A staging result can keep a window while changing another permanent feature nearby. The local REimagineHome notes make that distinction unusually concrete. For Room 07, they record the window and radiator as kept, while the top of the right pillar was rounded. An outlet and pipe were also added on the left. The input is a real living-room photograph credited as CC0, not an AI-generated interior.

These observations belong to the test-log entry dated 2026-10-21. The REimagineHome page and aggregate results do not fully represent that later round, so the local log and individual score files are the relevant evidence here. Proposed realism ratings for the round remain draft and pending review.

The technical task is to expose the measurements beside those observations without promoting an image-wide statistic into a diagnosis of a particular wall. Run this Python from the project root. It reads the selected local score files and prints the recorded fields:

import json from pathlib import Path fields = ("line_displaced_frac", "line_missing_frac", "edge_ssim", "aligned") for path in sorted(Path("results/reimaginehome").glob("[0-9][0-9]-score.json")): if path.name[:2] not in {"06", "07"}: continue data = json.loads(path.read_text()) block = data.get("architecture") source = "architecture" if not isinstance(block, dict): block = data.get("structure") source = "structure" if not isinstance(block, dict): block = {} source = None row = {"room": path.name[:2], "source": source} row.update({key: block.get(key) for key in fields}) row["framing_pass"] = (data.get("framing") or {}).get("pass") print(json.dumps(row)) 

Actual output from the local files:

{"room": "06", "source": "architecture", "line_displaced_frac": 0.3119, "line_missing_frac": 0.2751, "edge_ssim": 0.7194, "aligned": false, "framing_pass": null} {"room": "07", "source": "architecture", "line_displaced_frac": 0.2549, "line_missing_frac": 0.2776, "edge_ssim": 0.7116, "aligned": true, "framing_pass": true} 

Room 07 passes framing and has aligned measurements. Nevertheless, the visual note records the rounded pillar. Its line fractions summarize detected structural lines across the evaluated region; they do not measure the pillar individually or count changed windows. The preserved window and altered pillar can coexist in the same result.

Room 06 helps keep that interpretation disciplined. Its log says the window and stove were kept, although the room was re-rendered and brightened. Its architecture block explicitly reports failed alignment and an unaligned fallback. The script preserves that status. It also preserves JSON null, and uses the legacy structure block only when an architecture object is unavailable, reporting the source rather than blending fields. That compatibility fallback is separate from the scorer's unaligned fallback, which already lives inside architecture here.

A reviewer following the pillar concern should keep the original and staged image paired with this row. The row explains whether the measurements describe an aligned comparison; the log identifies the feature that prompted concern. Neither supplies a measured distance for the rounding, and the files do not establish how much of the reported line displacement belongs to that feature. Publishing such a distance would add precision the evidence does not contain. The recorded observation is sufficient to flag the image for inspection.

Under the MLS guide, last checked on 2026-10-19, MLS Now prohibits modifying permanent structural features. The pillar observation deserves review under that restriction even though the window was kept and framing passed. These measurements cannot certify compliance.

The methodology remains v0.2 draft. Results are partial, current editor assessments are not blind, and blind evaluation has not started. The value of this extraction is traceability: it connects a particular output to its measurement state while leaving the feature-specific conclusion anchored in the recorded visual observation.