phase: 113_source_chip_quality
Build and Push Containers / build-and-push-app (push) Successful in 2m2s
Build and Push Containers / build-and-push-db (push) Successful in 15s

All gates green — no defects found; this pass was verification only.

**Phase 113 final verification pass — report**

- Verified (no code changes needed): `select_documents_tiered` cited/related tiering + `select_documents` wrapper, `TurnPlan.related_docs`, `ChatDoneEvent.related` (additive, old payloads parse), `appendRelated` UI row (`.related-doc`, never `.source-chip`), done-frame + restore-path wiring, two settings with validators, `.env.example` entries
- `uv run pytest --cov=app --cov-report=term-missing` → 2422 passed, app/ coverage **99%** (>90% gate)
- `uv run pytest tests/e2e/test_source_chip_quality.py -v --no-cov` (isolated) → 2 passed
- Regression E2E `test_retrieval_quality.py` + `test_honest_deflection.py` + `test_chat_rag.py` + `test_sources_midstream_bug.py` → 17 passed
- `uv run ruff check . && uv run pyright` → clean (0 errors); `bash .agents/validate.sh` → "validation OK"

Completion criteria:
1. Single-doc question → exactly one `.source-chip` (E2E): ✅ passed
2. Weak 2nd doc only in de-emphasized related row, never `.source-chip` (unit + E2E): ✅ passed
3. Deflected turn → zero citation chips, weak hits in related row: ✅ passed
4. Full suite green, coverage >90%, isolated E2E green, lint/types clean: ✅ passed
5. `--no-gpg-sign` commit + phase dir move: left to harness per pass rules (task files already in `complete/`)

No deviations. Next pending phase: `114_embed_question_length`.
This commit is contained in:
2026-09-15 03:11:05 -04:00
parent 1374faf136
commit 97d663d16d
31 changed files with 2370 additions and 52 deletions
+65
View File
@@ -1395,6 +1395,51 @@ function appendSources(wrap, sources) {
}
}
/* Phase 113 (task 02): the DE-EMPHASIZED related-docs row — the done
* frame's second tier (phase 113 task 01): documents that scored but
* did not clear the usefulness bar. It must never read as a citation:
* the links carry the .related-doc class (NOT .source-chip — the
* citation surface stays appendSources' alone) while behaving exactly
* like the chips — the same documentUrl href (the /document.html escape
* hatch) and the same left-click → same-page modal (phase 26). The
* labeled row ("Nearby docs, in case:") joins the bubble's .msg-meta
* family and stacks below the citation row in the .msg-body flex gap.
* app.js appends it LAST among the meta rows — after appendTuneButton /
* appendSaveAsDocButton / appendRetryButton claimed the FIRST
* .msg-meta row — so a deflected turn (no citation row) never lets a
* meta action join this row. Empty/absent input → no DOM at all
* (pre-phase saved chats carry no `related` — the row is simply absent).
*/
function appendRelated(wrap, related) {
if (!related || !related.length) return;
const body = wrap.querySelector(".msg-body");
const row = document.createElement("div");
row.className = "msg-meta related-docs";
row.setAttribute("role", "list");
row.setAttribute("aria-label", "Nearby docs, in case");
const label = document.createElement("span");
label.className = "related-docs-label";
label.textContent = "Nearby docs, in case:";
row.appendChild(label);
for (const s of related) {
const docLabel = `${s.source}/${s.path}`;
const link = document.createElement("a");
link.className = "related-doc";
link.setAttribute("role", "listitem");
link.href = documentUrl(s.source, s.path, "/"); // back → the chat page
link.addEventListener("click", (e) => {
e.preventDefault(); // no new tab (phase 26) — the modal takes over
e.stopPropagation();
openDocumentModal(s.source, s.path, link);
});
link.textContent = docLabel;
link.title = docLabel; // full path as the native tooltip (chip pattern)
link.setAttribute("aria-label", docLabel); // the accessible name is the full path
row.appendChild(link);
}
body.appendChild(row);
}
/* "Maybe try:" chips under a deflected bubble (honesty gate, phase 04,
shared component + one-tap submit, phase 05). The group is accessible
(role=list + aria-label) and wraps cleanly at every width. */
@@ -1565,6 +1610,11 @@ function renderStoredMessage(m) {
// — no button (the live stop path adds none either).
if (!m.stopped) appendSaveAsDocButton(wrap, m.text);
if (m.stopped) appendStoppedNote(wrap); // phase 48: the stop marker restores
// Phase 113 (task 02): the related tier restores with the bubble
// (LAST — after the meta-row claimers, exactly like the live done
// path). Pre-phase records carry no `related` → appendRelated
// no-ops and the row is simply absent (graceful).
appendRelated(wrap, m.related);
lastBrainWrap = wrap; // phase 49: the LAST restored brain bubble wins
}
@@ -2433,6 +2483,15 @@ async function runTurn(text, { reask = false } = {}) {
appendMaybeTry(wrap, ev.suggestions);
}
appendSources(wrap, ev.sources);
// Phase 113 (task 02): the cited tier stays the citation
// surface (the chips above); the related tier (scored docs
// under the usefulness bar — and the weak hits of a DEFLECTED
// turn, whose `ev.sources` is empty → zero chips) renders as
// the de-emphasized labeled row, appended LAST among the meta
// rows below (after the appendTuneButton/SaveAsDoc/Retry
// claimers took the first .msg-meta row, so an action never
// joins the related row — a deflected turn with related docs
// still gets its own meta row for the buttons).
// Thinking-without-answer (reasoning can exhaust max_tokens): the
// bubble gets the empty-answer fallback — what the user saw is
// what gets persisted.
@@ -2457,10 +2516,16 @@ async function runTurn(text, { reask = false } = {}) {
tools: toolAcc.length ? toolAcc : undefined,
deflected: !!ev.deflected,
sources: ev.sources,
// Phase 113 (task 02): the related tier persists with the
// turn (undefined drops the key from the JSON — the house
// optional-meta pattern), so the restore path re-renders the
// row exactly as it looked live.
related: ev.related?.length ? ev.related : undefined,
suggestions: ev.suggestions,
}, leavePartialIndex);
lastBrainWrap = wrap; // this bubble is now the last brain answer
markLastRetryable(); // phase 49: the Retry button is last-bubble-only
appendRelated(wrap, ev.related); // phase 113 (task 02) — LAST meta row
} else if (ev.type === "tool_result") {
// Phase 95 (A15 extension, task 02): the truncation the LLM is
// told about is told to the USER. One frame per truncated read,
+45
View File
@@ -692,6 +692,51 @@ details.thinking .thinking-text ul { margin: 0 0 0.5rem; }
}
.source-chip:hover { background: var(--brand-soft); text-decoration: underline; }
/* Phase 113 (task 02): the related-docs row — the SECONDARY tier of
scored docs (phase 113 task 01's usefulness bar demotes the
sub-floor hits out of the citation surface; on a deflected turn the
weak hits live HERE and the chip row is absent). It must never read
as a citation: smaller mono links, DASHED border (the solid brand
pill is the citation look), transparent fill, muted --ink-soft text
— 8.6:1 on --bg (verified ≥4.5:1, WCAG 2.1 AA), palette variables
only (phase-92 zero-literal: no new hue, the monochrome theme grays
it automatically). The row joins .msg-meta (the .msg-body flex gap
stacks it below the citation row — no extra margin) but the meta
ACTIONS never do: app.js appends it last, after the first-row
claimers. :focus-visible via the global 3px outline rule. The hover
is deliberately flat — NO background swap (the chip's brand-soft
hover is the citation affordance): only the ink-soft → ink step-up
plus the underline. */
.related-docs-label {
font-size: 0.68rem;
font-weight: 600;
text-transform: uppercase;
letter-spacing: 0.05em;
color: var(--ink-soft);
}
.related-doc {
display: inline-flex;
align-items: center;
font-family: var(--mono);
font-size: 0.7rem;
color: var(--ink-soft);
background: transparent;
border: 1px dashed var(--line);
border-radius: 999px;
padding: 0.12rem 0.55rem;
text-decoration: none;
max-width: 100%;
/* min-width: 0 so the nowrap pill ellipsizes (same as the chips) */
min-width: 0;
overflow: hidden;
text-overflow: ellipsis;
white-space: nowrap;
}
.related-doc:hover {
color: var(--ink); /* ink on --bg 16.7:1 — the flat secondary hover */
text-decoration: underline;
}
/* "Maybe try" chips under a deflected bubble (phase 04). Unlike the
onboarding row (which scrolls horizontally on mobile), this group wraps
at every width: the chips are the actionable follow-up, not decoration.