docs(samples): document golden-score (VS/FS=100) reproduction in sample READMEs

[itsloganmann]

What

Adds a golden-score reproduction recipe to the samples docs so a reader can blindly regenerate the published VS 100 / FS 100 numbers.

• samples/README.md: a new "Reproduce the golden scores (VS / FS = 100)" section, alongside the existing task-shape summary, local-serve, run-inference, and run-evaluation sections.
• Each of the five per-sample READMEs (lumina-landing, meridian-dashboard, flux-field, prism-shader, cadence-board): a golden-reproduction block with the sample name pre-filled.

Why

The five calibrated golden outputs score VS 100 / FS 100, but the steps to reproduce that lived only in chat. This documents the exact, verified command sequence (stage the dataset, point a results dir at golden_output, evaluate in a fresh sandbox, analyze) so the numbers can be regenerated from a clean checkout.

Notes

• Docs-only: 157 insertions, 0 deletions across six files. No code, dependency, or schema touched.
• The GUI agent and VLM judge are non-deterministic, so the recipe states the acceptance bar: across three eval runs, at least one run with FS = 100 and at least one with VS > 80.
• Verified on a clean checkout: a fresh venv (pip install -e .) plus a fresh sandbox container per sample reproduces VS 100 / FS 100 for all five.

Test plan

• All six READMEs contain the golden-reproduction recipe.
• Overall README retains task-shape, local-serve, inference, and evaluation sections.
• Per-sample blocks carry the correct sample name in S=.
• No em-dashes in any edited file.
• Diff is additive only (157 insertions, 0 deletions).

[Copilot]

Pull request overview

Docs-only PR adding a "Reproduce the golden scores (VS / FS = 100)" recipe to samples/README.md and to each of the five per-sample READMEs. The recipe stages the sample dataset, copies golden_output/ into a results dir, and runs evaluate + analyze so a reader can regenerate the published golden numbers from a clean checkout.

Changes:

• New section in samples/README.md with the full recipe and 3-run acceptance bar.
• Identical per-sample block (with S= pre-filled) appended to each of the five sample READMEs.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
samples/README.md	Adds section 4 documenting the golden-score reproduction recipe across all samples.
samples/lumina-landing/README.md	Appends per-sample golden-score reproduction block with S=lumina-landing.
samples/meridian-dashboard/README.md	Appends per-sample golden-score block with S=meridian-dashboard.
samples/flux-field/README.md	Appends per-sample golden-score block with S=flux-field.
samples/prism-shader/README.md	Appends per-sample golden-score block with S=prism-shader.
samples/cadence-board/README.md	Appends per-sample golden-score block with S=cadence-board.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.