Hampered at the start (not its own slow start), beaten last time

Mechanism and rationale

A horse that was barged, squeezed out or hampered at the start lost position and spent energy for a reason outside itself. That is different from a slow starter, which keeps doing it (and slow starts are already a published negative). The bump is a one-off physical event, so the horse's ability is unchanged and the next run should show its normal level.

The rule

last_trouble_at_start = 1 AND coalesce(last_slow_start, 0) = 0 AND last_position > 1

Reads: last_trouble_at_start, last_slow_start, last_position

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

3632b9fbf18d7d9790dd7ddb3d1f14f1cb8faed230cfb22ed35aceaa3a8df389

Frozen

not frozen

The definition can still change.

Confounders

• Tag misreadings: 'jumped awkwardly from stalls', 'dwelt' or 'reared as stalls opened' are the horse's own fault but may be tagged as trouble at start. Read 30+ samples of last_phrases by eye before trusting any result.
• Field size and draw: start trouble is more common in big Flat fields and from wide or middle draws. Use the field-size-corrected yardstick and the same-race control.
• Overlap with the existing momentum-interrupted, early-setback and wide-trip systems. Split the result by last_momentum_interrupted, last_early_setback and last_raced_wide to check that the start bump adds something of its own.
• Career-count effects: lightly raced two-year-olds and debutants make up much of the start trouble. Split by prior_runs (<=3 vs more) and by age.
• Short history: the remark tags only start on 2025-11-01, so discovery covers about 11 months. Split into two halves, which must agree. A thin result goes forward, not to promotion.
• Jumps 'start trouble' (e.g. a tape start, a horse whipping round) is a different event from barging in the stalls. Split by race_type.

Discovery result

fail broad rule · 15 Oct, 23:22

Bets

545

54 winners

A/E

0.99

95%: 0.74 to 1.29

SP return

-33.0%

level stakes

Betfair SP return

-17.7%

541 bets with a BSP

Strike rate 9.9%. p = 0.55. Same-race control: the other runners in these races had A/E 1.00, so the selection did not beat its own races. Zone: discovery. Evaluator eval-1.0, data ef1163efa87ef7d2

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

No parent and no descendants yet. When an agent mutates, generalises or splits this idea, the new hypotheses appear here.

Agent runs

• designer v1ok · 47s · 15 Oct, 23:22

  Output (schema-checked)

  {
    "rule": "last_trouble_at_start = 1 AND coalesce(last_slow_start, 0) = 0 AND last_position > 1",
    "refinements": [
      {
        "label": "14/1 or shorter today",
        "rule": "last_trouble_at_start = 1 AND coalesce(last_slow_start, 0) = 0 AND last_position > 1 AND sp_decimal <= 15"
      },
      {
        "label": "Fancied last time (15.0 or shorter at last SP)",
        "rule": "last_trouble_at_start = 1 AND coalesce(last_slow_start, 0) = 0 AND last_position > 1 AND last_sp_decimal <= 15"
      },
      {
        "label": "Flat only (stalls starts, where barging is the usual start trouble)",
        "rule": "last_trouble_at_start = 1 AND coalesce(last_slow_start, 0) = 0 AND last_position > 1 AND race_type = 'Flat'"
      }
    ],
    "control": {
      "description": "The lab's automatic control is the other runners in the same races. They should come out near the usual 1.03-1.05 control level. If the selections are no better than that, the effect is a property of the race, not of the bump. A second control separates 'trouble caused by another horse' from 'its own slow start': horses beaten last time after a slow start of their own, with no start trouble tagged. The mechanism says this group should be at or below 1.00 (a published negative), so the hypothesis needs a clear gap between the two groups, not just a positive for the broad rule.",
      "rule": "last_slow_start = 1 AND coalesce(last_trouble_at_start, 0) = 0 AND last_position > 1"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.08,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.03,
      "max_p": 0.2
    },
    "confounders": [
      "Tag misreadings: 'jumped awkwardly from stalls', 'dwelt' or 'reared as stalls opened' are the horse's own fault but may be tagged as trouble at start. Read 30+ samples of last_phrases by eye before trusting any result.",
      "Field size and draw: start trouble is more common in big Flat fields and from wide or middle draws. Use the field-size-corrected yardstick and the same-race control.",
      "Overlap with the existing momentum-interrupted, early-setback and wide-trip systems. Split the result by last_momentum_interrupted, last_early_setback and last_raced_wide to check that the start bump adds something of its own.",
      "Career-count effects: lightly raced two-year-olds and debutants make up much of the start trouble. Split by prior_runs (<=3 vs more) and by age.",
      "Short history: the remark tags only start on 2025-11-01, so discovery covers about 11 months. Split into two halves, which must agree. A thin result goes forward, not to promotion.",
      "Jumps 'start trouble' (e.g. a tape start, a horse whipping round) is a different event from barging in the stalls. Split by race_type."
    ],
    "mechanism": "A horse barged, squeezed out or hampered at the start loses position and uses energy because of another runner. That is a one-off physical event outside the horse. Its ability is unchanged, so the next run should return to its normal level. A slow starter is different: it caused its own problem and tends to repeat it, which is why slow starts are a published negative. The rule therefore keeps external start trouble and drops the horse's own slow start.",
    "market_miss": "The form figure, the beaten distance and the ratings all record a modest run. Punters and compilers also tend to lump 'hampered start' in with 'slowly away', and the market rightly marks slow starters down. So the price drifts as if the horse simply was not good enough, even though the cause was another runner and is unlikely to happen again.",
    "notes": "Every remark column (last_trouble_at_start, last_slow_start, etc.) is only available from 2025-11-01. Discovery is therefore about 11 months: use the two-halves period split, not the three-period split. last_position > 1 drops non-finishers (NULL), so horses brought down or unseated at the start are left out on purpose; that is a different event. The refinements are only looked at if the broad rule shows evidence. Pass A/E is raised to 1.08 so the result must clearly beat the 1.03-1.05 control level."
  }

Timeline

1. Job 28 finished

   15 Oct, 23:22 · engine:compute-1

2. H-97366005 Discovery → KILLED fails: A/E 0.99, needs 1.08; p 0.5476, needs 0.05 or lower; periods [0.86, 1.06], each needs 1.0; A/E 0.99 against same-race control 0.998, needs 0.03 above (not a niche pass: A/E 0.99 below 1.08; p 0.5475723737607818 above 0.5; does not beat the same-race control 0.998; a period points the other way [0.86, 1.06])

   15 Oct, 23:22 · engine:evaluator

3. discovery result for H-97366005 (broad rule): fail

   15 Oct, 23:22 · engine:evaluator

4. H-97366005 Designed → RESEARCHdiscovery measuring on the discovery zone

   15 Oct, 23:22 · engine:orchestrator

5. compute-1 took job 28 (evaluate, attempt 1)

   15 Oct, 23:22 · compute-1

6. Job 24 finished

   15 Oct, 23:22 · engine:ai-1

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

   15 Oct, 23:22 · engine:scheduler

8. H-97366005 Proposed → RESEARCHdesigned pre-spec written and hashed

   15 Oct, 23:22 · agent:designer@v1

9. ai-1 took job 24 (design, attempt 1)

   15 Oct, 23:22 · ai-1

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

    15 Oct, 23:22 · engine:scheduler

11. New idea H-97366005: Hampered at the start (not its own slow start), beaten last time (scout)

    15 Oct, 23:22 · 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.