Rival: dropping in trip after a win over further pays at any track, stiff or not

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
   • H181-proven-over-further-at-a-stiff-track (H181: proven over further at a stiff track) was GREEN in the earlier research: A/E 1.08 on 9279 bets (periods 1.06 / 1.07 / 1.12); now PROMISING. O-7cdd45d9
2. AskedDoes a recent winner over at least 330 yards further beat its price at non-stiff tracks too, i.e. is the stiff finish needed at all? answered
3. MechanismPunters may mark down any horse dropping well below its winning trip as 'too slow', whatever the track. If so, the edge comes from the trip drop, not the hill.
4. Why the market misses itA drop in trip reads as a trip mismatch in the form book; the win over further is discounted as irrelevant to today.
5. Predictedpositive, A/E 1.05 to 1.08 on 15000+ bets at non-stiff tracks
6. Falsified ifA/E at or below 1.01 at non-stiff tracks, which would support the stiff-finish mechanism
7. Rule sketchtrack_stiff_finish = 0 AND last_win_distance_yards >= distance_yards + 330
8. question id

Mechanism and rationale

Punters may mark down any horse dropping well below its winning trip as 'too slow', whatever the track. If so, the edge comes from the trip drop, not the hill.

The rule

track_stiff_finish = 0 AND last_win_yards >= distance_yards + 330

Reads: track_stiff_finish, last_win_yards, distance_yards

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

ae08b47c1293bb3afca9817fa91e9d1f7acc20b9456dd40166b5d45b31f527ba

Frozen

not frozen

The definition can still change.

Confounders

• Sharp and speed tracks favour pace and actively penalise stayers, which could drag the non-stiff A/E down for reasons unrelated to market pricing (tested by the not-sharp refinement)
• The mix of codes differs: stiff tracks carry more jumps racing, where drops in trip behave differently from the Flat, so a code split is needed before comparing to the stiff-track parent
• Field size and race class differ between track types and change baseline A/E
• A longshot artefact: lift concentrated at 16/1+ on few winners would not be a credible mispricing (check the 14/1 band)
• Period and surface: the post-2025 window mixes all-weather winter with turf summer, so a period concentration needs a surface check
• track_stiff_finish NULLs are excluded by '= 0', so courses without a guide entry drop out of both halves

Control

{
  "description": "The lab's same-race comparison: other runners in the same non-stiff races should sit near A/E 1.00, so any lift is the horse, not the race. The decisive comparison is the parent rule on stiff tracks over the same window (H181 was A/E 1.08 on 9,279). If the non-stiff rule matches it (around 1.05 or more), the trip drop carries the edge on its own. If it sits at 1.01 or below while the stiff version holds, the hill is needed. Both rules use last_win_yards, so the two halves split the same population cleanly.",
  "rule": "track_stiff_finish = 1 AND last_win_yards >= distance_yards + 330"
}

Discovery result

fail broad rule · 16 Oct, 12:43

Bets

4,671

549 winners

A/E

1.05

95%: 0.96 to 1.14

SP return

-21.6%

level stakes

Betfair SP return

+2.9%

4,498 bets with a BSP

Strike rate 11.8%. p = 0.14. Same-race control: the other runners in these races had A/E 0.99, 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 · 48s · 16 Oct, 12:41

  Output (schema-checked)

  {
    "rule": "track_stiff_finish = 0 AND last_win_yards >= distance_yards + 330",
    "refinements": [
      {
        "label": "14/1 or shorter: a real trip-drop mispricing should show in credible bands, not 16/1+ artefacts",
        "rule": "track_stiff_finish = 0 AND last_win_yards >= distance_yards + 330 AND sp_decimal <= 15"
      },
      {
        "label": "non-stiff, not sharp or speed track: removes the tracks that actively penalise stayers (named confounder)",
        "rule": "track_stiff_finish = 0 AND last_win_yards >= distance_yards + 330 AND track_sharpness < 2 AND track_speed_track = 0"
      },
      {
        "label": "bigger drop (2f+): dose-response, since a larger trip mismatch should be discounted more",
        "rule": "track_stiff_finish = 0 AND last_win_yards >= distance_yards + 440"
      }
    ],
    "control": {
      "description": "The lab's same-race comparison: other runners in the same non-stiff races should sit near A/E 1.00, so any lift is the horse, not the race. The decisive comparison is the parent rule on stiff tracks over the same window (H181 was A/E 1.08 on 9,279). If the non-stiff rule matches it (around 1.05 or more), the trip drop carries the edge on its own. If it sits at 1.01 or below while the stiff version holds, the hill is needed. Both rules use last_win_yards, so the two halves split the same population cleanly.",
      "rule": "track_stiff_finish = 1 AND last_win_yards >= distance_yards + 330"
    },
    "pass_criteria": {
      "min_bets": 1000,
      "min_ae": 1.05,
      "max_p": 0.01
    },
    "forward_bar": {
      "min_bets": 200,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Sharp and speed tracks favour pace and actively penalise stayers, which could drag the non-stiff A/E down for reasons unrelated to market pricing (tested by the not-sharp refinement)",
      "The mix of codes differs: stiff tracks carry more jumps racing, where drops in trip behave differently from the Flat, so a code split is needed before comparing to the stiff-track parent",
      "Field size and race class differ between track types and change baseline A/E",
      "A longshot artefact: lift concentrated at 16/1+ on few winners would not be a credible mispricing (check the 14/1 band)",
      "Period and surface: the post-2025 window mixes all-weather winter with turf summer, so a period concentration needs a surface check",
      "track_stiff_finish NULLs are excluded by '= 0', so courses without a guide entry drop out of both halves"
    ],
    "mechanism": "Punters may mark down any horse dropping well below its winning trip (1.5f+ shorter than its last win) as 'too slow', whatever the track. If so, the edge comes from the trip drop itself and not from a stiff finish that brings stamina back into play. The question this answers: is the hill needed, or does the trip drop alone carry the H181 edge?",
    "market_miss": "In the form book a drop in trip reads as a trip mismatch, so the market discounts the win over further as irrelevant to today. If the market fully prices the speed shortfall at non-stiff tracks, there is no edge, and the H181 lift is about stiff finishes.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.05 to 1.08 on 15,000+ bets at non-stiff tracks, about level with the stiff-track parent (1.08)",
      "falsified_if": "A/E at or below 1.01 at non-stiff tracks (or the lift only appears at 16/1+) while the stiff-track control holds near 1.08. That result would support the stiff-finish mechanism over the trip-drop-alone rival."
    },
    "notes": "I changed last_win_distance_yards in the sketch to last_win_yards so the rule is the exact non-stiff complement of the parent H181. A lab finding flags that the two columns are not yet confirmed identical, and the clean split depends on using the same one. Both columns are available from 2020-09 in GB and Ireland."
  }

Timeline

1. Job 1421 finished

   16 Oct, 12:43 · engine:compute-1

2. H-9be58fc5 Discovery → KILLED fails: A/E 1.048, needs 1.05; p 0.1406, needs 0.01 or lower; periods [0.98, 1.14], each needs 1.0 (not a niche pass: A/E 1.048 below 1.05)

   16 Oct, 12:43 · engine:evaluator

3. discovery result for H-9be58fc5 (broad rule): fail

   16 Oct, 12:43 · engine:evaluator

4. H-9be58fc5 Designed → RESEARCHdiscovery measuring on the discovery zone

   16 Oct, 12:43 · engine:orchestrator

5. compute-1 took job 1421 (evaluate, attempt 1)

   16 Oct, 12:43 · compute-1

6. Job 1418 finished

   16 Oct, 12:42 · engine:ai-1

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

   16 Oct, 12:42 · engine:scheduler

8. H-9be58fc5 Proposed → RESEARCHdesigned pre-spec written and hashed

   16 Oct, 12:42 · agent:designer@v1

9. ai-1 took job 1418 (design, attempt 1)

   16 Oct, 12:41 · ai-1

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

    16 Oct, 12:41 · engine:scheduler

11. New idea H-9be58fc5: Rival: dropping in trip after a win over further pays at any track, stiff or not (rival-explanation)

    16 Oct, 12:41 · 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.