Document actual report previews and audit failure walkthroughs

README.md

@@ -41,6 +41,8 @@ The bundled math demo produces this self-contained report without an account, AP
 
 The screenshot is generated from `rigorgraph demo --scenario math`. The report keeps research text in its original language while its interface can switch between English, Traditional Chinese, Simplified Chinese, and Japanese.
 
+Open the [valid and deliberately invalid report walkthrough](docs/REPORT_WALKTHROUGH.md) to inspect actual exported examples before installing.
+
 ## Quick start (three minutes)
 
 RigorGraph requires Python 3.11 or newer. Install the published 1.0.1 package; the product status remains public beta.

docs/REPORT_WALKTHROUGH.md

@@ -0,0 +1,61 @@
+# Open a report, then make an audit fail
+
+These three self-contained reports were generated from the repository's built-in
+demo fixtures on September 5, 2026:
+
+- [Valid math demo](https://f0909172434.github.io/examples/rigorgraph/math.html)
+  — audit `PASS`, process exit 0.
+- [Invalid promotion demo](https://f0909172434.github.io/examples/rigorgraph/invalid.html)
+  — audit `FAIL`, process exit 1, with `RG_EVIDENCE_TYPE_MISSING` and
+  `RG_ACCEPT_EVIDENCE_UNCHECKED`.
+- [Evidence changed after review](https://f0909172434.github.io/examples/rigorgraph/tampered.html)
+  — a copy of the valid demo with one sentence appended to its evidence file;
+  audit `FAIL`, process exit 1, with `RG_HASH_MISMATCH`.
+
+They contain synthetic demo claims and review records, not a new independent
+review of a real research result. The report itself includes its data and viewer;
+it can be saved and opened offline. Hosted copies are convenience previews.
+
+## Five-minute inspection
+
+1. Open the valid report and select a claim. Follow its evidence and review record.
+2. Switch the interface language. User-authored research text stays in its
+   original language; the controls support English, Traditional Chinese,
+   Simplified Chinese and Japanese.
+3. Open the invalid report, select **Open gaps**, and read both audit codes.
+4. Explain why a numerical scan cannot satisfy a formal-proof evidence
+   requirement. A workflow `PASS` also does not establish scientific truth.
+5. Open the changed-evidence report and select **Open gaps**. The stored evidence
+   hash no longer matches the file bytes even though the old review record remains.
+
+## Reproduce locally
+
+```sh
+rigorgraph demo report-math --scenario math
+rigorgraph audit report-math --json
+rigorgraph demo report-invalid --scenario invalid
+rigorgraph audit report-invalid --json
+```
+
+The last command intentionally exits nonzero. Each demo directory contains
+`rigorgraph-report.html`; do not edit its embedded JSON by hand to change a result.
+Use the [end-to-end workflow](WORKFLOW.md) for a record-by-record example and a
+deliberate file-tampering check.
+
+To reproduce the third case in a disposable copy of `report-math`, append a
+sentence to `evidence/odd-sum-proof.md`, then audit it. Generate a new report with
+`rigorgraph report PATH --output PATH/rigorgraph-report.html`; the default output
+is relative to the current working directory, not automatically the project path.
+
+## Snapshot identity
+
+The hosted files were copied from the generated reports without editing them:
+
+| Report | SHA-256 |
+|---|---|
+| math.html | `192f730f65d35cff9b9acde272deda398ea26664beaf9ff41fc031f5ed9b9800` |
+| invalid.html | `558b056b63cc739046f9bf27eef798ae7c01db6762a679aa3514149c6313c154` |
+| tampered.html | `c45bcf68f5f2819c440930b28cc8abe0ffe0182ea603b0300a6a8b048e8ce6ca` |
+
+Demo timestamps and other record metadata can change on a fresh run. These hashes
+identify the published snapshots; they are not expected hashes for every rerun.