H98: brought down or unseated

Mechanism and rationale

Brought down or unseated its rider last time: nothing wrong with the horse, and it had no race.

The rule

last_code IN ('BD', 'UR')

Reads: last_code

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

not designed yet

Frozen

not frozen

The definition can still change.

Broad

last_code IN ('BD', 'UR')

Kind

remarks

Origin

research/experiments/overnight/results_v2.jsonl

Refinements

[
  {
    "name": "unseated",
    "rule": "(last_code IN ('BD', 'UR')) AND (last_code = 'UR')"
  },
  {
    "name": "no repeat jumping errors",
    "rule": "(last_code IN ('BD', 'UR')) AND (jumping_errors_last3 <= 1)"
  },
  {
    "name": "14/1 or shorter",
    "rule": "(last_code IN ('BD', 'UR')) AND (sp_decimal <= 15)"
  },
  {
    "name": "back within 35 days",
    "rule": "(last_code IN ('BD', 'UR')) AND (days_since_last_run <= 35)"
  }
]

Tier

A

Result (from the archive)

info back within 35 days · 15 Oct, 23:10

Bets

464

A/E

1.07

winners vs the prices

SP return

+15.3%

level stakes

Betfair SP return

+138.6%

464 bets with a BSP

Strike rate 9.9%.

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 · 52s · 16 Oct, 13:13

  Output (schema-checked)

  {
    "reflection": "The thread has one measured result so far. In the earlier research, runners that were brought down or unseated their rider last time scored A/E 1.10 on 612 bets. The two periods were 1.17 on 334 bets, then 1.02 on 278. Discovery gave 1.071 on 464. The unseated-only cut gave 1.109 on 556, and that was published as a live system.\n\nTwo things in the numbers weaken the story that the market under-reads an innocent 'U'.\n1. The lift shrank from 1.17 to 1.02 between periods. The later period on its own is almost exactly the price.\n2. SP ROI is +16.6% but BSP ROI is +140%. That gap means the profit comes from a few very big exchange-priced winners. The system's own note agrees: at 14/1 or shorter these runners only match their price.\n\nSo the honest belief is this. A broad, small edge exists in the record, but it may be the general longshot drift plus a few outliers rather than a mispriced 'nothing went wrong' signal.\n\nThe closest sibling, H118 ('out of the race early', within 28 days), was killed at 1.049 on 76 chase bets. That counts against the simple 'it had no race, so it is fresh' mechanism.\n\nThere is also a small inconsistency: the thread rule reads `last_code`, while the system reads `last_non_finish_code`. These should select the same runners. The lab should confirm they do before reading the live record against the old one.\n\nThe next useful tests are:\n- whether the edge survives a price cap and in the recent period;\n- whether a consequence the mechanism requires holds: horses brought back quickly, while still fresh, should do best.",
    "answers": [],
    "questions": [
      {
        "question": "Is the brought-down/unseated edge only a 16/1+ longshot effect, and is it fading (1.17 then 1.02)?",
        "rationale": "The gap between SP ROI (+16.6%) and BSP ROI (+140%), and the falling period A/E, both point to a few outsized winners. The test would split `last_non_finish_code IN ('BD','UR')` by `sp_decimal` bands (up to 15, and over 15) and by period. If the capped band sits at 1.0 in recent data, the live system is a longshot bet, not a reading-the-form edge. That would change how the system is staked and judged.",
        "observation_ids": [
          "O-9b549ef0"
        ],
        "priority": 0.65
      },
      {
        "question": "Does a quick return after an unseat or brought-down beat a slow return, as the 'had no race, still fresh' mechanism requires?",
        "rationale": "The mechanism is that the horse kept its fitness and form because it did no real racing. That predicts the lift is concentrated where `days_since_last_run` is short (28 days or less). After a long break, the incident no longer matters. Columns: `last_non_finish_code`, `days_since_last_run`, `race_type`. H118 (early exit within 28 days, chase) was killed at 1.049 on 76, so this needs the broader population.",
        "observation_ids": [
          "O-9b549ef0"
        ],
        "priority": 0.5
      },
      {
        "question": "Do `last_code IN ('BD','UR')` and `last_non_finish_code IN ('BD','UR')` select the same runners?",
        "rationale": "The hypothesis was measured on `last_code`, but the system is published on `last_non_finish_code`. If the two columns differ (for example, if `last_code` holds the jumping-shock Chase/Hurdle code), then the live record measures a different population from the one that was graded GREEN.",
        "observation_ids": [
          "O-9b549ef0"
        ],
        "priority": 0.4
      }
    ],
    "hypotheses": [
      {
        "title": "Unseated or brought down and back within four weeks: the fresh-horse consequence",
        "family": "letters",
        "operator": "test-mechanism",
        "question_id": null,
        "new_question": 1,
        "parent_id": "H98-brought-down-or-unseated",
        "reasoning": {
          "observation_ids": [
            "O-9b549ef0"
          ],
          "question": "Does a quick return after an unseat or brought-down beat a slow return, as the 'had no race, still fresh' mechanism requires?",
          "mechanism": "A horse that unseated its rider or was brought down did little racing. Its connections can run it again soon, while it is still at the fitness it had for the abandoned target. After a long break, the incident is irrelevant: the horse is just a returner.",
          "market_miss": "The form figure 'U' or 'B' reads as a failure, and the next run looks quick and unplanned. Punters do not credit that the horse arrives fresh and fit, with its last real form line still current.",
          "prediction": {
            "direction": "positive",
            "size": "A/E 1.10 to 1.18 on 250+ bets, above the same rule with days_since_last_run > 28",
            "falsified_if": "A/E below 1.03, or no better than the over-28-days complement"
          },
          "rule_sketch": "last_non_finish_code IN ('BD', 'UR') AND days_since_last_run <= 28"
        },
        "confounders": [
          "longshot winners at 16/1+",
          "chase versus hurdle mix",
          "yard habits of quickly re-running horses"
        ],
        "expected_survival": 0.3
      },
      {
        "title": "Rival: the unseated edge is only longshot drift; capped at 14/1 it matches the price",
        "family": "letters",
        "operator": "rival-explanation",
        "question_id": null,
        "new_question": 0,
        "parent_id": "SYS-brought-down-or-unseated",
        "reasoning": {
          "observation_ids": [
            "O-9b549ef0"
          ],
          "question": "Is the brought-down/unseated edge only a 16/1+ longshot effect, and is it fading?",
          "mechanism": "Runners coming off a non-finish drift to big prices. The A/E lift comes from the general underpricing of jump outsiders plus a handful of very big exchange winners, not from any misreading of the incident.",
          "market_miss": "If the rival is right, nothing is mispriced at sensible prices. The +140% BSP ROI against +16.6% SP ROI is a few outliers, and any real edge sits where the price is fair.",
          "prediction": {
            "direction": "positive",
            "size": "Rival predicts A/E 0.97 to 1.03 on 350+ capped bets; the thread mechanism predicts 1.08 or more",
            "falsified_if": "The capped rule scores A/E 1.08 or more with both periods above 1.03, which would rule the longshot rival out"
          },
          "rule_sketch": "last_non_finish_code IN ('BD', 'UR') AND sp_decimal <= 15"
        },
        "confounders": [
          "period decline (1.17 then 1.02)",
          "race_type mix",
          "field size inflating prices"
        ],
        "expected_survival": 0.35
      }
    ],
    "findings": [
      {
        "statement": "The brought-down/unseated lift depends on big prices. SP ROI was +16.6% against BSP ROI of +140% on 612 bets, so a few outsized exchange winners carry the profit.",
        "kind": "data",
        "observation_ids": [
          "O-9b549ef0"
        ]
      },
      {
        "statement": "The edge fell from 1.17 on 334 bets in the first period to 1.02 on 278 in the second, so the live system's forward record should be read against roughly 1.02, not the pooled 1.10.",
        "kind": "method",
        "observation_ids": [
          "O-9b549ef0"
        ]
      }
    ]
  }

Timeline

1. hypothesis note H98-brought-down-or-unseated

   15 Oct, 23:10 · importer

2. H98-brought-down-or-unseated Discovery → PROMISING import: tier A: strong candidate; best [unseated] GREEN A/E 1.109 on 556 bets (broad GREEN)

   15 Oct, 23:10 · importer

3. import result for H98-brought-down-or-unseated (back within 35 days): info

   15 Oct, 23:10 · importer

4. import result for H98-brought-down-or-unseated (14/1 or shorter): info

   15 Oct, 23:10 · importer

5. import result for H98-brought-down-or-unseated (no repeat jumping errors): info

   15 Oct, 23:10 · importer

6. import result for H98-brought-down-or-unseated (unseated): info

   15 Oct, 23:10 · importer

7. import result for H98-brought-down-or-unseated (broad): info

   15 Oct, 23:10 · importer

8. New idea H98-brought-down-or-unseated: H98: brought down or unseated (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.