Rival: GB fallers going well vs GB fallers not going well, the going-well premium within GB

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
   • H-6d4d8b5d (Sharpen: fell-going-well chasers with no other recent jumping errors): the country 'GB' carries the effect (A/E 1.46) while the rest score IRE 0.78. O-1a1228ef
   • H-f2244c8b (Rival: chase fallers/unseaters NOT going well show the same lift) passed discovery as a niche: A/E 1.08 on 270 bets (p 0.356), same-race control 0.99. O-39a98a2c
2. AskedDoes 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? answered
3. MechanismIf going well matters, GB non-going-well F/U should sit near par (the rival's GB figure was 1.035 on 191) while GB going-well sits near 1.4. If both are similar, the cause is the letter, not the run.
4. Why the market misses itThe F/U letter hides how far the horse had got in the race. Only in-running comments show that it was travelling well.
5. Predictednegative, A/E 0.95 to 1.05 on 190+ bets, i.e. no lift for GB non-going-well F/U
6. Falsified ifGB non-going-well chase F/U score A/E at or above 1.15, which would mean the going-well condition adds little in GB.
7. Rule sketch

Mechanism and rationale

If going well matters, GB non-going-well F/U should sit near par (the rival's GB figure was 1.035 on 191) while GB going-well sits near 1.4. If both are similar, the cause is the letter, not the run.

The rule

region = 'GB' AND race_type = 'Chase' AND last_fell_or_unseated = 1 AND COALESCE(last_fell_going_well, 0) = 0

Reads: region, race_type, last_fell_or_unseated, last_fell_going_well

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

1cf20b5b222df958d60b1b729cd0882e35e5c64d3bd39e478132fa0e5e9d170a

Frozen

not frozen

The definition can still change.

Confounders

• Calendar skew: the lift in earlier cuts sits largely after 27 April 2026, so check A/E before and after that date separately
• Price band mix: the going-well cut's profit rests on 4 winners from 37 bets at 16/1 or bigger, and the non-going-well group went 0 from 98 at 16/1+, so compare the groups within the same price bands
• Course: big-fence tracks (Aintree, Cheltenham) give more F/U from horses travelling well, and those runners may move to easier tracks next time
• Field size: big fields cause more interference-driven unseats, which say less about the horse
• Fell vs unseated mix: the non-going-well group may hold more unseats, which are more often down to the jockey or bad luck
• Repeat jumping errors (jumping_errors_last3): a non-going-well faller may be a chronic bad jumper

Control

{
  "description": "The main comparison is GB chase runners whose last run ended in a fall or unseat while going well. That is the condition this hypothesis is meant to separate out, and it scored about 1.46 on 92 in earlier cuts. If going well is the cause, the control should sit well above the test group (a gap of 0.25 or more in A/E). The lab's usual same-race comparison (the other runners in these races) should show about 1.0, which rules out a race-level property.",
  "rule": "region = 'GB' AND race_type = 'Chase' AND last_fell_going_well = 1"
}

Discovery result

fail broad rule · 17 Oct, 02:35

Bets

191

21 winners

A/E

1.03

95%: 0.64 to 1.58

SP return

-39.5%

level stakes

Betfair SP return

-33.7%

191 bets with a BSP

Strike rate 11.0%. p = 0.47. Same-race control: the other runners in these races had A/E 1.00, 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 · 37s · 17 Oct, 02:33

  Output (schema-checked)

  {
    "rule": "region = 'GB' AND race_type = 'Chase' AND last_fell_or_unseated = 1 AND COALESCE(last_fell_going_well, 0) = 0",
    "refinements": [
      {
        "label": "Price-matched to the going-well cut: 15.0 decimal (14/1) or shorter",
        "rule": "region = 'GB' AND race_type = 'Chase' AND last_fell_or_unseated = 1 AND COALESCE(last_fell_going_well, 0) = 0 AND sp_decimal <= 15.0"
      }
    ],
    "control": {
      "description": "The main comparison is GB chase runners whose last run ended in a fall or unseat while going well. That is the condition this hypothesis is meant to separate out, and it scored about 1.46 on 92 in earlier cuts. If going well is the cause, the control should sit well above the test group (a gap of 0.25 or more in A/E). The lab's usual same-race comparison (the other runners in these races) should show about 1.0, which rules out a race-level property.",
      "rule": "region = 'GB' AND race_type = 'Chase' AND last_fell_going_well = 1"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1,
      "max_p": 0.2
    },
    "confounders": [
      "Calendar skew: the lift in earlier cuts sits largely after 27 April 2026, so check A/E before and after that date separately",
      "Price band mix: the going-well cut's profit rests on 4 winners from 37 bets at 16/1 or bigger, and the non-going-well group went 0 from 98 at 16/1+, so compare the groups within the same price bands",
      "Course: big-fence tracks (Aintree, Cheltenham) give more F/U from horses travelling well, and those runners may move to easier tracks next time",
      "Field size: big fields cause more interference-driven unseats, which say less about the horse",
      "Fell vs unseated mix: the non-going-well group may hold more unseats, which are more often down to the jockey or bad luck",
      "Repeat jumping errors (jumping_errors_last3): a non-going-well faller may be a chronic bad jumper"
    ],
    "mechanism": "A chase F or U wipes out the form line whatever happened in the race. If the edge comes from the hidden ability of a horse that was travelling well when it fell, GB chasers that fell or unseated while not going well should have no hidden ability to reveal. They should run close to their price (A/E about 1.0). If instead they show a lift like the going-well group (about 1.4), the edge is a generic bounce-back after an F/U letter and the going-well comment adds nothing.",
    "market_miss": "Prices lean on the form figures, where every F/U looks alike. Only the in-running comment shows whether the horse was travelling well at the time. If the market treats all F/U the same, the mispricing should exist only where the comment hides real ability (going well), not in this group.",
    "prediction": {
      "direction": "negative",
      "size": "No lift: A/E about 0.95 to 1.05 on 190 or more bets, at least 0.25 below the GB going-well control",
      "falsified_if": "GB chase F/U runners that were not going well score A/E of 1.15 or higher, or within 0.15 of the GB going-well control. Either would mean the going-well condition adds little in GB and the system is a generic F/U bounce-back."
    },
    "notes": "This is a rival-explanation test: a null result (no pass) supports the going-well system, and a lift falsifies its specific mechanism. In the earlier rival cut the GB share of this group scored 1.035 on 191. The one refinement matches the price band to rule out an artefact from long-priced winners. Every column is a pre-off catalogue column: region, race_type, last_fell_or_unseated, last_fell_going_well, sp_decimal."
  }

Timeline

1. Job 2471 finished

   17 Oct, 02:35 · engine:compute-1

2. H-48103daf Discovery → KILLED fails: 191 bets, needs 300; A/E 1.035, needs 0.95 or lower; p 0.6194 (lower tail), needs 0.05; periods [1.03, 1.05], each needs 0.99 or lower; A/E 1.035 against same-race control 1.0, needs 0.03 below

   17 Oct, 02:35 · engine:evaluator

3. discovery result for H-48103daf (broad rule): fail

   17 Oct, 02:35 · engine:evaluator

4. H-48103daf Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 02:35 · engine:orchestrator

5. compute-1 took job 2471 (evaluate, attempt 1)

   17 Oct, 02:35 · compute-1

6. Job 2469 finished

   17 Oct, 02:34 · engine:ai-1

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

   17 Oct, 02:34 · engine:scheduler

8. H-48103daf Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 02:34 · agent:designer@v1

9. ai-1 took job 2469 (design, attempt 1)

   17 Oct, 02:33 · ai-1

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

    17 Oct, 02:33 · engine:scheduler

11. New idea H-48103daf: Rival: GB fallers going well vs GB fallers not going well, the going-well premium within GB (rival-explanation)

    17 Oct, 02:33 · 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.