H85: fell when going well

Mechanism and rationale

Fell or unseated last time while on the front end or travelling well, before coming under pressure.

The rule

lt_fell_going_well = 1

Reads: lt_fell_going_well

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

not designed yet

Frozen

not frozen

The definition can still change.

Broad

lt_fell_going_well = 1

Kind

remarks

Origin

research/experiments/overnight/results_v2.jsonl

Refinements

[
  {
    "name": "no repeat jumping errors",
    "rule": "(lt_fell_going_well = 1) AND (jumping_errors_last3 <= 1)"
  },
  {
    "name": "14/1 or shorter",
    "rule": "(lt_fell_going_well = 1) AND (sp_decimal <= 15)"
  },
  {
    "name": "chase",
    "rule": "(lt_fell_going_well = 1) AND (race_type = 'Chase')"
  }
]

Tier

B

Result (from the archive)

info chase · 15 Oct, 23:10

Bets

165

A/E

1.30

winners vs the prices

SP return

+45.8%

level stakes

Betfair SP return

+116.0%

164 bets with a BSP

Strike rate 17.0%.

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

• investigator v1ok · 47s · 17 Oct, 02:06

  Output (schema-checked)

  {
    "reflection": "This thread has one claim and no tests of its own yet. Earlier research found that horses which fell or unseated last time while going well beat their prices: A/E 1.13 on 405 bets. Both periods held up (1.15 on 255, 1.11 on 150). It made money at BSP (+36.5%) but lost slightly at SP (-3.0%). The lift sits in chases: A/E 1.297 on 165 bets, which became the live system SYS-fell-when-going-well-chase. Taking the chase bets out, the other 240 bets come to roughly par (about 1.0), so the effect looks specific to chases. That fits the mechanism: a chaser's F or U is read as failure when it was really an unfinished good run. Three things are untested. First, whether 'going well' matters at all, or whether any chase faller bounces back. Second, whether the effect is GB-led like the nearby chase-error thread (IRE non-finishers there scored 0.43 on 23). Third, whether habitual fallers dilute it. Until forward bets arrive, the 165-bet chase figure is one discovery result.",
    "answers": [],
    "questions": [
      {
        "question": "Does the chase lift need the horse to have been going well when it fell or unseated, or do all last-time chase fallers and unseaters beat their price?",
        "rationale": "If chasers that fell or unseated while NOT going well show the same lift, the mechanism is a general bounce-back after an F or U (the market over-punishing the letter), not a hidden good run. Columns: last_fell_or_unseated, last_fell_going_well, race_type, last_race_type.",
        "observation_ids": [
          "O-add18290",
          "O-1f8ceedf"
        ],
        "priority": 0.75
      },
      {
        "question": "Is the fell-when-going-well chase edge GB-led, as the chase-error edge is?",
        "rationale": "In the chase-error thread, IRE non-finishers scored 0.43 against GB 1.59. If the live system's edge is only in GB, it should be restricted to GB. Columns: region, race_type, last_fell_going_well.",
        "observation_ids": [
          "O-add18290",
          "O-3f2543b9"
        ],
        "priority": 0.55
      },
      {
        "question": "Is the edge concentrated in horses whose fall was a one-off, rather than horses with repeated jumping errors?",
        "rationale": "The market prices a known habit. It should not price a single error made while travelling well. Columns: jumping_errors_last3, f3_fell, last_fell_going_well, chase_runs.",
        "observation_ids": [
          "O-add18290",
          "O-bd96cc21"
        ],
        "priority": 0.5
      }
    ],
    "hypotheses": [
      {
        "title": "Rival: chase fallers/unseaters NOT going well show the same lift",
        "family": "trajectory",
        "operator": "rival-explanation",
        "question_id": null,
        "new_question": 0,
        "parent_id": "H85-fell-when-going-well",
        "reasoning": {
          "observation_ids": [
            "O-add18290"
          ],
          "question": "Is the lift specific to falls made while going well, or common to every last-time chase F or U?",
          "mechanism": "If punters simply over-penalise an F or U, the bounce-back would show in every chaser that failed to complete through a fall or unseat, whatever its position at the time.",
          "market_miss": "Form figures show F or U the same way whether the horse was going well or already beaten. So if the market overreacts to the letter itself, it overreacts to both groups.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.00 to 1.05 on 400+ bets if the mechanism is specific (the rival fails). A/E 1.15+ would mean the rival holds.",
            "falsified_if": "The rival is ruled out if A/E is at or below 1.05 with the same-race control near 1.0, which supports 'going well' as the active ingredient."
          },
          "rule_sketch": "race_type = 'Chase' AND last_fell_or_unseated = 1 AND last_fell_going_well = 0"
        },
        "confounders": [
          "horses that fell early never showed their ability",
          "price band differences",
          "field size"
        ],
        "expected_survival": 0.35
      },
      {
        "title": "Sharpen: fell-going-well chasers with no other recent jumping errors",
        "family": "trajectory",
        "operator": "sharpen",
        "question_id": null,
        "new_question": 2,
        "parent_id": "SYS-fell-when-going-well-chase",
        "reasoning": {
          "observation_ids": [
            "O-add18290"
          ],
          "question": "Is the edge strongest when the fall was a one-off?",
          "mechanism": "A chaser that otherwise jumps cleanly and fell while travelling well lost a run to bad luck. A habitual faller's falls predict more falls.",
          "market_miss": "A habit shows over several runs and gets priced. A single F next to otherwise clean form is read as failure, and the market does not see the good run behind it.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.25 to 1.40 on 100+ bets",
            "falsified_if": "A/E at or below the parent system's 1.30, or below 1.10."
          },
          "rule_sketch": "race_type = 'Chase' AND last_fell_going_well = 1 AND jumping_errors_last3 <= 1"
        },
        "confounders": [
          "jumping_errors_last3 may count the fall itself",
          "small sample",
          "horses with few chase runs"
        ],
        "expected_survival": 0.4
      },
      {
        "title": "Replicate: fell-going-well chase edge in IRE",
        "family": "trajectory",
        "operator": "replicate",
        "question_id": null,
        "new_question": 1,
        "parent_id": "SYS-fell-when-going-well-chase",
        "reasoning": {
          "observation_ids": [
            "O-add18290",
            "O-3f2543b9"
          ],
          "question": "Does the chase edge hold in Ireland, or is it GB-led like the chase-error thread?",
          "mechanism": "If the edge is a misread of the form figures, it should appear wherever punters read F and U beside a horse's name, so in IRE as well as GB.",
          "market_miss": "Irish form figures show the same F or U. But Irish chase markets are dominated by big yards and may price their runners differently.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.10 to 1.25 on 60+ bets",
            "falsified_if": "An IRE A/E at or below 1.0 would mean the system should be restricted to GB."
          },
          "rule_sketch": "region = 'IRE' AND race_type = 'Chase' AND last_fell_going_well = 1"
        },
        "confounders": [
          "small IRE sample",
          "festival concentration",
          "dominant yards"
        ],
        "expected_survival": 0.4
      }
    ],
    "findings": [
      {
        "statement": "Fell when going well carries a lift overall (A/E 1.13 on 405, both periods above 1.1), but the profit is at BSP only (+36.5%) while SP ROI is -3.0%. The edge needs exchange prices to pay.",
        "kind": "holds",
        "observation_ids": [
          "O-add18290"
        ]
      },
      {
        "statement": "The lift concentrates in chases (1.297 on 165), which leaves the other roughly 240 bets near par. That is consistent with an F or U meaning more in chase form.",
        "kind": "data",
        "observation_ids": [
          "O-add18290"
        ]
      }
    ]
  }

Timeline

1. hypothesis note H85-fell-when-going-well

   15 Oct, 23:10 · importer

2. H85-fell-when-going-well Discovery → PROMISING import: tier B: interesting, unproven; best [chase] GREEN A/E 1.297 on 165 bets (broad GREEN)

   15 Oct, 23:10 · importer

3. import result for H85-fell-when-going-well (chase): info

   15 Oct, 23:10 · importer

4. import result for H85-fell-when-going-well (14/1 or shorter): info

   15 Oct, 23:10 · importer

5. import result for H85-fell-when-going-well (no repeat jumping errors): info

   15 Oct, 23:10 · importer

6. import result for H85-fell-when-going-well (broad): info

   15 Oct, 23:10 · importer

7. New idea H85-fell-when-going-well: H85: fell when going well (import)

   15 Oct, 23:10 · importer

Every change is an event with an actor: an agent and its version, an engine, a human, or the importer. Current stage: Promising.