[P3] render_event_drawer() rebuilds a fresh array + join() on every event_rev bump

[itsmiso-ai]

Ask: Cache the event-drawer expanded-log lines in main.gd render_event_drawer() so the per-render string-array allocation only happens when the event list changes rather than on every event_rev bump.
Expected files: scripts/main.gd

Problem: scripts/main.gd:1829-1848 render_event_drawer() runs whenever sim.event_rev ticks. The collapsed-label branch sets a single event_drawer_label.text and is cheap. The expanded-log branch, however, builds a fresh lines := [] Array, runs six "t%02d %s" % [...] format calls, then calls "\n".join(lines) to assign to event_drawer_log.text. Each event_rev bump (every push_event, which fires on every ambient drop, milestone completion, goal completion, recruit, build queue, build finish, food upkeep tick, break completion, and stale-reservation cleanup) pays that allocation cost. The 2026-07-15 audit (#279 finding 8) flagged this as a future concern; it has not been addressed.

Evidence:

• scripts/main.gd:1829 — func render_event_drawer() -> void: — early-return gate at _drawer_event_rev == sim.event_rev
• scripts/main.gd:1845 — comment Update expanded log with recent history (last 6 events)
• scripts/main.gd:1846 — var lines := [] — array allocated per call
• scripts/main.gd:1847 — lines.append("t%02d %s" % [int(entry.tick), String(entry.get("text", ""))]) — format string per event in the loop
• scripts/main.gd:1848 — event_drawer_log.text = "\n".join(lines) if not lines.is_empty() else "No events yet." — second string allocation from join()
• The gate only prevents rerun when nothing changes; event_rev increments on every push_event, so the churn scales with how busy the colony is
• 2026-07-15 audit (Weekly tech debt audit: windowstead - 2026-07-15 #279 finding 8) called this out: "render_event_drawer() creates string arrays every tick" — still valid

Acceptance:

• render_event_drawer caches the joined drawer text (e.g. cached_text: String + last_rendered_size: int keyed by event list identity) and only recomputes when the last 6 events actually changed
• Inactive colonies (no recent pushes) do zero allocation in render_event_drawer beyond the event_rev compare
• Visual output for the drawer matches the current behaviour byte-for-byte (event_drawer_log.text is the same string for the same event list)
• A simple test scenario: queue a push_event, assert the cached text updated once; idle for a tick, assert no recomputation

[itsmiso-ai]

Not worth doing yet

Listed here so the next audit doesn't re-propose them — each is already covered by an open umbrella or has been explicitly deferred.

• Replace release-please with a hand-rolled version-bumper script. release-please-config.json is the project's single source of truth for release flow; the standing-PR pattern is stable and the team has standardised on it. Replacing it would burn a sprint for negligible return.
• Stand up a real repo-level canary/smoke probe instead of the 15-second headless smoke. The current CI smoke is intentionally tiny; promoting it to a multi-hour scenario would need a runner pool that the project hasn't budgeted for.
• Migrate the web export to a PWA installable manifest + service worker. Web export is a nice-to-have side outcome of Add Web export preset and CI coverage for localStorage persistence path #327; the team has not committed any web-first UX. Adding service-worker plumbing now would be premature.
• Add a Wayland/macOS-native edge-snapping policy beyond DisplayServer.screen_get_usable_rect. edge_snap_cooldown + EDGE_SNAP_THRESHOLD already handle all current supported platforms; a Wayland-specific path would require per-DE testing the project can't justify yet.
• Rewrite main.gd rendering into an MVVM-style view-model/binding layer. The scene-free renderer split (TileRender, WorkerRenderer, LayoutMath) is the agreed-on direction for this codebase; introducing a separate binding framework would explode blast radius without solving a concrete user-visible problem.
• Add automated accessibility (screen-reader, high-contrast) coverage to the dock UI. Windowstead's UX is a tiny desktop companion with no current a11y commitment; coverage would need an a11y test plan that doesn't exist yet.

[itsmiso-ai]

Grooming pass: this issue is ready. The automation note above lists other items that were parked; #361 is not among them. The ask is concrete (cache the joined drawer string, skip recomputation when the last 6 events are unchanged), the target is a single function in scripts/main.gd, and acceptance criteria include a byte-for-byte visual-output check plus a simple push_event/idle test. No blockers detected.