{
  "id": "SR2",
  "question": "Starting from the same proposition (S21's block-in), with critique in both arms, does stating intent before painting and answering for every decision before completion (the intent ledger) change what painters do with it? Read as distributions of choices and of the defaults the work kept falling into, not scores.",
  "start": {
    "handoff": "SEL2-easel",
    "masks": "The masks are Lua globals in canvas units (1000 wide, 800 high, origin top left): m0 sky, m1 far ground band (a half-plane: everything below its curve), m2 snow (a half-plane below its curve), m3 the open water (brook), m4 the ruin's gable, m5 the ruin's side wall, m6 the second wall fragment, m7 walker coat, m8 back leg, m9 front leg, m10 head. Masks combine with + and -, e.g. m1 - m2 is the far strip only."
  },
  "conditions": {
    "critique_only": "After your 2nd, 4th and 6th material sitting a critic reads the painting (Marrowline and Sketcher) and writes a critique that cites your sittings; you will find it in your painting's critiques folder before you continue. When you propose completion, a fresh critic runs a completion test on a 2400px render of the proposed state: at most three failures, each with cited sittings, a visible passage and its consequence. A substantiated failure returns the painting to active. You decide what to do with every critique.",
    "critique_intent": "The same critique as every run receives, plus the intent ledger. Before your first material sitting, write what the painting is for (what it is about, what each element does: the brook, the walker, the ruin, the near snow, the light, the palette) and save it with painting declare; the studio refuses material sittings until you have. Before you propose completion, open the ledger for the exact state (painting intent) and answer every item as the painter: intended (you chose it, and why), accepted (it happened and you keep it, and why), or unintended. The ledger lists what the studio measures in the frame and the brief's clauses. An unanswered or unintended item blocks completion: paint it or withdraw the claim. In the completion test the critic checks each intended item is visible. A challenge to a visible intended decision does not block completion unless the critic argues it from a named clause of the brief; if your reading of that clause is visible too, the arbiter (Russell) decides the reading."
  },
  "control_definition": "critique_only is the comparison arm: it keeps every rule, record, critique and completion test of the studio, and the same method notes and allowance. It has no declared intent and no ledger: the studio refuses both on its runs, and its completion needs no ledger. Disputes about the brief cannot reach the arbiter in critique_only; a substantiated failure argued from the brief returns the work to active as any other failure does.",
  "runs_per_condition": 5,
  "waves": {
    "1": "the first three runs of each condition, in run order (R-numbers), run before any of wave 2",
    "2": "the remaining two runs of each condition; wave 2 runs only if wave 1 is judged worth extending, and that decision is recorded before wave 2 starts. If wave 2 does not run, the series reports 3 + 3 and says so."
  },
  "allowance": {
    "replay_seconds": 1200,
    "sittings": 12,
    "width": 600,
    "completion_tests": 3
  },
  "end_rule": "A run ends when the painting is completed, held by the painter, stopped by the allowance (checkpointed), or returned to active by its third completion test (recorded as end state hold, final = the state the third test read). A run awaiting the arbiter waits; it is not counted as ended until the ruling.",
  "painter": {
    "backend": "easel",
    "agent": "a Claude subagent with a fresh context per run, given only its packet; it never sees another run, the pilots folder, or P1-P4",
    "self_review": "You may look at your own work at any time (value, squint, mirror, crops) and judge it yourself. Before you stop, look at the whole canvas once in value and once in squint, and say in the look's decision what you see."
  },
  "method": [
    "Never repair over dry paint: every local repair shows its seam at full resolution. Branch from a clean state and re-run the whole coat instead.",
    "Lay the ground through, then the figure on top. Painting the ground around a figure leaves a margin that reads as a halo.",
    "Two full coats before a blend. One blended coat lets the block-in's grey back through and fresh marks read as pale dashes.",
    "Texture from the ground's own paint; a lighter texture mix becomes pebbles or dashes.",
    "A shadow darkens what it lies on: a fixed shadow colour is right on one surface only; a neutral transparent film darkens every surface it crosses. Saturated blue reads as water.",
    "Working areas need soft edges: a rectangular working area leaves a ruled seam; feather it (rect(...):blur(40)) or follow the picture's own shapes.",
    "One light means consistent shadows for every body; a large dark film reads as a band, not a shadow.",
    "Drawn accents go on last, over dry paint; small touches laid into wet paint mix away.",
    "Masks from the block-in are half-planes or exact shapes; know which before you paint with one.",
    "The 600px state hides textures, ghosts and halos; completion tests read a 2400px render."
  ],
  "critic": {
    "agent": "a Claude subagent with a fresh context per critique and per completion test, given the brief, the proposition (SEL2) and that painting's record only; never other runs",
    "voice": "Marrowline (pressure, collapse tags, non-repair when warranted) and Sketcher (structure); every claim cites a sitting and a visible passage",
    "schedule": "after material sittings 2, 4 and 6, and at every proposed completion (on a 2400px render, made by the orchestrator for both arms)",
    "same_in_both_arms": "the critic prompt is identical in both arms except that in critique_intent it is told to check the ledger's intended items (the studio refuses a test that skips them)"
  },
  "arbiter": {
    "who": "Russell",
    "scope": "reading disputes in critique_intent (a brief clause read two ways, both visible). Rulings are collected and brought to him in batches; the run waits.",
    "limit": "Ruling on a dispute tells the arbiter that run is in critique_intent; his blind annotation of that final is flagged as arm-aware in the report."
  },
  "categories": [
    {
      "id": "brook_broken",
      "name": "The brook's course",
      "question": "Is the brook broken into pieces, or does it take a different route from walker to ruin than a single continuous band?",
      "values": [
        "yes",
        "no"
      ],
      "definition": "Is the brook broken into pieces, or does it take a different route from walker to ruin than a single continuous band?"
    },
    {
      "id": "light_rim",
      "name": "The light behind the ruin",
      "question": "Look at the brightest area of sky near the ruin. Does that bright area itself have a rim you could trace with a pen (not the edge of a cloud next to it)?",
      "values": [
        "yes",
        "no"
      ],
      "definition": "Look at the brightest area of sky near the ruin. Does that bright area itself have a rim you could trace with a pen (not the edge of a cloud next to it)?"
    },
    {
      "id": "near_patchy",
      "name": "Near snow: patchy",
      "question": "Is the snow surface in the lower third patchy or textured (blotches, mottling, marks), rather than smooth?",
      "values": [
        "yes",
        "no"
      ],
      "definition": "Is the snow surface in the lower third patchy or textured (blotches, mottling, marks), rather than smooth?"
    },
    {
      "id": "near_cut",
      "name": "Near snow: cut",
      "question": "Does a path, channel or object cut across the lower third? Shadows do not count.",
      "values": [
        "yes",
        "no"
      ],
      "definition": "Does a path, channel or object cut across the lower third? Shadows do not count."
    },
    {
      "id": "ruin_traceable",
      "name": "The ruin's outline",
      "question": "Could you trace the main ruin's outline with a pen all the way round, even where it is close to the sky's value? Ignore small detached fragments.",
      "values": [
        "yes",
        "no"
      ],
      "definition": "Could you trace the main ruin's outline with a pen all the way round, even where it is close to the sky's value? Ignore small detached fragments."
    },
    {
      "id": "walker_glance",
      "name": "The walker at a glance",
      "question": "At a quick glance at 600px, is the walker a person or just a mark?",
      "values": [
        "a-person",
        "just-a-mark"
      ],
      "definition": "At a quick glance at 600px, is the walker a person or just a mark?"
    }
  ],
  "measures": {
    "pixels": "measure.py relations (value key, glow at horizon, dark reaching the ruin, the brook's fade) and intent.facts (path continuity to the ruin's foot, taper, figure against path, hue relations, tangents at the destination) on every final, regions from S21's spec, computed after unblinding",
    "process": "material sittings, branches, critiques, completion tests to end, end state, replay used; in critique_intent also items first answered unintended and arbiter rulings",
    "headline": "how often each arm's finals fall into the defaults SR1 and P1-P4 kept meeting (glow behind the ruin, the path delivering to the ruin's foot, the figure sinking into the path, a tangent at the ruin, shared warmth read as a symbol), counted per arm"
  },
  "annotators": "Russell and a fresh blind subagent, on the frozen six questions (categories-SR2-frozen v3.1), from blind/ only",
  "limits": [
    "Five runs per arm (three in wave 1) describe; they do not prove.",
    "Arm B will probably use more tests and sittings; 'more care' and 'intent' are not fully separated.",
    "The arbiter learns the arm of any run he rules on."
  ],
  "seed": 20261011,
  "frozen": "2026-10-11T15:57:06+00:00"
}