shared/SPLIT_MANIFEST_v2.json

{
  "schema_version": "outer-loop-split-manifest-v2",
  "protocol": "protocol/outer_loop_method_development_protocol.md rev-2 §2.1/§2.3",
  "domain": "pricing (Twin-2K-500 / Mega-Study wave3+wave4 purchase decisions)",
  "created": "2026-08-18",
  "supersedes": {
    "file": "runs/SPLIT_MANIFEST_v1.json",
    "manifest_sha256": "04797d0bba7c678710255d5d273c3ee67dd83cdcd0be6ae9b672cadcd40862c8",
    "reason": "v1 had no final-test reservation (Q_test = P_test = empty). Re-split to reserve a two-axis held-out final test, per user decision 2026-08-18."
  },
  "hash_rule": "sha256 over json.dumps(manifest_without_manifest_sha256, sort_keys=True, separators=(',',':'), ensure_ascii=False).encode()",
  "split_generation_rule": "D_dev inherited verbatim from the v0 release (seed 20260810) so that v0's archived R1/R2/R3 development trace stays valid. The v0 validation pool (30 personas x 20 products) is then partitioned by a deterministic seeded shuffle of the ID lists alone: personas 18 validation / 12 final test (seed 20260818), products 12 validation / 8 final test (seed 20260819). No response labels, accuracies, category labels, or per-unit outcomes were consulted.",
  "seed": {
    "dev_inherited": 20260810,
    "persona_partition": 20260818,
    "product_partition": 20260819
  },
  "dev_question_ids": [
    2,
    3,
    6,
    7,
    9,
    12,
    13,
    16,
    17,
    20,
    23,
    24,
    27,
    28,
    29,
    32,
    35,
    36,
    37,
    38
  ],
  "dev_persona_ids": [
    3,
    6,
    8,
    11,
    12,
    14,
    16,
    17,
    21,
    22,
    23,
    24,
    28,
    38,
    40,
    42,
    44,
    45,
    46,
    51
  ],
  "validation_question_ids": [
    1,
    4,
    5,
    8,
    11,
    15,
    18,
    19,
    22,
    30,
    33,
    39
  ],
  "validation_persona_ids": [
    2,
    4,
    5,
    9,
    10,
    18,
    25,
    26,
    27,
    32,
    33,
    35,
    37,
    41,
    43,
    47,
    49,
    50
  ],
  "final_test_question_ids": [
    10,
    14,
    21,
    25,
    26,
    31,
    34,
    40
  ],
  "final_test_persona_ids": [
    7,
    13,
    15,
    19,
    20,
    29,
    30,
    31,
    34,
    36,
    39,
    48
  ],
  "final_test_status": "RESERVED_SEALED",
  "cells": {
    "D_dev": 400,
    "D_val": 216,
    "D_test": 96
  },
  "historical_contamination_disclosure": "The pricing pool is capped at 50 respondents x 40 products; no additional labelled personas or products exist. Every unit reserved for D_test therefore appeared in the v0 release evaluation, whose aggregate results were published and inspected. D_test is sealed from v1 onward and its cell-level outcomes have never been used to choose or tune any AutoPipeline version, but it is NOT contamination-free in the strict §2.4 sense. Any final-test result must be reported with this disclosure.",
  "manifest_sha256": "36aea05770a9ae346e3fccc9b5968e85690fa4699c6cd93406495a2a93e449bb"
}