wcag: preserve report dates and correct W3C hand-off

README.md

@@ -1165,9 +1165,16 @@ pages before a second post), load each in headless Chrome, run
 [axe-core](https://github.com/dequelabs/axe-core) over it, and print one row
 per success criterion: how many pages fail it, how many need a look, and the
 rule with the most elements in violation. `report` turns that into the tool's
-own evaluation file, which "Open evaluation" loads with the scope, the sample
-and one assertion per page and criterion filled in, so the evaluator starts
-at the judgement calls rather than at an empty form.
+own evaluation file, which "Open report" on the Overview page loads with the
+scope, the sample and one assertion per page and criterion filled in, so the
+evaluator starts at the judgement calls rather than at an empty form.
+
+In "4. Evaluate sample set", select the pages under "Add results for samples"
+(or "Select all"), then expand "Show samples to enter individual results"
+beneath a criterion to see its imported findings. The "Entire sample set"
+outcomes remain "Not checked" for the evaluator to complete, so the tool's
+overall progress starts at 0 even when page findings are present. Save with
+"View report" > "Download report (JSON)".
 
 **Nothing is ever marked passed.** axe can prove a failure — an image with no
 alternative fails 1.1.1 wherever it is — but "no rule fired" proves nothing

bin/wcag.ts

@@ -217,7 +217,7 @@ async function runReport(argv: string[]): Promise<number> {
 
   const assertions = (evaluation.auditSample as unknown[]).length;
   const sampled = ((evaluation.selectSample as { structuredSample: unknown[] }).structuredSample).length;
-  out(`${outFile}: ${sampled} pages in the sample, ${assertions} assertions. Open it at ${REPORT_TOOL_URL} with "Open evaluation".`);
+  out(`${outFile}: ${sampled} pages in the sample, ${assertions} assertions. Open it at ${REPORT_TOOL_URL} with "Open report" on the Overview page. Run \`wcag open\` for the full steps.`);
   return 0;
 }
 

src/wcag.ts

@@ -33,7 +33,7 @@
  *   `reportFindings`, a `Webpage` subject per sampled page whose id is its
  *   URL, and an `Assertion` per page and criterion with `test` set to the
  *   tool's own criterion ids (`WCAG22:non-text-content`). A file of that
- *   shape goes through the tool's "Open evaluation" rather than the beta
+ *   shape goes through the tool's "Open report" rather than the beta
  *   assertion import, which is the path that restores the sample as well.
  */
 
@@ -1363,7 +1363,7 @@ const describe = (outcome: CriterionOutcome, page: PageResult, report: Report):
 
 /**
  * The report tool's own evaluation file, with the sample and the automated
- * results filled in. Open it in the tool with "Open evaluation".
+ * results filled in. Open it in the tool with "Open report".
  */
 export function toEvaluation(report: Report, options: EvaluationOptions = {}): Record<string, unknown> {
   const { wcagVersion: version } = report;
@@ -1456,7 +1456,9 @@ export function toEvaluation(report: Report, options: EvaluationOptions = {}): R
     reportFindings: {
       documentSteps: [{ '@id': '_:about' }, { '@id': '_:defineScope' }, { '@id': '_:exploreTarget' }, { '@id': '_:selectSample' }],
       commissioner: options.commissioner ?? '',
-      date,
+      // The tool reads reportFindings.date['@value']; an untyped string is lost.
+      // Use a calendar date for the report and keep precise assertion timestamps.
+      date: { type: 'http://www.w3.org/TR/NOTE-datetime', '@value': date.slice(0, 10) },
       evaluator: options.evaluator ?? '',
       evaluationSpecifics: '',
       summary,
@@ -1483,11 +1485,16 @@ export const OPEN_STEPS = `The WCAG-EM Report Tool is a web page with no command
   1. wcag audit https://example.org --pages 8           writes wcag-report.json
   2. wcag report wcag-report.json                       writes evaluation.json
   3. open ${REPORT_TOOL_URL}
-  4. "Open evaluation" in the menu, choose evaluation.json
+  4. "Open report" on the Overview page, choose evaluation.json
+  5. In "4. Evaluate sample set", select the pages under "Add results for samples"
+     (or "Select all"), then expand "Show samples to enter individual results"
+     beneath a criterion to see its imported findings.
 
 Steps 1 (scope) and 3 (sample) are then filled in, and step 4 (audit) holds
 one assertion per page and criterion: "failed" where axe proved a failure,
 "cannot tell" where it found something to look at or only checked a part.
 Nothing is marked "passed" — that is the evaluator's call, made in the tool.
+The "Entire sample set" outcomes stay "Not checked" for the evaluator to finish,
+so the report's overall progress remains 0 until those outcomes are recorded.
 The tool keeps the evaluation in the browser and saves it back out as JSON
-from step 5.`;
+with "View report" > "Download report (JSON)".`;

test/wcag.test.ts

@@ -432,6 +432,15 @@ describe('toEvaluation', () => {
     expect(evaluation.reportFindings.evaluator).toBe('Anthony');
   });
 
+  it('preserves the evaluation date in the typed value the report tool imports', () => {
+    // W3C's importer reads reportFindings.date['@value']; a plain string
+    // silently becomes an empty date. Use YYYY-MM-DD for the report date.
+    expect(evaluation.reportFindings.date).toEqual({
+      type: 'http://www.w3.org/TR/NOTE-datetime',
+      '@value': '2026-09-13',
+    });
+  });
+
   // The tool derives a sampled page's id from the URL in its description and
   // finds it again by title, so both have to be exactly this.
   it('puts every loaded page in the structured sample with its URL as id and description', () => {