sorokin.oleg.edu/learnharness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 6 Wiki Activity

fix(pdf): split pagesRead from pagesConsulted — the honest badge count

Browse Source 
Live-smoke follow-up the user approved (2026-09-19): the GEB read showed
"78 pages consulted" when the reader genuinely READ ~12 - pdf_search hit
snippets inflated the count. PdfResult now carries both:

- pagesRead: extract/render output entered the loop - the honest count
  the badge and brief footer show ("read N pages", "Pages read: ...").
- pagesConsulted: every page whose content entered the loop, search-hit
  context lines included - unchanged, still the citation audit set (the
  reader's instructions still say "Pages consulted: N, N, ..." and the
  footer still lists it when it exceeds the read set).

The unread-finish gate deliberately stays on CONSULTED: search hits are
genuine consultation (anti-fabrication duty), so a search-only brief is
honest and must deliver - pagesRead is a reporting refinement, not a
delivery gate. Pinned by the search-only faux test (read [], consulted
[1], delivered).

UI tolerates pre-split persisted records via ?? fallbacks (old steers
lack pagesRead).
This commit is contained in:
amosblomqvist
2026-09-19 16:52:09 +03:00
parent 7a3e19e95f
commit 1b9122c88b
8 changed files with 57 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/agent/subagents/pdf-reader.ts
+20 -5
View File
@@ -172,7 +172,7 @@ export async function runPdfReaderAgent(
// for renders — the brief never starts on a broken source.
const openError = await probeReadable(pdfPath);
if (openError) {
return { path: pdfPath, question: input.question, brief: `PDF failed: ${openError}`, pagesConsulted: [] };
return { path: pdfPath, question: input.question, brief: `PDF failed: ${openError}`, pagesRead: [], pagesConsulted: [] };
}
const chromeReason = headless.unavailableReason();
// Chrome missing does NOT fail the run: text-side tools still work; the
@@ -192,6 +192,10 @@ export async function runPdfReaderAgent(
let messages: Message[] = [{ role: "user", content: prompt, timestamp: Date.now() }];
/** Every page the reader extracted or rendered (the citation set). */
const consulted = new Set<number>();
/** Pages genuinely READ — extract or render output entered the loop.
* Search-hit context lines consult but don't read (the "78 pages
* consulted" overstatement split, sources v1 2026-09-19). */
const read = new Set<number>();
let finalText = "";
let nudged = false;
@@ -211,6 +215,10 @@ export async function runPdfReaderAgent(
.map((c) => c.text)
.join("")
.trim();
// The gate stays on CONSULTED, not read: search-hit context
// lines are genuine consultation (anti-fabrication duty) — a
// search-only brief is honest. pagesRead is the UI's honest
// count, not a delivery gate.
if (consulted.size === 0 && !nudged) {
nudged = true;
messages = [
@@ -219,7 +227,7 @@ export async function runPdfReaderAgent(
];
continue;
}
return { path: pdfPath, question: input.question, brief: finalText || "(no brief text)", pagesConsulted: [...consulted].sort((a, b) => a - b) };
return { path: pdfPath, question: input.question, brief: finalText || "(no brief text)", pagesRead: [...read].sort((a, b) => a - b), pagesConsulted: [...consulted].sort((a, b) => a - b) };
}
const toolResults: ToolResultMessage[] = [];
@@ -239,7 +247,10 @@ export async function runPdfReaderAgent(
opts.onActivity?.(`pdf_extract: ${spec}`);
const r = await tools.pdf_extract.execute!({ pages: spec }, { toolCallId: call.id });
toolResults.push(toolText(call.id, call.name, String(r)));
for (const p of specNumbers(spec, Infinity)) consulted.add(p);
for (const p of specNumbers(spec, Infinity)) {
consulted.add(p);
read.add(p);
}
} else if (call.name === "pdf_search") {
const query = String(args.query ?? "");
const contextLines = numOr(args.context_lines, 3);
@@ -268,7 +279,10 @@ export async function runPdfReaderAgent(
const wanted = specNumbers(spec, RENDER_BATCH_CAP);
opts.onActivity?.(`pdf_render: ${spec} @${dpi}dpi`);
const pages = await headless.renderPdfPages(pdfPath, wanted, { dpi });
for (const p of pages) consulted.add(p.page);
for (const p of pages) {
consulted.add(p.page);
read.add(p.page);
}
if (pages.length === 0) {
toolResults.push(toolText(call.id, call.name, "ERROR: no pages rendered (spec matched nothing?)"));
} else {
@@ -315,7 +329,7 @@ export async function runPdfReaderAgent(
}
} catch (err) {
const msg = err instanceof Error ? err.message : String(err);
return { path: pdfPath, question: input.question, brief: `PDF failed: the reader crashed — ${msg}`, pagesConsulted: [...consulted].sort((a, b) => a - b) };
return { path: pdfPath, question: input.question, brief: `PDF failed: the reader crashed — ${msg}`, pagesRead: [...read].sort((a, b) => a - b), pagesConsulted: [...consulted].sort((a, b) => a - b) };
}
// Iteration limit: deliver whatever was honestly consulted (never invention).
@@ -323,6 +337,7 @@ export async function runPdfReaderAgent(
path: pdfPath,
question: input.question,
brief: finalText || `PDF failed: iteration limit (${maxIterations}) reached without a brief.`,
pagesRead: [...read].sort((a, b) => a - b),
pagesConsulted: [...consulted].sort((a, b) => a - b),
};
}