# RGI-quick: results

Protocol: `protocol.json` (frozen, read-only). Measures: `results.json`. Finals beside S21's block-in: `finals.jpg`.
Two runs per arm: this describes four paintings; it does not establish an effect.

## Measured against the pre-registered rule

| run | arm | final | geometry (edge corr. with S21; escaped < 0.45) | snow | glow | lit brook | escaped |
|---|---|---|---|---|---|---|---|
| RG-L1 | rgi_late | clean:1 | 0.415 | 0.760 | no | yes | geometry, glow |
| RG-L2 | rgi_late | quiet:1 | 0.527 | 0.784 | yes | yes | none |
| RG-E1 | rgi_early | sun2:1 | 0.447 | 0.573 | yes | yes | geometry, high key |
| RG-E2 | rgi_early | dark:6 | 0.417 | 0.527 | yes | no | geometry, lit brook, high key |

Reference: all twelve SR1 and SR2 finals 0.52 to 0.91; studies built on S21 0.74 to 0.95; unrelated wall studies 0.04 to 0.10.

**Verdict under the rule: BROKEN.** RG-L1, given RGI only after two sittings of its own plan, escaped geometry and one other
default (glow). The early arm also escaped geometry (both runs), so the test is not inconclusive.

## What the paintings show

- All four moved the brook from S21's lateral S across the frame to a channel that enters from the bottom edge and recedes
  to the ruin. RG-L2 did this too; it scores 0.527 because S21's lateral band also stays in the picture beside the new channel.
- All four kept S21's ruin (shape and place) and horizon, and all four painters' first stated questions (two of them under
  RGI from before the first stroke) set a low sun behind the ruin with the brook as its path of light.
- The late runs made the change after RGI arrived: RG-L1 "invented a new frozen brook that comes in wide at the bottom
  edge"; RG-L2 "opened a wide dark channel rising from the bottom edge". Both had painted S21's lateral brook in sittings 1-2.
- Margins are narrow: RG-L1 sits 0.035 under the threshold, RG-E1 0.003.
- Region measures (glow, lit brook, snow) read S21's regions; where the brook moved, "lit brook" describes S21's brook region.

## The light, measured afterwards (rgi/light.py, not pre-registered)

Brightest 3% of S21's sky region, its x-centroid against the ruin's (dx), plus the glow measure. RG-L1: dx 0.088, no glow.
RG-L2: dx 0.403 (the brightest sky sits left, the glow is faint). RG-E1: dx 0.056, glow. RG-E2: dx 0.063, glow. For
reference, SR1 and SR2 finals: dx 0.007 to 0.22, except SR1 R03 (0.559); S21's block-in 0.326; P4 (overcast) 0.427.
The idea held in all four first questions (a low sun behind the ruin); in the finals it held in the early pair and
loosened in both late runs.

## Not determined here

Whether the move came from timing or from permission: RGI (with the bleeding-ulcer note and "nothing you have painted is
protected") changed what the painters were allowed to undo, and late and early arms behaved alike. All four arrived at the
same new move (a receding channel from the bottom edge), which may itself be a default.