# SR2: findings (wave 1, 3 + 3)

Records only. Read beside REPORT.md (every run, both annotations), MEASURES.md (pixels) and
finals-unblinded.jpg. Three runs per arm describe; they do not establish an effect.

## What both arms did alike

- **Geometry kept.** In all six finals the brook keeps S21's S-curve and the horizon stays where it was.
  Walker and ruin stay at or near their block-in places; R02 (critique_intent) moved the walker onto the far
  bank and rebuilt the ruin smaller.
- **Key inverted.** S21 is low-key (snow 0.559, sky 0.676, brook 0.165 as the darkest mass). All six finals
  are high-key (snow 0.68 to 0.86, sky 0.79 to 0.84).
- **Brook turned to light.** S21's dark open water became a light path in 5 of 6 (brook end lighter than the
  snow: R01, R03, R04, R06, R08). R02 kept a dark brook.
- **Light behind the ruin.** Five of six first stated questions set a low sun behind or beside the ruin
  (R01, R03, R04 in critique_only; R06, R08 in critique_intent). R02 (critique_intent) asked about distance
  measured by the walker, with a low sun to the left of the view, and kept the brook dark. Measured glow at the horizon: 5 of 6 (not R03). SR1, with critique on
  against off, also measured a horizon glow in 5 of 6.
- **Annotation agreement.** Both annotators: no brook broken (6 of 6), the walker reads as a person (6 of 6),
  a path or channel cuts the lower third in R01 and R06. Agreement was 28 of 36 answers; 4 of the 8 splits
  were the near-snow texture question.

## Where the arms differ (descriptive)

| | critique_only (R01, R03, R04) | critique_intent (R02, R06, R08) |
|---|---|---|
| End | 3 hold | 1 completed (R06), 2 hold |
| Completion tests | 5 (2, 1, 2) | 7 (2, 2, 3) |
| Material sittings | 12, 11, 12 | 12, 10, 12 |
| Branches | 4, 2, 3 | 9, 5, 2 |
| Reading disputes to the arbiter | none possible | 6, all upheld for the critic |
| Light rim traceable (blind subagent) | 0 of 3 | 2 of 3 |

- In critique_intent, the ledger stopped completion in three tests on intended items judged not visible
  (five items: R02 T1 two, R06 T1 two, R08 T1 one). The painter answered each one in paint before the next test.
- The arbiter's note on R08 (b2:12): "Nothing unexpected", "No visual tension, unexpected surprise".

## Reading the measures

Regions come from S21's block-in. R02 rebuilt the ruin at about half its height, so its "ruin" region (0.843) mostly
measures sky, and its ruin-darker-than-sky "no" says that, not that the ruin is pale.

## Deviations

D1 (R06, a sitting past the pause), D2 (usage-limit re-runs of five tests), D3 (orchestrator's allowance
count on R03), D4 (RP's annotation is arm-aware), D5 (correction to the wave-2 reason). Wave 2 (R05, R07, R09, R10) was not run (waves.json).

## Not determined here

Whether the ledger changes what is painted: in this wave the composition was set in sitting 1, by the same opening
move in five of six runs across both arms, and the ledger acted on how the painter answered for that composition, not on the composition
chosen.