3 questions from the Opus spec run (unblock the build) + 5 standing decisions. Recommendations pre-starred where clear. Add notes on each slide.
← → or tap arrows to navigate★ = my recommendation
Depth/Expand · Decision 1 of 3
Where does the expand trigger live in v1?
Context
Agent 2 designed triggers on both the map node and the citation chip in the answer stream. They sync visually — hovering one highlights the other.
A
Both — map node + citation chip
Full spec
Triggers on both surfaces. Hover-link syncs them. The two panels read as one instrument.
B
Map node only in v1
Faster
Citation chip trigger added in v1.1. Risk: map feels disconnected from answer.
C
Citation chip only in v1
Fastest
Answer-centric. Map shows boundary but isn’t interactive yet.
—
Shelve it
Park for now
Not blocking right now. Revisit next session.
Notes
Depth/Expand · Decision 2 of 3
DOCX export — client or server?
Context
ShareMenu.tsx has Teams/Outlook/clipboard/PDF. DOCX is missing entirely. Client-side: docx.js in-browser (~3h). Server-side: new worker route (~6h, cleaner for large outputs).
A
Client-side (docx.js) now
~3h
No new worker route. Migrate server-side in M4 alongside PDF — they share the same document model.
B
Server-side worker route
~6h
Better for large outputs. Clean parity with PDF. Extra 3h upfront.
—
Shelve it
Park for now
Not blocking right now. Revisit next session.
Notes
Depth/Expand · Decision 3 of 3
Expansion limits — right numbers?
Context
5 expansions/answer max, $40/day per-user cap. At Opus pricing: 5 expansions = ~$2.00 worst case. Both are env vars — no redeploy to change them.
A
Ship 5/$40, tune after real use
Conservative
Right for day one. Adjust via env var once you see real usage. No redeploy.
B
Higher — 10 expansions, $100/day
Power users
$100/day is the existing global cap. 10 expansions = ~$4.00 worst case per answer.
C
Lower — 3 expansions during beta
Cost-safe
Conservative while in maintenance mode. Raise after wider rollout.
—
Shelve it
Park for now
Not blocking right now. Revisit next session.
Notes
Standing Decisions · 1 of 5
Floor plan analysis — where to start?
Context
NCC floor plan upload + compliance check. Reuses existing image upload + corpus search infrastructure.
A
MVP — Floor Plan + Quick NCC Check
~2 days · ~$0.25/query
Image upload → visual analysis → plain-language NCC flags. Build to extend cleanly to B — no throwaway architecture.
B
Structured Analysis — Room Schedule
~5 days · ~$0.85/query
Structured JSON extraction, per-room deep-dive, pass/fail per room. Build after A proves demand.
Current: API keys, no expiry, no MFA, no audit log. 8 users have keys ready but aren’t active. CF Access = email OTP, session expiry, audit log. Free on CF plan. Needs ~2h in CF dashboard.
A
Set Up CF Access Now
~2h · Free
Life-safety corpus. Do it before the remaining 8 users onboard. Email OTP is also better UX than pasting keys.
B
Wait Until More Users Onboard
Deferred
Only 2 of 10 active. But each additional key is another risk while this is deferred.
—
Shelve it
Park for now
Not blocking right now. Revisit next session.
Notes
Standing Decisions · 3 of 5
Checklist Phase 2 — which features, what order?
Context
v1 shipped with Pass/Fail/N/A, notes, MANDATORY/GUIDANCE, D1 persistence. Phase 2 adds 4 independent features. ~12 days total.
A
All four, in priority order
~12 days
Photos (3d) → Signatures (2d) → PDF Report (3d) → Corrective Actions (4d). QLD Form 16 requires photographic evidence and inspector signature.
B
Photos + Signatures first
~5 days
Ship the evidence layer fast, revisit PDF + corrective actions next session.
C
Defer Phase 2
Focus elsewhere
Park checklists until Option C + depth/expand is live.
—
Shelve it
Park for now
Not blocking right now. Revisit next session.
Notes
Standing Decisions · 4 of 5
User guide — what format?
Context
8 users have keys but aren’t active yet. A guide is needed before wider rollout. Onboarding tour exists but no comprehensive written reference.
A
Web page on cockpit site
~3h · Linkable
Build once, always current. PDF generated from the web page when needed — not a separate doc. In-app help deferred until UI stabilises.
B
All three — web + PDF + in-app
~10h
Maximum coverage. Over-engineering while UI is still changing (Option C is being built now).
—
Shelve it
Park for now
Not blocking right now. Revisit next session.
Notes
Standing Decisions · 5 of 5
Research Brief rollout — already resolved.
Context
The deployed worker already has RESEARCH_BRIEF_ENABLED=“all”. Option C removes the Brief pill entirely — the 3-stage pipeline becomes the auto-format for complex queries. No separate rollout decision needed.
✓
Acknowledged — no action needed
Already “all” in the deployed worker. detectComplexQuery handles depth automatically as part of Option C. Move on.