Engagements

$106 million in Portland housing funds sat uncommitted with no architecture connecting them to validated interventions — while the shelter system exited 17 percent of people to permanent housing against its own 41 percent goal, and more than 18,000 people went unhoused.

No commission preceded the work. The engagement was a deliberate live test of the Sarif methodology: whether one analyst with an AI-augmented production stack could produce civic intelligence at the standard a council floor demands, delivered cold to the officials who would have to act on it.

Council passed the Keep Portland Housed ordinance days later, committing approximately $56 million — close to the figure the matrix had marked as deployable. We claim the sequence and the analysis, not the causation: two sitting councilors and the executive director of a regional housing coalition responded directly, and the suite was revised twice inside the live budget window to keep pace with implementation.

Lloyd Center is 29.3 acres on three light-rail lines — one of the most transit-rich sites on the West Coast — approved for demolition of a seismically sound civic core while 2.4 million square feet of parking garages remained unexamined. The staff findings never analyzed the asymmetry.

The coalition had public will — 10,000+ petition signatures, 600+ pages of testimony — but no instrument an institution could act on: no financial architecture, no legal crosswalk, no phased alternative. We built the missing instrument: a 29-card friendly amendment, a source-verified financial model, the approval-criteria crosswalk that carried the appeal, and the rendering set that made the alternative visible.

Council denied the appeal unanimously on July 8, 2026. The filing still did its work: the substantive claims are permanent civic record, the alliance architecture behind the approval is now public, and Phase 2 pivots from remand to transaction — a coalition-aligned acquisition inside the owner's announced strategic-review window.

A B Corp personal injury firm was losing after-hours callers to whichever firm answered first. The obvious fix — an automated voice answering as the firm — collided with the brand: trauma-informed, human-first, bilingual. The wrong intake voice would cost more than the missed calls.

The product was a bespoke intake intelligence layer: one regulated voice across three intake channels — Local Services Ads, Website Inquiries, After-Hours Calls — unified into a single record. One voice. One memory. One handoff.

Legal intake is a regulated act. Oregon State Bar Formal Opinions 2026-208 and 2025-205 bound what an automated system may say, collect and promise — and no off-the-shelf intake product was built to that constraint set. We designed around it: a five-node architecture — Reception, Listening, Gathering, Discernment, Handoff — with constrained decoding, bilingual parity and a morning handoff that returns judgment to the attorneys. AI captures. Humans decide. The complete design architecture — voice, compliance, routing, escalation — was delivered as a fixed-fee first phase, with build scoped as a separate decision.

A solo practitioner had the rarer half of the problem solved. She could run a retreat that changed people — and had no way to say what she was selling, to whom, or why it beat the alternatives a buyer would price her against.

The market gives that buyer two options and neither one works. Executive coaching delivers rigor without immersion; wellness retreats deliver immersion without integration. Roughly 123,000 coach practitioners compete on the first and a crowded, largely undifferentiated retreat market competes on the second. The intersection — coaching rigor, retreat immersion, sustained integration — was structurally unoccupied.

So we did not write a marketing plan. We built the category, then the business inside it: a document system where the pricing architecture, the ideal-client definition, the risk register and the sprint plan all resolve to the same thesis. Retreat-first, integration-backed, deliberately small.

A K-8 public charter school — small, place-based, built on an academic model with no local equivalent — was forced off its campus in the middle of an enrollment collapse: a 29 percent enrollment gap, roughly $630K of annual revenue gone, reserves measured in days. The mission was excellent. Excellence was not the problem.

Mission strength does not guarantee institutional sustainability. What the school lacked was not conviction but architecture: no revenue diversification, no enrollment engine, no operational systems, no sequenced path from survival to stability.

We built the architecture: seven modules across six interdependent domains — narrative, financial, enrollment, partnership, operational, implementation — sequenced across 36 months and costed against probability rather than hope. The seventh module is the one most consultancies omit: a register of thirty-five assumptions, each flagged for validation, delivered with the standing instruction to treat the plan as hypotheses to be tested rather than certainties to be executed.

The brief was a website. What we built was an environment — a real-time 3D lobby you walk into rather than a page you scroll, rendered in the browser at device pixel ratio. First light took 48 working hours, concept to production deployment.

Then we spent twice that again on the part nobody sees. The interesting engineering was not the lobby; it was making a WebGL environment behave like a text document — fast on first paint, legible without JavaScript, navigable without a mouse, and safe under a strict Content-Security-Policy.

The site is now its own proof surface. It is the reference implementation for how we work: every claim on it is checked by a build gate, and the gates were written because each of them once caught something we had gotten wrong.