Resuming a long session (196k events, 2.2k steps, 1.8k tool cards) took
~12s to render and ~800ms to echo one keystroke:
- Every step's timing footer called stepTimingAt, which replayed the whole
event log per footer - O(steps x events) on the initial render.
- pi-tui re-renders every component each frame and relies on per-component
line caches, but ToolCardComponent/ContextCardComponent built throwaway
Text/Markdown instances inside render(width), re-wrapping every settled
card's output on every keystroke.
Replace the per-footer replay with one shared StepTimingTracker per chat
mount (single O(events) cursor over the append-only log), and cache card
rows by width via CardLineCache, dropped by every state mutator and
invalidate().
Measured (tmux 200x50, 196k-event session): resume prompt-ready ~12s -> ~7.6s;
per-keystroke echo ~800ms median -> ~11ms.
session-title already registers a title projection unit; /resume now
reads it instead of scanning logs: live rows from the registry
snapshot, persisted rows from the durable checkpoint row
(cachedSnapshot, zero I/O), and only rows without a usable checkpoint
pay a coldSnapshot — checkpoint plus readFrom tail, written back so the
next scan is metadata-only. Cold reads are bounded by the new
resumeScanConcurrency config; compositions without the cache fall back
to the bounded readTitleSnapshots batch. The TUI overlay mounts the
projection registry, storage, and projection-cache rows over the same
storages root the web surface uses, so checkpoints serve both.
- turn boundary now returns false for an empty admitted batch (claimed
input removed before the wake) instead of opening a turn and spending a
model call on nothing; the step boundary already had the symmetric guard.
- a max-token step stays sticky when steering or injected work continues
the turn: a later completed step no longer downgrades the outcome,
matching the TurnEndReasonMap contract.
- session/queue wire schema accepts the context placement (previously the
zod union rejected injected-context snapshots wholesale and the client
silently dropped the whole frame); schema tests cover all placements.
- headless runs settle at whole-agent idle instead of the first turn/end,
honoring the one-shot idle-to-idle contract.
- flush JSDoc names the real callers (checkpoint policy, goal-session,
teardown, self-flushing consumers); apiproxy zh README loses its stale
duplicate history section; ACP note/README record the delivered error
rejection and turnless-cancelled behaviors.
Making `.sources` a scroll container turned its `padding-left` from spacing
into a correctness constraint. A scroll container clips inline-start overflow
with no way to scroll it back, and `::marker` is right-aligned to the content
edge, so past nine sources the markers rendered as `0.` and `1.` where `10.`
and `11.` belonged.
`searchMaxResults` is an unbounded positive integer, so size the padding in
`em` against the list's own font to hold a three-digit marker. The browser e2e
measures a `999. ` marker in that inherited font and requires the computed
padding to be at least that wide, pinning the room against the widest marker
rather than one fixture's source count.
- gen-cordis-catalog / gen-persistence-catalog / gen-doc-graphs after the
error-reason and steering/message changes (line anchors and event
producer/consumer graph drifted).
- verify-translation-pairing --write --all: every bilingual pair touched
by the PR (agent notes, session/core docs, READMEs) plus the merge.
- queue-actions preserved golden re-recorded against the rebuilt client
(branch disabled on non-completed-turn tails is back in effect).
jsdom resolves no CSS Modules layout and reports scrollHeight === clientHeight
for every element, so the unit test cannot witness the scroll the .sources
max-height produces. Drive it in the real browser instead.
The web-search-round search double now returns 12 provider results with a
title, snippet, and date each. That pins the seam's cap end to end for the
first time in a real composition — searchMaxResults keeps 8, the model-visible
text carries the kept titles and none of the dropped URLs, meta.truncated is
true — and 8 such rows overflow the 320px container. A case after the aria
golden expands the row and asserts 8 <li>, no button, the truncation
indicator, and computed max-height/overflow-y with scrollHeight > clientHeight.
The recorded stream and the aria golden are unchanged: replay is a positional
cursor over the fixture's chunks and the search double is a separate local
endpoint, while the card is collapsed at capture time.
Rows no longer read logs for anything but the batch title fold: the
activity timestamp is a live session's last in-memory event time or the
artifact mtime via the optional sessionPersistence.locate(), falling
back to creation time; the last-turn, route, and goal columns are gone.
Route availability moves to the Enter-time preflight, which already
fully reads and replay-validates the one chosen log. The projectSessions
public API this PR had added to session-query is reverted — the change
is now confined to the TUI package.
The READMEs of ui-primitives and ui-conversation still documented the
removed maxSources prop, DEFAULT_WEB_MAX_SOURCES, CHAT_WEB_MAX_SOURCES,
and the head/tail collapse; the 2026-07-30 frontend note described the
same as current behavior. Restate all of them as the fixed-height scroll
container and cross-link the two notes both ways.
Bound the model-visible/card-visible correspondence: spill-policy's
tools/post-execute replaces an oversized result's content while leaving
presentationMeta whole, so the card's contract is the view it receives,
not the model's context.
Narrow the unit test's name and header comment to what jsdom establishes
(one <ol>, no expand control); jsdom does not resolve CSS Modules layout,
so the scroll geometry belongs to an assembled browser case.
_Kerman