Q2 2026 Lookback

Our Q2, as analyzed and summarized by Claude.

Q2 was the quarter Momentum stopped being a pilot and became a platform. NYCA launched and grew to 500+ teams, vendor matching took us to DC, revenue nearly doubled year-over-year with five straight profitable months — and the way we build software changed as much as what we built. This is the internal version of the story: real numbers, real credit, and the honest challenges.

The quarter by the numbers

PRs merged (momentum)	2,030 — 481 in April, 598 in May, 951 in June. Accelerating, not just fast
Releases	87 tags in the quarter; from “release day” to push-button daily. 9 releases in one 14-day stretch in May
Calcs versions	~24 minor versions (1.26 → 1.50)
NYCA	500+ teams registered since April 14; daily actives ~25 → ~150
May user metrics	400+ unique users, 300+ scopes, 2x projects vs April — all records
Backlog	597 stale issues closed in one day; open issues now mean “we’re doing this.” We open ~150/week
P0/P1 bugs	Single digits all quarter; touched zero in late May
Architecture	Arch-test exemption ratchet down 42% in a two-week stretch; UBID is now the canonical building identifier end-to-end

What we shipped — Momentum

• Vendor matching & the contractor marketplace — the quarter’s biggest new surface: owner match requests, the matching engine, vendor response and contact-reveal flows, IMT bulk import, multi-building requests. Launched for real users at IMT’s Building Innovation Hub in DC.
• Team Messaging v0 + unified inbox — team threads, project Conversations, multi-party threads, vendor conversations converging onto the same infrastructure.
• Questionnaires — generalized, versioned, with a builder, responder flow, PII guard, and export. Went permission-gated GA in June.
• The new owner experience — building landing page, scope detail, scope comparison, notes where you work plus a team Notes hub, phase snapshots for a real project audit trail.
• LL97/BPS surfaces — the NYCA LL97 public map v0, fines through 2050, and generalization of LL97-specific UI to multi-standard BPS keys (the Boston enabler).
• Program platform — Programs bounded context, configurable registration landing pages, combined signup, program logos and theming, Salesforce-shaped activity exports, program-config export/import with versioning.
• Registration overhaul — co-op/condo board-member flow, BKB resolution that trusts CalcService, NY-metro NYCA auto-suggestion.
• Financing groundwork — lender-matching engine v0 with a real PACE/NYCEEC catalog, the coin-sorter UX, Contractor Outcomes v0.
• Platform integrity — the UBID migration (prod backfills, NOT NULL, hash URLs retired), the DDD wave (ProjectService god class deleted, CalcsIntegration reframed as a gateway), notification standardization + user preferences, CSP enforced, upload scanning, Labs + dark mode.

What we shipped — Calcs & data

• Metering, tariffs & rate-change OPEX — real ConEd electric and gas rate schedules, OPEX computed at scope and attributed back to measures, savings gated on data confidence. This is what makes electrification and submetering financially legible.
• Daisy Chain submetering measure with a new recurring-cost layer — directly monetizing the new partnership.
• Measure portioning — partial-coverage measures and a portion-aware LL97 optimizer; the prerequisite for scopes-over-time.
• Whole-building accuracy — rebuilt energy end-uses, EUI percentile scores, envelope-physics baseline corrections, the MFEEP/AMEEP rebate verification pass.
• The LL97 agent — productized with session persistence, observability, and an eval suite; in-context regulatory delivery is now an ADR.
• Sweep-as-a-service — statistically trustworthy PR regression sweeps on ephemeral apps, blue-green deploys, BKB-compatibility gates.

How we shipped it

The vibe-maxxing experiment that started March 30 became the way we work. PR output doubled in week one and kept climbing. Ana, Ilayda, and Chuck made the rotating tech-lead model real; Naina landed 10 PRs in three days during launch crunch and Mike shipped to prod in his first weeks; Stefan built a genuine multi-tool review practice; Chuck and Maks executed the database migration off Managed Postgres that ended a whole class of incidents; Reuben kept infrastructure standing through launch (and answered the “Reuben signal”); Jeff turned release QA into a repeatable practice. Erika and Reuben got us a clean SOC1 preliminary. Jon ran the NYCA webinar circuit that onboarded hundreds of teams, Marc turned customer-success embedding into two signed revenue agreements, and Jason kept an accelerating machine pointed at the right work.

We also invested in the system itself: role playbooks (Tech Lead, Release Lead, Bug Squasher, First Responder), grooming automation, release automation with regression sweeps, the coverage dashboard, and an automated UAT E2E suite. The goal is a team that runs its own releases, quality, and priorities — and Q2 built most of that scaffolding.

What was hard (and what we learned)

1. We create buildings; we don’t yet create enough projects. In the 32 days after NYCA launch, users added 22x more buildings — but scope creation didn’t keep up: buildings-per-scope ballooned from 3:1 to 36:1, and only ~20% of registration visitors reach the buildings page. April cohort churn was 65%. The users showed up; the product has to convert them. This is the top product problem for H2, and the funnel is now instrumented.

2. No project has gone end-to-end yet. Scoping → procurement → implementation entirely on-platform hasn’t happened. CPC CFHF taught us that mandates don’t create buy-in — the tool has to make people’s actual work easier. The pieces all exist now; H2 is about lighthouse partners who want to run their work here.

3. Velocity had a bill. Week one of vibe-maxxing produced ~20 release blockers. We had a 3-hour downtime before launch, the PgBouncer outage in April, and a 30-minute calcs incident in June. Each one led to real hardening (revert drills, the MPG migration, deploy gates) — but testing automation trailed shipping speed all quarter. The UAT E2E suite only landed in the final two weeks; it needs to grow up fast in Q3.

4. “Fusion” features are still slow. Features that need baton-passing between Momentum, Calcs, and Data (rate changes, LL97 pathways, tariffs) drag compared to contained ones. The measures-refactor post-mortem — four failed attempts across three engineers — is the sharpest evidence that some of our cross-service seams need design work, not just effort.

5. Too much of the quarter ran through one person. Honest accounting: too many of Q2’s commits were Bomee, including hotfixes, release surgery, and late-night landings. The team demonstrated all quarter that it can own releases, quality, and incidents; the H2 plan moves Bomee off the critical path for real. Velocity may dip while habits set. That’s the price, not a failure.