test(presentation): cover real unknown native retention

examples/personal-workspace-browser/research-history-backend.py

@@ -158,6 +158,9 @@ def do_GET(self):
                         metrics["analysis_reads"] += 1
                         language = query.get("output_language", ["original"])[0]
                         record = capture["analysis_readbacks"].get(language)
+                        if "pack_id" in query:
+                            record = next((item for item in capture["analysis_readbacks"].values()
+                                if query == parse_qs(urlparse(item["request_path"]).query)), None)
                         if record and query.get("pack_id", [record["response"]["pack"]["pack_id"]]) == [record["response"]["pack"]["pack_id"]]:
                             return self._send_json(record["response"])
                     elif path == "/api/chat/research-analysis/citation":

examples/personal-workspace-browser/research-history.md

@@ -159,6 +159,30 @@ request identities. Other cited ranges return an explicit unavailable response;
 no transcript or failure stack is substituted. These analysis Goals have no
 Explore logs: their empty histories are preserved, separately from the recent
 real-history cases above.
+The same fixture also freezes the original registered feedback snapshot, whose
+native retention is unknown. Select it through **Evaluation scope / 评估范围**:
+the best valid value remains `0.849544264387785` and the last valid candidate
+remains `0.8485288027628158`, while initial and retained results stay unknown.
+No retained-improvement value is supplied by that snapshot. Its saved 36 process
+judgments and analysis usage are identical to the later declaration; neither
+is rerun or relabeled. Switching scope and Back must restore each snapshot's own
+native results, without borrowing a retained value from the other. The additional
+original/Chinese/English responses are captured through the real read-only HTTP
+owner and pinned independently. This is a reader regression, not independent
+verification of the later provider's retention declaration.
+
+The packaged reader keeps the original snapshot's unknown retention visible
+beside its measured best and final values. The narrow view below is scrolled to
+that same result area; it preserves the same values and on-demand sources.
+
+![Original Graph snapshot, desktop](../../docs/development/assets/runtime-source-analysis/original-native-retention-desktop.png)
+
+![Original Graph snapshot, narrow result area](../../docs/development/assets/runtime-source-analysis/original-native-retention-mobile.png)
+
+The original snapshot includes the exact best/final native-result citations and
+the same native-report-4 excerpt as the later snapshot. Other cited ranges return
+an explicit unavailable response; no transcript, failure stack or other snapshot's
+evidence is substituted. The selected Goal has no Explore log: its empty history is preserved.
 Pass `--fixture apps/presentation/dashboard/src/data/fixtures/edgebench-graph-saved-analysis.json`
 to the same read-only preview to inspect this case. All mutation verbs are denied.
 

examples/personal-workspace-browser/research-real-benchmark.mjs

@@ -61,16 +61,63 @@ async function readNativeResults(page, history, capture, language) {
     assert.equal(await history.locator("[data-analysis-judgment]").count(), 0);
   }
   await progress.locator(".research-progress-basis > summary").click();
+  const name = pack.scope.case_id.split("_")[0];
   for (const width of [1440, 390]) {
     await page.setViewportSize({width, height: 900});
     await progress.scrollIntoViewIfNeeded();
     assert.equal(await page.evaluate(() => document.documentElement.scrollWidth > innerWidth), false);
-    const name = pack.scope.case_id.split("_")[0];
     await page.screenshot({path: resolve(outputDir, `research-real-native-${name}-${language}-${width}.png`), animations: "disabled"});
   }
   await page.setViewportSize({width: 1440, height: 900});
 }
 
+async function readOriginalNativeSnapshot(page, history, capture, language) {
+  const initial = capture.analysis_readbacks.initial_original.response.pack;
+  const current = capture.analysis_readbacks.original.response.pack;
+  const progress = history.locator(".research-progress");
+  const scope = progress.getByRole("combobox", {name: language === "zh" ? "评估范围" : "Evaluation scope"});
+  const selectOriginal = async () => {
+    await scope.selectOption(initial.pack_id);
+    await progress.locator(`[data-pack-id="${initial.pack_id}"]`).waitFor();
+    await history.locator('.research-analysis [data-analysis-judgment]').first().waitFor({state: "attached"});
+  };
+  await selectOriginal();
+  for (const key of ["initial", "retained"]) {
+    const row = progress.locator(`[data-native-point="${key}"]`);
+    assert.match(await row.innerText(), language === "zh" ? /未知/ : /Unknown/);
+    assert.equal(await row.getByRole("button").count(), 0, "An absent native result is not borrowed from another snapshot");
+  }
+  assert.match(await progress.locator('[data-native-point="best_valid"] button').getAttribute("aria-label"), /0\.849544264387785/);
+  assert.match(await progress.locator('[data-native-point="final_valid"] button').getAttribute("aria-label"), /0\.8485288027628158/);
+  assert.equal(await progress.locator("[data-retained-improvement]").count(), 0, "A best measurement does not supply missing retained improvement");
+  assert.equal(await history.locator("[data-analysis-judgment]").count(), 36, "The same original process judgments remain available");
+  const missing = history.locator('[data-analysis-episode="native-feedback-1"]');
+  await missing.locator(":scope > summary").click();
+  for (const detail of await missing.locator(".analysis-judgments > details").all()) {
+    if (!await detail.evaluate(element => element.open)) await detail.locator(":scope > summary").click();
+  }
+  await missing.getByRole("button", {name: /native-report-1/}).first().click();
+  const source = history.locator(".analysis-source");
+  await source.locator(".analysis-error").waitFor();
+  assert.equal(await source.locator("pre").count(), 0, "An uncaptured source is not borrowed from a different snapshot");
+  await source.getByRole("button", {name: language === "zh" ? "关闭原文" : "Close source", exact: true}).click();
+  await progress.locator('[data-native-point="best_valid"] button').click();
+  await source.locator("pre").waitFor();
+  assert.equal(JSON.parse(await source.locator("pre").innerText()).value, 0.849544264387785);
+  await source.getByRole("button", {name: language === "zh" ? "关闭原文" : "Close source", exact: true}).click();
+  await page.goBack();
+  await progress.locator(`[data-pack-id="${current.pack_id}"]`).waitFor();
+  assert.match(await progress.locator("[data-retained-improvement]").innerText(), /\+0\.0323/);
+  await selectOriginal();
+  for (const width of [1440, 390]) {
+    await page.setViewportSize({width, height: 900});
+    await progress.scrollIntoViewIfNeeded();
+    assert.equal(await page.evaluate(() => document.documentElement.scrollWidth > innerWidth), false);
+    await page.screenshot({path: resolve(outputDir, `research-real-graph-retention-unknown-${language}-${width}.png`), animations: "disabled"});
+  }
+  await page.setViewportSize({width: 1440, height: 900});
+}
+
 async function readUnanalyzedNativeResults(browser, collectCoverage) {
   const coverageEntries = [];
   for (const name of ["portfolio", "ad"]) {
@@ -164,6 +211,7 @@ async function readSavedGraphAnalysis(browser, collectCoverage) {
         await page.screenshot({path: resolve(outputDir, `research-real-graph-analysis-${language}-${width}.png`), animations: "disabled"});
       }
       await page.setViewportSize({width: 1440, height: 900});
+      await readOriginalNativeSnapshot(page, history, capture, language);
     }
     const readback = await (await page.request.get(`${owner.base}/fixture/readback`)).json();
     assert.equal(readback.source_unchanged, true);

tests/presentation/test_research_history_fixture.py

@@ -105,6 +105,9 @@ def read(path):
                             "original": "0e4bb4b931784abe1faf8cba49922aeb36f6c96f9e1730fdc77d82e7ebd3da2f",
                             "zh": "cb8fdf1a61915f06168816c82367e9d037bdf49424028cd2793dfe61364def64",
                             "en": "d76b579672d87f8b071d871ecfd5797b986f295b8dc93bc967471c0e1aab0b2b",
+                            "initial_original": "7b68615db558732a9a78b4d83befcb3e31d92840f4d4944c37f5589288fcfcd2",
+                            "initial_zh": "be0621f5d24cc61a9717611caf7b0ef89c7acdfc5eb03da0d763da7a1c87e531",
+                            "initial_en": "c44c78a2404ea55183a42d3ae1c2e08b90920666b49f159d8f841a84d8f5a4da",
                         },
                         "edgebench-portfolio-native-results.json": {
                             "original": "b33ebfcc7c233f8cf87b70e23b3bc76e34d5b91b8fcf11393e532afa9597b717",
@@ -145,6 +148,17 @@ def read(path):
                     assert "solver_usage" not in view and not events
                     for citation in capture["analysis_citations"]:
                         assert read(citation["request_path"]) == citation["response"]
+                    citation = capture["analysis_citations"][0]
+                    if fixture_name == "edgebench-graph-saved-analysis.json":
+                        initial = read(capture["analysis_readbacks"]["initial_original"]["request_path"])["pack"]
+                        initial_native = initial["native_outcomes"]["projection"]
+                        assert initial_native["retained"] is None and initial_native["initial"] is None
+                        assert "retained_improvement" not in initial_native
+                        assert initial_native["best_valid"]["candidate_ref"] == "agent-16"
+                        assert initial_native["final_valid"]["candidate_ref"] == "agent-18"
+                        assert initial["result"] == view["result"]
+                        assert initial["job"] == view["job"]
+                    assert read(citation["request_path"]) == citation["response"]
                     for query in [
                         {"output_language": "fr"}, {"pack_id": "wrong-pack"},
                     ]: