Roadmap Review 2026-08-24b — Endgame Sequencing

Status: Planning audit only. No feature code touched. Scope: Second strategy rotation on UPGRADE_ROADMAP.md. Review #1 (roadmap-review-2026-08-24.md) set the sequencing; that sequence is now mostly EXECUTED — Mission Control shipped + ticked, gateway bridge validated + ticked, live console shipped + validated + ticked, budget ledger slice 1 + session replay backend shipped, budget slice 2 + replay view building concurrently. Evidence base: git log 92021f9..HEAD (22 commits, 01:15–10:21), CHANGELOG [Unreleased], docs/briefs/{budget-ledger,session-replay}.md, routes/cost-routes.js, scripts/docs-drift-check.js (exit 0, 0 errors), DB-free suite 42/42 green at audit time.

---

1. Velocity Assessment

Measured since review #1: 22 commits in ~9 hours, three parallel lanes (build, validation, docs). Composition:

Outcome	Count	Items
Roadmap items fully closed	3	Mission Control (112b224, 0a667ab), gateway bridge (6184097, d568a07), live console (83919c4, 1c8acf7)
Partial lands on open items	2	replay backend part 1 (49eef27), budget slice 1 (0a1ed9b)
Design briefs	3	session-replay, live-console, budget-ledger
Research/spike	2	streaming spike (ed3d733), market scan refresh (bfd184e)
Security dep fixes	2	GHSA-5p4m-2wfm-xmqj (ad1fbc5, PR #19)

Cadence held at roughly one roadmap item closed per ~3 hours, with briefs landing 2–3 h ahead of their builds. Review #1’s “~2× commit overstatement” rule of thumb continues to hold: of 22 commits, ~11 were bookkeeping/docs.

Verdict: velocity is healthy and increasing in feature density — Phase 1’s three heaviest integration items (MC, bridge, console) all closed inside this window, each validated against the live gateway before ticking. No stalled lane.

2. Unchecked Inventory — What Genuinely Remains

15 unchecked boxes remain. Honest state per item:

Phase 0 leftovers (2) — both small, one is a live blocker

Item	Effort	Notes
Cost/token history backfill	S — ≤1 run	No backfill script exists in the repo (verified). Gates budget slice 2 enforcement AND the Phase 2 analytics residuals. Highest-leverage small item left on the board.
npm audit as CI gate	XS — ≤½ run	Foldable into any run already touching .github/workflows/ci.yml; otherwise standalone micro-run.

Phase 1 (6 boxes; 2 in flight)

Item	State	Remaining effort
Session replay inspector	Backend shipped (reader + events routes, 49eef27); view building now	≤1 run to land + validate. Brief AC1–AC9 ready for qa-auditor.
Budget ledger + auto-pause	Slice 1 shipped (migration 023, budgets API, pure eval, 0a1ed9b); slice 2 (dispatcher enforcement) building now; slice 3 (MC bars + budget_breach flag + SSE) not started	Slice 2 ≤1 run (⚠ see §4 ordering flag); slice 3 = 1 run.
One-click agent actions	Not started. Brief does not exist — the “write it during bridge/console runs” instruction (review #1 run 7) expired unused when both those runs completed this morning	Brief = 1 docs run (can overlap build lanes); build = 1–2 runs (write paths, idempotency, protected-action preview + receipts per market scan 2026-08-24). Longest remaining pole.
Memory browser 2.0	Not started; semantic-search foundation live in memory-view	1 filler run. Lowest daily-value item in phase.

Done-by-side-effect verdicts (the “can we tick it?” question)

• Cost & token analytics UI — PARTIALLY covered, do not tick. Fleet-level aggregate (GET /api/costs/summary?days=7) + Mission Control cost panel ship today/7d totals with correct degradation states; budget slice 3 adds budget bars to the same panel. What does NOT exist anywhere: per-agent / per- department / per-workflow-type rollups (verified: cost-routes.js has no GROUP BY owner_agent_id-class query) and sparkline feeding. Remaining true scope is one rollup endpoint + UI wiring ≈ 1 run. Annotate the checkbox accordingly (done in this commit) rather than tick.
• Session replay inspector — backend half done-by-side-effect of the console work (session-reader infrastructure), but the checkbox describes the stepper view; stays open until the view lands.
• Nothing else qualifies. Snapshot/restore’s foundation (export-routes.js) exists but the item is genuinely unbuilt; MCP exposure has zero code.

Phase 2 (5) and Phase 3 (4)

No drift found. Phase 3 items remain low-coupling polish. See §5 for Phase 2 readiness verdict.

3. Recommended Final Phase 1 Order (endgame queue)

In-flight lanes finish; then:

Order	Run	Item	Why
1	next available	Cost/token backfill	Unblocks slice 2 merge + analytics residual; small; zero collision with in-flight lanes
2	in flight	Replay view lands + validates	Closes its box
3	in flight, after #1	Budget slice 2 lands	Enforcement over non-sparse history
4	+1 run	Budget slice 3 (MC bars, budget_breach flag, SSE typeMap) + fold npm audit gate if a CI-touching run is open anyway	Closes budget box
5	docs lane NOW	One-click actions brief (action set, confirmation UX, idempotency, protected-action preview + receipts; also defines the gating machinery NL command bar reuses)	Removes the long pole’s blocker; pure docs, no lane collision
6	after #5	One-click actions build (1–2 runs)	Last big Phase 1 lift
—	filler	Memory browser 2.0 whenever a DB-dependent run blocks (working rule 8)	Explicitly flex (§6)

4. Sequencing Flag — slice 2 vs backfill

docs/briefs/budget-ledger.md §4/R2 is explicit: backfill “must land before Slice 2 (enforcement)” because budgets over sparse history silently under-enforce. Slice 2 is building right now while no backfill exists in the repo. Two acceptable resolutions, in preference order:

1. Backfill lands before slice 2 merges (it is a ≤1-run job and touches disjoint files — no lane collision).
2. If backfill slips, slice 2 merges with the under-enforcement caveat documented in CHANGELOG and the budgets API copy, and backfill becomes the mandatory next run.

Option 1 costs nothing; take it. Do not let slice 2 sit merged-and-silent.

5. Phase 2 Readiness Verdict

Ready to open in parallel — for two of five items — before Phase 1’s box is fully clear:

• MCP server exposure: zero Phase 1 dependencies; read-only toolset over shipped REST routes; still the single most defensible roadmap item (review #1 slot-protection stance holds). Start once the two in-flight build lanes drain (cap: 2 build lanes + 1 docs lane — see risk 3).
• Snapshot/restore: builds on existing export-routes.js; independent of everything in flight. Cheap, operator-valuable, safe interleave.
• Cost & token analytics UI: unblocked the moment backfill lands; 1 run at its reduced (annotated) scope.
• NL command bar: sequenced AFTER one-click actions — it reuses the same confirmation/receipt machinery; building it first would fork the gating design. This dependency is why the brief (§3 order 5) matters beyond Phase 1.
• Workflow visual editor Stage 1: last in phase, first willing-cut (§6).

Phase 1 completion ETA: functionally complete (replay + budgets + one-click actions) in ~4–5 further runs ≈ within the next working day at observed cadence. Not “next run or two” — that horizon covers only the two in-flight items. Box-and-all including memory browser: ~5–7 runs, unless memory browser is flexed out of the exit criteria (recommended, §6).

6. Cut / Fold Recommendations

1. FOLD — cost analytics UI scope: mark fleet-level coverage as delivered by side effect (cost summary API + MC cost panel + upcoming budget bars); reduce the checkbox to rollups + sparklines (annotated in this commit). Saves ~½ run of rediscovery and prevents a phantom “big” Phase 2 item.
2. FLEX — memory browser 2.0 out of the Phase 1 exit criteria: keep the box in Phase 1 as designated filler, but explicitly not blocking Phase 2 start (annotated). Single-operator daily value is the lowest in phase; market scan already ruled memory-graph parity can trail. Prevents a filler item from holding the phase gate hostage.
3. FOLD — gating design, write once: one-click actions brief owns the confirmation/idempotency/receipts machinery; NL command bar consumes it. Recorded on both checkboxes (annotated). Prevents duplicate design work in Phase 2.
4. KEEP — one-click actions and MCP exposure: differentiator core; no cut.
5. WILLING-CUT — workflow visual editor Stage 1: if Phase 2 slips behind schedule, this is the first item dropped entirely; read-only graph render is eye-candy next to MCP depth. No roadmap change yet — decision point at next review.
6. Phase 3 unchanged.

7. Risk Register — Top 3 (delta from review #1)

1. Slice 2 / backfill ordering violation, in flight. Self-inflicted process drift against the budget brief’s own dependency gate (§4). Mitigation: backfill takes the next available run; resolution recorded in CHANGELOG either way.
2. One-click actions brief debt. Its scheduled writing window (during bridge/console runs) closed unused; it is now the long pole of Phase 1. Every half-day of slip pushes phase completion a half-day. Mitigation: docs-lane run for the brief immediately — it collides with nothing.
3. Concurrent-lane collision surface growing. Budget slice 2 (dispatcher + api-reference docs) and replay view (app-registry + views-reference docs) both carry docs-drift gates; a third build lane would create merge friction on shared registry/reference files. Mitigation: hard cap 2 build lanes + 1 docs lane until both in-flight items land.

Watchlist carried from review #1: cost-data sparsity keeps anomaly flag 4 quiet until backfill lands (now doubly true — it also gates slice 2); CHANGELOG [Unreleased] backlog (~20 entries) should be folded out at the next release cut — candidate trigger: Phase 1 completion.

8. Actions Taken in This Commit

1. This review document.
2. UPGRADE_ROADMAP.md, three annotations, nothing ticked or unticked:
   • Cost & token analytics UI: fleet-level marked delivered-by-side-effect; remaining scope narrowed to per-agent/department rollups + sparkline wiring; backfill dependency restated. Justification: §2/done-by-side- effect audit — prevents the hourly cron from re-planning shipped scope.
   • Memory browser 2.0: marked flex — does not block Phase 2 start. Justification: §6/fold 2 — keeps the phase gate honest.
   • One-click agent actions: brief-overdue note (window expired) + shared-gating-machinery link from NL command bar. Justification: §4 risk 2 and §6/fold 3 — the cron picker needs to see the brief as the actionable next step for this item.

Review only. No .js/.mjs/.sql/.yml changes. node scripts/docs-drift-check.js exit 0 (0 errors, 10 pre-existing warnings); DB-free suite 42/42 at audit time.