🗺️ Situation Report — Where We Are
✅ Five blocks shipped
Blocks A (bug fixes), B (admin/infra), D (features), E (corpus expansion), and G (UX) are all complete and deployed. Core product is stable.
⚠️ Block C — corpus work, gated chain
C1→C2→C3 must run in strict sequence with a Grant checkpoint between C2 and C3. C4 (depth review) is standalone. This is the only block that touches the live corpus.
📋 Checklist v2 — clean slate
4 features already approved by Grant (2026-06-17): PDF export, photo evidence, signed/locked checklists, Docker deploy. Zero risk to existing search functionality.
📋
Section 1 — What's Left
Two remaining blocks · 8 tasks · ~14–18 hours total wall clock
5/7
Blocks done
2
Remaining
Already shipped
✓ Block A — Bug Fixes & Polish
✓ Block B — Admin & Infra
✓ Block D — Feature Improvements
✓ Block E — Corpus Expansion
✓ Block G — User Experience
C
NCC Corpus Enrichment
Search & Corpus Quality · Strict gated chain C1→C2→C3
~6–8h
Wall clock
~$28.90
API cost
Execution Chain (strictly sequential)
C1 Design → C2 NCC Pilot → ⚡ Grant checkpoint → C3 Full rollout
C4 is standalone — can run anytime
C1
Enrichment Design + Test Harness
Design the metadata enrichment schema for NCC clauses. Build before/after benchmark harness to measure search quality impact.
standalone no corpus write
~2h
gate: none
C2
NCC-Only Rollout (1,806 clauses)
Enrich NCC corpus only. Run benchmark comparison. Report quality delta to Grant. This is the pilot before touching all 5,600+ clauses.
corpus write: NCC only checkpoint follows
~3h
needs C1
⚡
Grant checkpoint required — C2 delivers benchmark report. Grant reviews before C3 starts. C3 cannot run without explicit approval.
C3
Full Corpus Enrichment (5,600+ clauses)
Rolls out enrichment to all corpus clauses if C2 benchmark proves positive. Largest single API cost task in Phase 2.
corpus write: ALL standards ~$20 API cost
~3h
needs C2 + Grant ✓
C4
Response Depth Review (5-agent report)
Standalone deep review of whether response depth is being artificially limited. 5 parallel reviewer agents, produces a report. No corpus writes.
standalone no corpus risk 5-agent report
~1h
independent
F
Checklist v2 + Deployment
4 features · All decisions approved 2026-06-17 · Zero corpus risk
~8–10h
Wall clock
~$0.50
API cost
Decision status — all approved
✓ HTML template for PDF ✓ Keep photos forever ✓ Lock after signing ✓ Void/unlock escape
F1
PDF Export (HTML template approach)
Generate printable PDF-ready HTML from checklist data. Uses server-side template rendering, not a PDF library. Approved: HTML template.
new component zero search risk
~3h
independent
F2
Photo Evidence Attachment
Attach photos to checklist line items as evidence. Photos stored permanently (never purged). Decision: keep photos forever.
new feature R2 storage
~2.5h
independent
F3
Signed & Locked Checklists
Sign-off flow that locks the checklist. Decision: lock after signing with void/unlock escape hatch for corrections. Signature stored as evidence.
new feature lock mechanism
~2.5h
independent
F4
Docker / On-Prem Deployment Prep
Package app for Docker deployment. Enables on-premise installs for enterprise clients who can't use cloud-hosted version.
infra Docker on-prem path
~1.5h
independent
Why two very different blocks? Block C is high-value corpus work that could meaningfully improve every answer the system gives — but it carries real risk. Checklist v2 is pure feature shipping: new capabilities in isolated components with no shared state. The question is sequencing: which block earns the next session?
⚠️
Section 2 — Risk Analysis
Corpus enrichment risk vs feature delivery risk · Side-by-side comparison
C
Block C — Corpus Enrichment Risk
HIGH RISK
Risk Factor Level
Search quality degradation
HIGH
Blast radius if C3 goes wrong
ALL 5,600+
Rollback complexity
MEDIUM
Risk of accidental C3 trigger
LOW
API cost exposure
~$28.90
Mitigations in place
🧪 C1 builds before/after benchmark harness — quantifies quality impact before any writes
🔬 C2 pilots NCC-only (1,806 clauses) — controlled test before touching full corpus
⚡ Hard checkpoint: Grant must review C2 benchmark before C3 can start — no auto-proceed
⏪ Rollback: re-upload original embeddings from backup. Restores pre-enrichment state.
F
Checklist v2 — Feature Delivery Risk
LOW RISK
Risk Factor Level
Impact on existing search/query
NONE
Blast radius if feature fails
ISOLATED
Rollback complexity
EASY
Decision ambiguity
NONE
API cost exposure
~$0.50
Why this is genuinely low risk
🏗️ New components (PDF renderer, photo uploader, sign-off UI) are isolated from search pipeline
✅ All decisions already approved 2026-06-17 — no ambiguity, no mid-task pivots
🔄 Worst case: a feature doesn't work → users continue existing workflow, zero regression
💰 $0.50 total API cost — basically free to run relative to Block C
The core tension: Block C has the highest potential upside (better answers for every query) but the highest risk (metadata injection into 5,600+ embeddings could degrade global search quality). The gating chain (C1→C2→checkpoint→C3) is designed to make the risk manageable, but the risk is real. Block F is pure upside with no systemic risk — the question is whether delaying it costs users anything.
🎯
Section 3 — Three Options
Click a card to select your preferred approach · Options are mutually exclusive
A
★ Recommended
Block C Only
Tackle the corpus risk in isolation. Everything else is already shipped — give C the focused attention it deserves.
Rationale
Grant's "C last due to risk" instinct is correct. With A, B, D, E, G all shipped, there's a clean rollback baseline. If C3 degrades search quality, there's nothing else in flight to contaminate.
Execution Order
1.C4 first — depth review report, standalone, no risk, clears the deck
2.C1 — enrichment design + test harness
3.C2 — NCC pilot (1,806 clauses) + benchmark report
4.⚡ Grant checkpoint — review benchmark, approve/reject C3
5.C3 (if approved) — full corpus enrichment
→Next run: Checklist v2 (F1–F4), zero corpus risk in context
Timeline
~6–8 hours 1–2 sessions ~$28.90 API
Pros & Cons
Pros
Clean risk isolation — corpus work gets full attention
Benchmark is definitive — nothing else in flight to muddy results
Aligns with Grant's stated preference
Clear rollback baseline
Cons
Checklist v2 delayed by 1–2 sessions
Users wait longer for PDF export and signatures
B
Parallel Streams
Block C and Checklist v2 run simultaneously in two independent streams. Fastest total completion.
Rationale
Block C touches corpus files. Checklist v2 touches checklist components. Zero file overlap — they can run in parallel. Stream 1 handles C1→C2→C3. Stream 2 handles F1→F2→F3→F4 independently.
Execution (2 parallel streams)
S1C track: C4 → C1 → C2 → [checkpoint] → C3
S2F track: F1 → F2 → F3 → F4 (fully independent)
→Both blocks finish in a single run
Timeline
~8–10 hours (wall clock) both done together ~$29.40 API
Pros & Cons
Pros
Fastest total completion — both blocks done simultaneously
Checklist v2 ships sooner for users
No file overlap between streams
Cons
If C2 shows degradation, debugging corpus issues while also reviewing new feature deploys creates cognitive overhead
Higher cost (2 parallel Sonnet agents)
Parallel requires explicit approval — Grant originally said sequential only
⚡ Approval needed: Grant's 2026-06-17 directive was "sequential only." He approved 3-parallel for the last run but that was a one-off. Confirm parallel is still OK before choosing this option.
C
Checklist v2 First
Ship user-visible features now. Corpus enrichment gets its own focused run after — properly resourced, not an afterthought.
Rationale
Ship what users will see and use. PDF export, photos, signatures — these are tangible, demo-able improvements. Block C is backend infrastructure. Users don't see enrichment directly.
Execution Order
1.Next run: F1 (PDF) → F2 (Photos) → F3 (Signed) → F4 (Docker)
2.Run after: C4 → C1 → C2 → [checkpoint] → C3
Timeline
~8–10h (Checklist v2) then ~6–8h (Block C)
Pros & Cons
Pros
Users see new features sooner
More "demo-able" progress for near-term
Block C gets full focused attention in its own run
Cons
Block C keeps getting pushed back — it's been "next" for a while
Every day without enrichment is suboptimal search answers for all users
C's gated chain (C1→C2→checkpoint→C3) needs focused attention, not afterthought treatment
📅
Section 4 — Timeline Mockup
Visual Gantt view of each option's execution order · Colour-coded by block
Block C tasks (corpus)
Block F tasks (Checklist v2)
Grant checkpoint / approval gate
Next run (deferred)
Option A — Block C Only (Recommended)
~6–8 hours · Single focused run · Checklist v2 in next run
Run 1 Run 2 After C4 depth → C1 design → C2 NCC pilot → ⚡ Grant reviews → C3 full enrichment ✓ C done F1 PDF → F2 Photos → F3 Signed → F4 Docker ✓ Checklist v2 done Run 1 (~6–8h): Block C corpus enrichment. Gated chain with checkpoint. Run 2 (~8–10h): Checklist v2. Zero corpus risk. Clean context. → Phase 2 complete after Run 2.
Option B — Parallel Streams (Fastest completion)
~8–10 hours · Both blocks finish simultaneously · Requires parallel approval
C Track F Track C4 → C1 design → C2 NCC pilot → ⚡ checkpoint → C3 full parallel F1 PDF → F2 Photos → F3 Signed → F4 Docker ✓ done independently Both tracks run in the same session. C track is gated (C2 → checkpoint → C3). F track is fully independent. → Phase 2 complete in a single run. Requires parallel agent approval from Grant.
Option C — Checklist v2 First, Block C After
~8–10h (Checklist v2) then ~6–8h (Block C) in a separate run
Run 1 Run 2 F1 PDF → F2 Photos → F3 Signed → F4 Docker ✓ Checklist v2 done C4 depth → C1 design → C2 NCC pilot → ⚡ checkpoint → C3 full ✓ C done Run 1 (~8–10h): Checklist v2. PDF export, photos, signatures, Docker. Zero corpus risk. Run 2 (~6–8h): Block C corpus enrichment. Focused, clean context, no other work in flight. → Phase 2 complete after Run 2. Slightly longer overall but both blocks get full attention.
💬
Section 5 — Your Call
Select your preferred option and add any notes · No submit needed — just think it through