Engineering Roadmap

1. The map

Read it like an airport departures board — each row is a team, each block is a window of work, and the greyed-out row already got cancelled.

• Western Slope goes live Thursday 9/17: voice, tours, leasing, application link, rescheduling.
• Next ~3 weeks, two tracks run at once: Maintenance and Renewals.
• Platform: onboarding-import engine finishes this weekend, then scale to 10,000 units.
• Clara becomes a coworker — text-to-reschedule first, then AppFolio guest cards.
• The Promises platform is killed; that work moves into Clara as coworker.

Proves: Fede's dictated rulings, 2026-09-12 evening — memory files cto-top-priority, western-slope-mvp-launch, ci-speed-10x, appfolio-onboarding-import-module.

2. This week (Sep 14–18)

Like a pre-flight checklist taped to the cockpit door — four boxes need a check mark before Thursday, and only one of them is engineering's pen.

• Top priority: Western Slope live on voice, SMS and email by Thursday Sep 17.
• Situs Group onboarding starts this week — their first build is after-hours/weekend phone coverage.
• Settings cleanup waits for after go-live, except Situs's day-one tour gap and confirmation window.
• Four Thursday blockers: mailbox receive path, Jay's Microsoft approval, empty knowledge base, inbound-text webhook.
• Touring engine design and the testing-cadence proposal are linked — design, not this week.

Proves: Fede's rulings, 2026-09-13, plus the readiness audit at leasing-all-channels-readiness-ws-2026-09-13.md (§0, §7 shortest path).

3. Top priority

Like widening the runway before adding in-flight service — until the platform can carry more clients, better features don't move the needle.

• Fede's personal top priority as CTO: onboard clients fast, scale to 1,000 then 10,000 units.
• Winning is measured by onboarding speed, not feature count.
• Credits (~$9k GitHub, ~$5k AWS) are free grants — spend them for speed, 6-month horizon.
• Every item below is ranked by whether it shortens onboarding or removes a scale ceiling.

Proves: user_cto_top_priority_2026-09.md, Fede dictated ~8:30 PM, 2026-09-12.

4. Product now

Like flipping a shop's sign to "Open" — Thursday the phone line starts taking real calls for real.

• Western Slope's MVP launches Thursday, September 17, 2026.
• Scope: voice line, tour booking, leasing conversation, application link, rescheduling.
• Same window: a call auto-creates a guest card in AppFolio.
• Fede's recommendation: run fully autonomous like Camellia, not Situs's human-in-loop model.
• Autonomous vs. ease-in stays a separate, still-open decision — see chapter 10.

Proves: project_western_slope_mvp_launch_2026-09-17.md, Fede dictated ~8:45 PM, 2026-09-12.

5. Next 3 weeks

Two lanes on the same highway, not a single-file line — Maintenance and Renewals move at the same time.

• Right after launch, two workstreams run in parallel for about three weeks (est.).
• Maintenance replaces Situs Group's call center — Fede calls it the biggest win.
• Renewals: Fede personally drives this one.
• ConAm onboarding queues behind these — guaranteed deal, capacity reserved when they're ready.

Proves: user_cto_top_priority_2026-09.md ("product priorities") + project_new_clients_situs_western_slope.md.

6. Platform

Like widening a highway before rush hour, not after the jam starts.

• Onboarding import + sync scale engine is in flight, finishing this weekend.
• The old every-minute-per-property sync was the wrong tool for a first pull.
• Next: scale the platform to 1,000 units, then 10,000.
• CI 10x: proof-of-concept ran 9/12; goal is a PR live in prod every ~15 minutes.
• Funded by ~$9k GitHub + ~$5k AWS credits, spend-for-speed, 6-month horizon.
• Follow-up (Smith, 2026-09-22): close the "test that scans the repo" gap in the every-push safety net. Main went red on 9/21 after a merge that CI had called green: the broken test finds its inputs by scanning the repo (a git search), so it has no import link to the new file and the changed-files lane never picked it up — and it was also never added to the guard lane that runs on every push for exactly this kind of test. Six tests sit on this gap today (among them the org-knowledge readers pin, the dead work-orders links check, the tour-duration wiring check; Smith's recount on 9/22 found three more written in a form his first search missed); the same shape reddened main on 9/13. Fix: teach the guard-lane selector that a test shelling out to a git scan belongs in it by construction — +6 files on an ~844-file lane, no full-suite escalation on merges. Merged and live (#10724, 9146303f, 2026-09-22). One tiny follow-up still owed: two shipped code comments describe husky-runtime backwards (say it was never derived-selected; it always was, off a coincidental literal) — Smith flagged it himself post-approval and is fixing it in a comment-only PR rather than reopening review.

Proves: project_appfolio_onboarding_import_module.md (overnight chain, 2026-09-12) + project_ci_speed_10x_2026-09.md.

7. Clara as coworker

Like giving a new hire more responsibility only after they've nailed the basics.

• Step 1: a manager texts Clara to reschedule a tour — she does it, basics first.
• Step 2: guest cards get worked inside AppFolio, Clara as assignee — Situs's model.
• Humans, VAs and Clara share one screen; a PM decides card-by-card who owns it.
• When stuck, Clara emails the team and learns from the answer — no review queues.
• Staff-reply meaning judge (decision / takeover / working), Sonnet 5.5, staff loop v2 PR-C — PR #13384 open on the integration branch, in review (refs #13379); next: PR-D hand-off execution (wires both judges), then PR-E knowledge

Proves: user_cto_top_priority_2026-09.md ("medium-term priorities" #4, "further rulings"), 2026-09-12 ~8:50 PM.

8. Integrations & partners

Like choosing which loading dock to widen based on which trucks actually show up.

• Invest in becoming a certified Google Workspace partner, not Outlook.
• Keep deepening the AppFolio sync — coverage and how much data comes in.
• Explore a partnership with Orca, a Denver field-service tracker for maintenance techs.
• Orca priority: maintenance data first, then turnovers; tour-assignment is low priority.

Proves: user_cto_top_priority_2026-09.md, Fede ~9:00 PM, 2026-09-12.

9. Clients & pacing

Like deciding whether to finish the house you're building before starting a second one.

• Western Slope (~250 doors, Grand Junction): launches 9/17, recommended fully autonomous.
• Situs Group (645 units, Denver): second centralized client, onboarding starts this week.
• Situs's first agreed build: after-hours and weekend phone coverage, humans in the loop otherwise.
• ConAm: guaranteed deal, lower priority, capacity reserved when they're ready.
• Risk profile decides the build: scrappy clients go autonomous, cautious ones ease in.

Proves: user_cto_top_priority_2026-09.md (pacing decision) + project_new_clients_situs_western_slope.md + Fede's 2026-09-13 rulings.

10. Decisions open

Like a short punch list on Fede's desk — each one a pick, not an open question.

• Pacing: (A) 6–8 weeks heads-down, recommended, vs. (B) add one apartment client now.
• Clara-as-AppFolio-seat: confirm guest-card assignment works, recommended, before building Situs ease-in.
• Orca outreach: recommended after Maintenance ships (~3 weeks), not before.
• Kill-Promises execution: remove fully now, recommended — Fede said kill it completely.

Proves: user_cto_top_priority_2026-09.md, Fede's rulings ~8:50–9:20 PM, 2026-09-12.

Deeper → Appendix: full item table

11. The last eight weeks

Like a weekly time-card for the whole team — one bar per week, taller means more engineering time went in.

Proves:

Deeper → Appendix: the full weekly grid.

12. The same thing, as numbers

Like a bank statement next to the calendar it came from — the same weeks, now as totals anyone can check.

Proves:

13. Where the time actually went

Like reading two people's last two weeks of receipts side by side, instead of asking them what they think they spent.

Proves:

Deeper → Appendix: the lane-by-lane table, with what is underway and what is next.

A separate read-only pull of the last two weeks of PRs (09-15 to 09-28) looked at why work piles up and where the hours actually go. Short version: review itself got faster and cleaner week over week (fewer rounds, fewer holds), even while volume rose 34%. What got slower is the last mile — merge to live now waits on a 30-minute deploy batch. The single biggest time sink, by far, is about 40 PRs sitting green and unmerged in Gera's lane. Three fixes are proposed, pending Fede's review: refuse PRs over 600 lines, clear the parked green PRs same day, and stop re-reviewing unchanged findings. Full breakdown, tables, and the proposal → Always Shipping: PR flow, 09-15 to 09-28.

14. Who is doing what next

Like two lists pinned side by side — what each person has half-built right now, and what they pick up next.

Proves:

Deeper → Appendix: every lane, its owner, its measured share, what is underway, what is next, and where that came from.

A1 · Appendix — full record

Every table behind the chapters above: this week's decisions, the lane-by-lane roadmap, the weekly investment grid, and the work that matched no lane.

Decisions open — full option table

Windows past this week are estimates ("est.") — none were dictated with a specific date. No item here authorizes turning a feature on for a real customer; that stays Fede's explicit call under the standard rollout playbook.

Every lane: share, underway, next, and where "next" came from

Weekly investment by lane (lines of code changed in finished work)

Source repositories: the product and the agent fleet. Lockfiles and generated files are left out on purpose.

The biggest single piece of work, per lane per week

Work that matched no lane