{
  "title": "Locked design, transcribed from the original written plan",
  "provenance": "The author recorded the plan as written 2026-08-15 at 18:39 UTC-07:00. It entered version history with the results at 19:09:58 UTC-07:00. No pre-run copy survives, so that history cannot independently prove the claimed order. This transcription is not an earlier timestamp.",
  "universe": {
    "as_of": "2004-01-02",
    "top_n": 60,
    "minimum_dollar_volume": 5000000,
    "price_floor": 5,
    "type": "US common stock",
    "qualification": "at least 250 daily bars in each window"
  },
  "split": {
    "is_start": "2004-01-02",
    "is_end": "2014-12-31",
    "oos_start": "2015-01-01",
    "oos_end": "2024-12-31"
  },
  "families": {
    "CHANNEL": "close breaks above highest high of prior `sig` bars",
    "MACROSS": "SMA(`sig`) crosses above SMA(200)"
  },
  "slow_ma": 200,
  "headline_grid": {
    "sig": [
      10,
      20,
      30,
      40
    ],
    "k_atr": [
      1.0,
      1.5,
      2.0,
      2.5
    ],
    "r_mult": 3.8,
    "n": 16
  },
  "extended_grid": {
    "sig": [
      5,
      10,
      15,
      20,
      25,
      30,
      40,
      50
    ],
    "k_atr": [
      0.75,
      1.0,
      1.5,
      2.0,
      2.5,
      3.0
    ],
    "r_mult": [
      1.5,
      2.5,
      3.8
    ],
    "n": 144
  },
  "risk_skeleton": {
    "RISK_F": 0.02,
    "ATR_N": 14,
    "long_only": true,
    "one_position_at_a_time": true,
    "gaps_honoured_at_open": true
  },
  "friction": {
    "COST_R": 0.05,
    "meaning": "fraction of 1R paid per round trip"
  },
  "fair_null": "per-ticker buy-and-hold over the same window; alpha = strategy net% - B&H net%",
  "best_of_N": {
    "N_GRID": [
      1,
      2,
      4,
      8,
      16,
      32,
      64,
      144
    ],
    "draws_per_N": 200,
    "seed": 20260815
  },
  "predictions": [
    "The in-sample optimum looks like an edge. For the median instance, best-of-16 IS alpha over buy-and-hold is positive and materially above the median knob's IS alpha.",
    "THE KILL — the same knobs do not carry. For the median instance, the best-of-16 knobs' OOS alpha over buy-and-hold is ≤ 0, and the median instance's (OOS alpha − IS alpha) is strongly negative.",
    "The surface is a spike, not a plateau. For the median instance the spike gap is large relative to the spread of the 16 cells, and fewer than half of the 16 knob settings beat buy-and-hold in-sample.",
    "Best-of-N rises with N purely from the count. Mean best-of-N IS net increases monotonically in N, while the mean OOS net of the selected setting does not increase with N.",
    "Friction bites. With `COST_R = 0.05` the median knob's IS net falls materially."
  ],
  "conditional_dispositions": [
    "If P2 AND P3 both fail — the surface is a broad plateau (most of the 16 beat B&H IS, small spike gap) *and* the best knobs' OOS alpha holds at or above zero for the median instance — then the registered thesis is falsified, and that is a *better* episode, not a dead one: the strategy has real structure and the knobs are cosmetic. The lesson flips to *\"here is how you tell a fitted EA from a robust one — and here is one that passes.\"* Report it as the headline, in those words.",
    "If P2 holds but P3 fails — a broad plateau that still dies out-of-sample — then the parameter surface is not the buyer-facing tell we thought it was, the \"lone spike\" beat is struck from the spine, and the held-out half carries the episode alone. Say so explicitly.",
    "If P1 fails — the best-of-16 does not even look like an edge in-sample — the design is too weak to demonstrate anything and the result is inconclusive. Report it as inconclusive and do not narrate a verdict from it."
  ],
  "data": "Daily prices, the delisted-name history and the ticker reference file used for the renames come from Massive.com (formerly Polygon.io), pulled once on 2026-06-22 and split-adjusted as of that pull. The supporting opening-range study used one-minute bars from the same vendor. EODHD, the market-data affiliate linked in the video description, did not supply these measurements.",
  "separation": "Cleaning, joined-history exclusions and open-position re-reading were later work. They are not part of this original locked design."
}