Rival: the festival-course lift is a long-price artefact

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
   • H228-dual-purpose-horse-in-an-irish-summer-flat-handicap (H228: dual purpose horse in an irish summer flat handicap) was GREEN in the earlier research: A/E 1.05 on 1890 bets (periods 1.03 / 1.08 / 0.99); now PROMISING. O-a317bdab
2. AskedIs the festival-course lift (A/E 1.2 on 544) carried by long-priced winners rather than by the field generally? answered
3. MechanismIf the Flat mark really lags behind hurdle form, the advantage should show among the horses the market fancies, not only among a few big-priced winners. The +40.7% BSP ROI against -23.7% SP ROI points to the rival: big BSP winners.
4. Why the market misses itIf the mechanism is real, the market under-rates the whole profile, so shorter-priced runners should also beat their prices.
5. Predictedpositive, A/E 1.05 to 1.12 on 300+ bets at under 10/1
6. Falsified ifA/E below 1.00 at under 10/1 on 250+ bets, which would mean the system's lift is price-band variance
7. Rule sketchregion = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND hurdle_runs >= 3 AND course IN ('GALWAY', 'LISTOWEL', 'KILLARNEY', 'BALLINROBE') AND sp_decimal < 11

Mechanism and rationale

If the Flat mark really lags behind hurdle form, the advantage should show among the horses the market fancies, not only among a few big-priced winners. The +40.7% BSP ROI against -23.7% SP ROI points to the rival: big BSP winners.

The rule

region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND prior_hurdle_runs >= 3 AND course IN ('GALWAY', 'LISTOWEL', 'KILLARNEY', 'BALLINROBE') AND sp_decimal < 11

Reads: region, race_type, is_handicap, month, distance_yards, prior_hurdle_runs, course, sp_decimal

Pre-registration

Spec hash (sha256 of the canonical pre-spec)

293ef7fe91b9b22c9e1e8067db1dbfa16d6e7c8593f43ba5767855034bb5fe5e

Frozen

not frozen

The definition can still change.

Confounders

• The four courses were picked after seeing the numbers on a parent rule that failed (0.986 on 744 in the lab window). Discovery on overlapping data is therefore partly in-sample, and only the forward record is clean.
• Small sample after the price cut: perhaps 200 to 300 bets out of the slice's 544, so the p-value will be weak even if the A/E is real.
• Galway week (late July and early August) may dominate the sample. A Galway-only effect, such as its very large fields and heavy betting, would pass for a festival-course effect. The ex-Galway refinement checks this.
• Field size: festival staying handicaps have big fields, which inflate SP overrounds and can lift A/E for any profile (rival H-7f585029). The same-race rivals measure this.
• Column choice: the system uses prior_hurdle_runs and the parent used hurdle_runs. This spec uses prior_hurdle_runs, which the catalogue clearly defines as before today. If the two columns disagree on Flat runs, the result is not comparable with H228.
• SP against BSP: the imported ROI gap (+40.7% BSP, -23.7% SP) shows that the price measure matters. A/E is judged against SP-implied expectation, so BSP-only profits are not evidence here.

Discovery result

fail broad rule · 17 Oct, 03:21

Bets

32

4 winners

A/E

0.83

95%: 0.23 to 2.13

SP return

-25.4%

level stakes

Betfair SP return

-11.6%

32 bets with a BSP

Strike rate 12.5%. p = 0.71. Same-race control: the other runners in these races had A/E 1.08, 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 · 48s · 17 Oct, 03:16

  Output (schema-checked)

  {
    "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND prior_hurdle_runs >= 3 AND course IN ('GALWAY', 'LISTOWEL', 'KILLARNEY', 'BALLINROBE') AND sp_decimal < 11",
    "refinements": [
      {
        "label": "fancied: 5/1 or shorter",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND prior_hurdle_runs >= 3 AND course IN ('GALWAY', 'LISTOWEL', 'KILLARNEY', 'BALLINROBE') AND sp_decimal <= 6"
      },
      {
        "label": "ex-Galway: festival lift not one meeting",
        "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND prior_hurdle_runs >= 3 AND course IN ('LISTOWEL', 'KILLARNEY', 'BALLINROBE') AND sp_decimal < 11"
      }
    ],
    "control": {
      "description": "Two comparisons. (1) The lab's standard check: the other runners in the same races. If the hurdlers have a real edge, those rivals should come out at or a little under A/E 1.0. If the rivals also beat their prices, the effect belongs to the race (for example festival overround or a field-size effect), not to the dual-purpose horse. (2) The long-price complement: the same four-course profile at sp_decimal >= 11. The rival explanation predicts that most of the system's 1.2 lift sits here, carried by a few big BSP winners, while the under-10/1 rule sits near 1.0. The mechanism predicts both bands beat their prices. If the main rule is flat and only the complement lifts, the system's lift is a long-price artefact.",
      "rule": "region = 'IRE' AND race_type = 'Flat' AND is_handicap = 1 AND month BETWEEN 6 AND 9 AND distance_yards >= 2640 AND prior_hurdle_runs >= 3 AND course IN ('GALWAY', 'LISTOWEL', 'KILLARNEY', 'BALLINROBE') AND sp_decimal >= 11"
    },
    "pass_criteria": {
      "min_bets": 150,
      "min_ae": 1.05,
      "max_p": 0.05
    },
    "forward_bar": {
      "min_bets": 100,
      "min_ae": 1.02,
      "max_p": 0.2
    },
    "confounders": [
      "The four courses were picked after seeing the numbers on a parent rule that failed (0.986 on 744 in the lab window). Discovery on overlapping data is therefore partly in-sample, and only the forward record is clean.",
      "Small sample after the price cut: perhaps 200 to 300 bets out of the slice's 544, so the p-value will be weak even if the A/E is real.",
      "Galway week (late July and early August) may dominate the sample. A Galway-only effect, such as its very large fields and heavy betting, would pass for a festival-course effect. The ex-Galway refinement checks this.",
      "Field size: festival staying handicaps have big fields, which inflate SP overrounds and can lift A/E for any profile (rival H-7f585029). The same-race rivals measure this.",
      "Column choice: the system uses prior_hurdle_runs and the parent used hurdle_runs. This spec uses prior_hurdle_runs, which the catalogue clearly defines as before today. If the two columns disagree on Flat runs, the result is not comparable with H228.",
      "SP against BSP: the imported ROI gap (+40.7% BSP, -23.7% SP) shows that the price measure matters. A/E is judged against SP-implied expectation, so BSP-only profits are not evidence here.",
      "Trainer clustering: a few dual-purpose yards (for example Mullins and Elliott types) may supply much of the sample, so the result may reflect those yards rather than the profile."
    ],
    "mechanism": "Horses with three or more prior hurdle runs carry a Flat handicap mark that lags behind their current staying ability, because hurdle form does not feed into the Flat rating and stamina built over jumps carries over to 14f+ Flat trips. At the Irish summer festivals (Galway, Listowel, Killarney, Ballinrobe), staying Flat handicaps attract dual-purpose yards that target them. If this edge is real, it applies to the whole profile, so runners the market fancies (under 10/1) should beat their SP too, not only a few big-priced winners.",
    "market_miss": "Flat handicappers and Flat punters price these horses off their Flat ratings and Flat form, and give little weight to recent hurdle form and stamina. If the mechanism is real, the market under-rates the whole profile, so short-priced runners should beat their SP as well. If the lift exists only at 10/1+, the plain explanation is variance from a few big BSP winners, not a systematic mispricing.",
    "prediction": {
      "direction": "positive",
      "size": "A/E 1.05 to 1.12 on roughly 200 to 300+ bets at under 10/1 (sp_decimal < 11)",
      "falsified_if": "A/E below 1.00 at under 10/1 on 250+ bets. Also falsified if the main rule is below 1.03 while the long-price complement (sp_decimal >= 11) carries the slice's lift, which would make the system's 1.2 price-band variance."
    },
    "notes": "I kept the rule sketch but swapped hurdle_runs for prior_hurdle_runs. That column is clearly defined as before today and matches the live system's rule, which this rival is testing; an earlier finding flagged the column mismatch. The cut sp_decimal < 11 is strictly shorter than 10/1, which is the band named in the prediction. The four-course list comes from data already seen, so a discovery pass is weak evidence and the forward record carries the weight; for that reason the forward A/E bar is raised slightly to 1.02. Refinements are reasoning-led only: a stricter fancied band (5/1 or shorter) and an ex-Galway cut for the Galway-dominance confounder."
  }

Timeline

1. Job 2625 finished

   17 Oct, 03:21 · engine:compute-1

2. H-c2fd2f08 Discovery → KILLED fails: 32 bets, needs 150; A/E 0.833, needs 1.05; p 0.7064, needs 0.05 or lower; periods [], each needs 1.0; A/E 0.833 against same-race control 1.082, needs 0.03 above (not a niche pass: A/E 0.833 below 1.05; p 0.7063763912388672 above 0.5; does not beat the same-race control 1.082)

   17 Oct, 03:21 · engine:evaluator

3. discovery result for H-c2fd2f08 (broad rule): fail

   17 Oct, 03:21 · engine:evaluator

4. H-c2fd2f08 Designed → RESEARCHdiscovery measuring on the discovery zone

   17 Oct, 03:21 · engine:orchestrator

5. compute-1 took job 2625 (evaluate, attempt 1)

   17 Oct, 03:21 · compute-1

6. Job 2622 finished

   17 Oct, 03:17 · engine:ai-1

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

   17 Oct, 03:17 · engine:scheduler

8. H-c2fd2f08 Proposed → RESEARCHdesigned pre-spec written and hashed

   17 Oct, 03:17 · agent:designer@v1

9. ai-1 took job 2622 (design, attempt 1)

   17 Oct, 03:16 · ai-1

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

    17 Oct, 03:16 · engine:scheduler

11. New idea H-c2fd2f08: Rival: the festival-course lift is a long-price artefact (rival-explanation)

    17 Oct, 03:16 · 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.