Negative: short-priced recruits to a strong new yard after a heavy beating are over-bet on the yard's name

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-53ab001f (Recruit to a yard good with new arrivals, after a heavy beating for it) was predicted positive but came out the other way: A/E 0.93 on 422 bets (p 0.399 for the opposite direction). O-0f113025
   • H-432686d7 (Negative: short-priced planned moves from yards proven poor at that mo) pointed the predicted way but failed the bar: A/E 0.87 on 195 bets (p 0.82), control 1.05. O-bc2c08d8
2. AskedAre recruits to a strong new yard, after a heavy beating for their old trainer, over-bet when they go off short-priced? In other words, does the market price the yard's name rather than the horse? answered
3. MechanismA change of trainer is the most visible event in the form book. After a heavy beating, punters and tipsters read the move to a fashionable yard as the explanation for the bad run and back the horse to bounce back straight away. Yet the horse has shown nothing on the course since the beating, and many recruits need time in their new yard.
4. Why the market misses itThe beating looks like a fresh start because of the new trainer's name, so money comes in on the yard's reputation and the price shortens beyond the horse's real chance. The market is not missing a hidden positive; it is paying too much for a story everyone can see.
5. Predictednegative, A/E 0.80 to 0.90 on 150+ bets
6. Falsified if

Mechanism and rationale

A change of trainer is the most visible event in the form book. After a heavy beating, punters and tipsters read the move to a fashionable yard as the explanation for the bad run and back the horse to bounce back straight away. Yet the horse has shown nothing on the course since the beating, and many recruits need time in their new yard.

The rule

trainer_id <> last_trainer_id AND tx_new_yard_runs >= 15 AND tx_new_yard_wins >= 1.3 * tx_new_yard_exp AND last_field_size >= 6 AND last_position >= 0.6 * last_field_size AND sp_decimal <= 8

Reads: trainer_id, last_trainer_id, tx_new_yard_runs, tx_new_yard_wins, tx_new_yard_exp, last_field_size, last_position, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

be5812ebfb4385441ca6e620bafdb8b74edcb8cf7561dae94ff6e1d6b0f98fd4

Frozen

not frozen

The definition can still change.

Confounders

• The 7/1 cut was chosen after seeing the parent rule (0.93 on 422), whose lift seemed to come from a few big-price winners. The short-price subset is not independent of that look: treat it as a post-hoc band and hold it to the forward record.
• GB/IRE mix: in the parent rule, GB scored 0.81 on 316 and IRE 1.35 on 106. A negative result could simply be the GB part, so the GB-only refinement and an IRE check are required.
• Favourite-longshot bias: short-priced runners already run close to fair, and fair/A-E should absorb the general bias. Any shortfall must also beat the same-race control, whose short-priced runners carry the same bias.
• Novice/maiden races, where recruits are unexposed and the beating may say little; the handicap-only refinement checks this.
• Trainer-change information the market prices correctly, such as a known wind operation or a move to a yard with more suitable targets. The price-matched non-strong-yard control separates this from money following the yard's name.
• Overlap with the parent rule H-53ab001f (a strict subset). The two are not independent lines of evidence.

Discovery result

fail broad rule · 17 Oct, 05:05

Bets

84

13 winners

A/E

0.81

95%: 0.43 to 1.39

SP return

-34.6%

level stakes

Betfair SP return

-19.2%

78 bets with a BSP

Strike rate 15.5%. p = 0.81. Same-race control: the other runners in these races had A/E 1.04, so the selection did not 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 · 1m · 17 Oct, 05:03

  Output (schema-checked)

  {
    "rule": "trainer_id <> last_trainer_id AND tx_new_yard_runs >= 15 AND tx_new_yard_wins >= 1.3 * tx_new_yard_exp AND last_field_size >= 6 AND last_position >= 0.6 * last_field_size AND sp_decimal <= 8",
    "refinements": [
      {
        "label": "GB only",
        "rule": "trainer_id <> last_trainer_id AND tx_new_yard_runs >= 15 AND tx_new_yard_wins >= 1.3 * tx_new_yard_exp AND last_field_size >= 6 AND last_position >= 0.6 * last_field_size AND sp_decimal <= 8 AND region = 'GB'"
      },
      {
        "label": "handicaps only (exposed horses, removes unexposed novice/maiden recruits)",
        "rule": "trainer_id <> last_trainer_id AND tx_new_yard_runs >= 15 AND tx_new_yard_wins >= 1.3 * tx_new_yard_exp AND last_field_size >= 6 AND last_position >= 0.6 * last_field_size AND sp_decimal <= 8 AND is_handicap = 1"
      },
      {
        "label": "5/1 or shorter (where the yard-name money should be heaviest)",
        "rule": "trainer_id <> last_trainer_id AND tx_new_yard_runs >= 15 AND tx_new_yard_wins >= 1.3 * tx_new_yard_exp AND last_field_size >= 6 AND last_position >= 0.6 * last_field_size AND sp_decimal <= 6"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) The standard same-race control: the other runners in the same races should come out near 1.00. The rule needs to come in at least 0.05 below them. (2) A price-matched rule-out of the 'yard name' story: recruits at 7/1 or shorter after a heavy beating whose new yard does NOT have a strong recruit record (tx_new_yard_runs >= 15 and wins below 1.3 x exp). If those are over-bet just as much, the cause is short-priced recruits after a beating in general (the market pricing in the trainer change), not money following a fashionable yard's name. Then the mechanism as stated fails.",
      "rule": "trainer_id <> last_trainer_id AND tx_new_yard_runs >= 15 AND tx_new_yard_wins < 1.3 * tx_new_yard_exp AND last_field_size >= 6 AND last_position >= 0.6 * last_field_size AND sp_decimal <= 8"
    },
    "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": [
      "The 7/1 cut was chosen after seeing the parent rule (0.93 on 422), whose lift seemed to come from a few big-price winners. The short-price subset is not independent of that look: treat it as a post-hoc band and hold it to the forward record.",
      "GB/IRE mix: in the parent rule, GB scored 0.81 on 316 and IRE 1.35 on 106. A negative result could simply be the GB part, so the GB-only refinement and an IRE check are required.",
      "Favourite-longshot bias: short-priced runners already run close to fair, and fair/A-E should absorb the general bias. Any shortfall must also beat the same-race control, whose short-priced runners carry the same bias.",
      "Novice/maiden races, where recruits are unexposed and the beating may say little; the handicap-only refinement checks this.",
      "Trainer-change information the market prices correctly, such as a known wind operation or a move to a yard with more suitable targets. The price-matched non-strong-yard control separates this from money following the yard's name.",
      "Overlap with the parent rule H-53ab001f (a strict subset). The two are not independent lines of evidence."
    ],
    "mechanism": "A change of trainer is the most visible event in the form book. After a heavy beating (finishing in the bottom 40% of a field of 6 or more), punters and tipsters read the move to a yard with a strong recruit record as the explanation for the bad run. They back the horse to bounce back first time. But the horse has shown nothing on the course since the beating, and many recruits need runs in the new yard before they come right. The yard's strong record is spread across all its recruits, not front-loaded on horses that have just been beaten heavily.",
    "market_miss": "The market is not missing a hidden positive. It is overpaying for a story everyone can see. The new trainer's name turns the beating into a 'fresh start', and reputation money shortens the price (to 7/1 or shorter) beyond the horse's real chance. The longer-priced recruits in the parent set, which the story did not shorten, are where any yard edge survived. So the overpricing should sit in the short band.",
    "prediction": {
      "direction": "negative",
      "size": "A/E about 0.80 to 0.90 on 150+ bets, at least 0.05 below the same-race control and below the price-matched recruits from non-strong yards",
      "falsified_if": "A/E at or above 0.97 on 150+ bets; or not at least 0.05 below the same-race control; or the price-matched non-strong-yard recruit control is over-bet by as much or more, which would mean the yard name is not the cause"
    },
    "notes": "The rule is kept exactly as sketched. All its columns (trainer_id, last_trainer_id, tx_new_yard_*, last_field_size, last_position, sp_decimal) are pre_off, and NULL last_position or last_trainer_id rows drop out naturally. min_ae 1.05 is read in the pre-registered negative direction, i.e. the rule's A/E must be about 0.95 or lower (1/1.05) with p <= 0.05. tx_new_yard_* data start 2020-11/12, so discovery begins there. The refinements come straight from the stated confounders (GB/IRE split, novice/maiden exposure, and the short-price story getting stronger as prices shorten) and are only looked at if the broad rule shows evidence."
  }

Timeline

1. Job 3137 finished

   17 Oct, 05:05 · engine:compute-1

2. H-61d14772 Discovery → KILLED fails: 84 bets, needs 300; p 0.2723 (lower tail), needs 0.05; periods [0.71], each needs 0.99 or lower

   17 Oct, 05:05 · engine:evaluator

3. discovery result for H-61d14772 (broad rule): fail

   17 Oct, 05:05 · engine:evaluator

4. H-61d14772 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 05:05 · engine:orchestrator

5. compute-1 took job 3137 (evaluate, attempt 1)

   17 Oct, 05:05 · compute-1

6. Job 3136 finished

   17 Oct, 05:04 · engine:ai-1

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

   17 Oct, 05:04 · engine:scheduler

8. H-61d14772 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 05:04 · agent:designer@v1

9. ai-1 took job 3136 (design, attempt 1)

   17 Oct, 05:03 · ai-1

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

    17 Oct, 05:03 · engine:scheduler

11. New idea H-61d14772: Negative: short-priced recruits to a strong new yard after a heavy beating are over-bet on the yard's name (rival-explanation)

    17 Oct, 05:03 · 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.