp_decimal <= 15 AND region = 'GB'"
      },
      {
        "label": "Last run at a mile or more (the mechanism's own stamina threshold)",
        "rule": "race_type = 'Flat' AND distance_yards <= 1340 AND track_stiff_finish = 1 AND last_distance_yards >= distance_yards + 660 AND last_distance_yards >= 1760 AND sp_decimal <= 15"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) Same-race rivals: the other runners in these races should score about 0.95 to 1.00, so any lift is not a property of the races. (2) Dose-response control: the smaller-drop band (330 to 659 yards further last time) at the same stiff Flat sprints and the same 14/1 cap. The hypothesis requires the 660+ group to beat this band; if the 330-659 band scores the same or higher, drop size does not matter and the lift is the general trip-drop effect.",
      "rule": "race_type = 'Flat' AND distance_yards <= 1340 AND track_stiff_finish = 1 AND last_distance_yards >= distance_yards + 330 AND last_distance_yards < distance_yards + 660 AND sp_decimal <= 15"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.1,
      "max_p": 0.2
    },
    "confounders": [
      "Drop size goes with age and class: horses dropped from a mile to a sprint are often older, exposed handicappers, so a difference between drop bands may reflect horse type, not stamina",
      "Course mix: there are few stiff GB sprint courses, and big drops may cluster at particular courses (and draw positions), so a course quirk could pose as a dose effect",
      "Region: the parent edge is GB only (IRE about 0.74 to 0.85), and a different GB/IRE mix between bands would distort the comparison",
      "Thin sample: the 330+ parent had only 83 discovery bets at 14/1 or shorter, so the 660+ subset will probably miss 150 bets and only the vault and forward record can judge it",
      "Overlap with the parent: every 660+ bet is also in the 330+ rule, so the 330-659 control band, not the parent figure, is the clean dose comparison",
      "Going and season: big trip drops may cluster in the late season on soft ground, which adds a separate stamina demand"
    ],
    "mechanism": "An uphill finish turns the last furlong of a Flat sprint into a stamina test. A horse whose last run was 660+ yards further (in effect a mile or more against a 5-6f sprint) brings more stamina than one dropped from seven furlongs. It should therefore keep going up the hill while the speed horses tire, so the edge should rise with the size of the drop. That is a dose-response on the parent stiff-sprint trip-drop effect (vault 1.17 on 415).",
    "market_miss": "Punters read a big drop in trip as wrong-trip form. The last run was slow early and often beaten, over a trip they see as irrelevant to a sprint, so the horse is priced as lacking speed. The price ignores that a stiff finish rewards the stamina the horse showed, and the bigger the drop, the more irrelevant the form looks and the bigger the discount.",
    "prediction": {
      "direction": "positive",
      "size": "A/E about 1.20 to 1.40 at 14/1 or shorter, higher than the 330+ rule's vault 1.17 and clearly higher than the 330-659 yard control band",
      "falsified_if": "A/E at or below 1.10, or no higher than the 330+ rule's vault figure of 1.17, or no higher than the 330-659 yard control band. Any of these would mean drop size does not matter and the lift is general trip-drop drift."
    },
    "notes": "I kept the rule sketch as written: it uses only pre-off columns and reads nothing from the result. The 14/1 cap carries over from the parent because the edge is carried by shorter-priced runners. I expect fewer than 150 discovery bets, so the main evidence will be the dose-response gap against the 330-659 band, checked in the vault and forward. The forward A/E bar is set at 1.10, above the floor, because the hypothesis claims more than the parent's modest edge."
  }

Timeline

1. Job 1583 finished

   16 Oct, 13:22 · engine:compute-1

2. H-717f83d2 Discovery → KILLED fails: 2 bets, needs 150; p 0.1796, needs 0.05 or lower; periods [], each needs 1.0 (not a niche pass: only 2 bets)

   16 Oct, 13:22 · engine:evaluator

3. discovery result for H-717f83d2 (broad rule): fail

   16 Oct, 13:22 · engine:evaluator

4. H-717f83d2 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 13:22 · engine:orchestrator

5. compute-1 took job 1583 (evaluate, attempt 1)

   16 Oct, 13:22 · compute-1

6. Job 1581 finished

   16 Oct, 13:20 · engine:ai-1

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

   16 Oct, 13:20 · engine:scheduler

8. H-717f83d2 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 13:20 · agent:designer@v1

9. ai-1 took job 1581 (design, attempt 1)

   16 Oct, 13:19 · ai-1

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

    16 Oct, 13:19 · engine:scheduler

11. New idea H-717f83d2: Dose-response: a big trip drop (660+ yards) into a stiff Flat sprint pays more (test-mechanism)

    16 Oct, 13:19 · 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.