System: One jumping mistake last time, and well beaten

Mechanism and rationale

A chaser that jumped cleanly in its two runs before, made a mistake last time and finished well beaten or not at all. The mistake cost it the race, and the form figure hides how it was going. It wins more often than its prices imply but loses at bookmaker starting price: it pays at Betfair SP. Twelve months of data only.

The rule

race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)

Reads: race_type, remarked_last3, jumping_errors_last3, last_jumping_error, last_position, last_beaten_lengths

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

not designed yet

Frozen

not frozen

The definition can still change.

Backing

exchange

Kind

system

Live from

2026-10-13

Matched

{
  "hypothesis_id": "H24-one-off-jumping-error-well-beaten",
  "similarity": 0.83,
  "variant": "chase"
}

Published

true

Rules

• The race is a chase.
• Last time out the horse made a jumping error and was beaten more than 10 lengths or failed to finish.
• No jumping error was noted in its two runs before that.

Source

In-running remarks, October 2026

System id

one-off-jumping-error-chase

Tier

none

Where

race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10)

Discovery result

Not measured yet.

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 · 53s · 17 Oct, 01:02

  Output (schema-checked)

  {
    "reflection": "This thread has one measurement so far, the imported result, and it is a strong one. A chaser or hurdler that jumped cleanly in its two runs before, then had a jumping error noted last time and finished more than 10 lengths behind or did not finish, beat its price: A/E 1.17 on 1,570 bets. It held in both periods (1.12 and 1.23) and made +15% at BSP against -14.1% at SP (O-6e6fe2b6). The chase slice is the stronger one (1.224 on 536) and is now live as SYS-one-off-jumping-error-chase, with no forward bets yet. So hurdles carry less of the lift. Imported siblings found by search (these are not observations in this run) complicate the stated mechanism, that the mistake cost the run. First, among chasers with a remark error, the cost measured by the figures adds little: error with large cost scored 1.057 and error with little cost 0.901. Second, inside this exact rule, runners whose earlier Jump Index was poor scored 1.879 on 161, against 1.186 on 179 for good earlier jumpers. That is the opposite of the 'good jumper, one bad day' story. Two other things could explain the lift, and neither is ruled out. The pool includes non-finishers, which overlaps the known value in fallers and unseaters. The remark itself may simply be a public excuse the market over-weights or under-weights. We do not yet know which part of the pool carries the edge. All remark columns start in November 2025, so every test here is on about a year of data.",
    "answers": [],
    "questions": [
      {
        "question": "Does the H24 chase edge come from finishers beaten more than 10 lengths, or from runners that did not finish (fell, unseated, pulled up)?",
        "rationale": "The rule's last_position IS NULL branch merges fallers and unseaters with genuinely well-beaten finishers. If the lift is only in non-finishers, it repeats the known fell/unseated forgive story rather than a remark-based edge. The scanner should split on last_position IS NULL, last_fell_or_unseated, last_pulled_up and last_code, inside race_type = 'Chase' with jumping_errors_last3 = 1.",
        "observation_ids": [
          "O-6e6fe2b6"
        ],
        "priority": 0.7
      },
      {
        "question": "Within the H24 chase pool, is the lift larger when the sectionals show the error actually cost ground (low l_lgj_pct / last_ground_gained_pct)?",
        "rationale": "The mechanism says the mistake cost the run. If that is true, a costly error should pay more than an error that cost nothing by the figures. The imported siblings found by search (error with large cost 1.057, error with little cost 0.901) suggest it may not. The scanner should read l_lgj_pct, last_ground_gained_pct and l_speed_lost_pct.",
        "observation_ids": [
          "O-6e6fe2b6"
        ],
        "priority": 0.6
      },
      {
        "question": "Why do previously poor jumpers by Jump Index outperform good ones inside this rule (1.879 vs 1.186 in the imported overlays)?",
        "rationale": "This runs against the 'one-off' story. Two explanations are possible: the market already marks down poor jumpers for their figures as well as for the remark, or the 'clean in the two runs before' remarks wrongly reassure on horses whose figures say they jump poorly. The scanner should read usual_jump_index_pct, s_jump_index_pct_prev and sp_decimal within the H24 chase pool.",
        "observation_ids": [
          "O-6e6fe2b6"
        ],
        "priority": 0.45
      }
    ],
    "hypotheses": [
      {
        "title": "Rival: H24 chase edge is only the faller/unseater story; finishers beaten 10L+ do not beat their price",
        "family": "story",
        "operator": "rival-explanation",
        "question_id": null,
        "new_question": 0,
        "parent_id": "SYS-one-off-jumping-error-chase",
        "reasoning": {
          "observation_ids": [
            "O-6e6fe2b6"
          ],
          "question": "Does the one-off jumping error edge survive when the run is restricted to horses that completed and were beaten more than 10 lengths?",
          "mechanism": "If the jumping-error remark is what misleads the market, completed runs where the mistake left the horse well beaten should carry the lift too. If not, the edge comes from fallers and unseaters, whose zero in the form figures is a known mispricing that has nothing to do with this remark.",
          "market_miss": "A finisher beaten 15 lengths shows a poor figure and a big beaten distance. Punters read that as a lack of ability, not as the one mistake in the remark.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.10 to 1.20 on 250+ bets if the remark-based story is real",
            "falsified_if": "A/E of 1.03 or lower on the finishers-only pool, while the full system stays above 1.15. That would mean the lift belongs to non-finishers."
          },
          "rule_sketch": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND last_position IS NOT NULL AND last_beaten_lengths > 10"
        },
        "confounders": [
          "non-finishers' prices are already discounted for the fall",
          "field size of the last race",
          "going on the last run",
          "class of the last race"
        ],
        "expected_survival": 0.45
      },
      {
        "title": "Test-mechanism: one-off chase error that the sectionals show cost ground",
        "family": "story",
        "operator": "test-mechanism",
        "question_id": null,
        "new_question": 1,
        "parent_id": "SYS-one-off-jumping-error-chase",
        "reasoning": {
          "observation_ids": [
            "O-6e6fe2b6"
          ],
          "question": "If the mistake cost the run, is the lift stronger when the figures confirm the error lost ground?",
          "mechanism": "A one-off error that physically cost lengths at the fences explains the heavy defeat. The horse's real level is better than the result, so it should win more often than its price says next time.",
          "market_miss": "The remark only says 'mistake'. The market cannot tell a mistake that cost five lengths from one that cost nothing, so it marks every remark error the same way.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.25 to 1.40 on 150+ bets, above the parent's 1.224",
            "falsified_if": "A/E at or below the parent's 1.224, or below 1.10, on this subset. That would mean the measured cost does not carry the edge."
          },
          "rule_sketch": "race_type = 'Chase' AND remarked_last3 = 3 AND jumping_errors_last3 = 1 AND last_jumping_error = 1 AND (last_position IS NULL OR last_beaten_lengths > 10) AND l_lgj_pct <= 0.33"
        },
        "confounders": [
          "l_lgj_pct missing for non-finishers",
          "ground lost may reflect tiredness rather than the error",
          "small sample after November 2025"
        ],
        "expected_survival": 0.35
      }
    ],
    "findings": [
      {
        "statement": "The one-off jumping error, well beaten idea held at A/E 1.17 on 1,570 bets in both periods (1.12 and 1.23), with +15% BSP ROI. The chase slice (1.224 on 536) is stronger than the hurdle part and is live with no forward record.",
        "kind": "holds",
        "observation_ids": [
          "O-6e6fe2b6"
        ]
      },
      {
        "statement": "The SP ROI of -14.1% against +15% at BSP shows that this edge depends on exchange prices: the edge is real only at BSP, not at bookmaker SP.",
        "kind": "method",
        "observation_ids": [
          "O-6e6fe2b6"
        ]
      }
    ]
  }

Timeline

1. hypothesis note SYS-one-off-jumping-error-chase

   15 Oct, 23:10 · importer

2. New idea SYS-one-off-jumping-error-chase: System: One jumping mistake last time, and well beaten (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: Live.