Spectral Engineering | Alex Wolf

Your org chart says one thing.
Your communication graph says another.
The graph is right.
The offering
Bi-directional socio-technical engineering. DGSF-certified. Supervised.
The human informs the tech. The tech informs the human. That's the loop. I'm the translation node.
What this is, structurally: I help you see the topology your organization actually has. Your team does its own work from there. The engagement transfers recognition, not direction.
I embed with your team. I measure your communication topology with mirror. Not your org chart. The actual communication topology of your organization. The code. The pull requests. The tickets. The slack messages. All local. On your infrastructure. We measure what's there. Not what's described. Not "who says what?" But "who talks to whom when in which function?"
We don't look at what people say. That data stays invisible. We look at how your organization connects.
The measurement informs the intervention. The intervention produces a new graph. The new graph refines my understanding of socio-technical systems, which shapes how I improve mirror's architecture — not its data. Your graph stays on your infrastructure. Only the patterns I learn from working with you cross over, and only via the practitioner.
This is not consulting. This is not a tool. It's the practice of deriving technology from human behavior and feeding the learning back. Bi-directional.
What I measure
Fiedler value.
How close your organization is to splitting in two. The number is computed automatically. I show you what it means.
Star-graph topology.
One person connected to everyone, everyone connected only through them — the bus factor made mathematical, the architect who is the architecture. The eigenvalue profile makes the shape visible; the engagement is where we name the specific node and decide what to do about it.
Communication bottlenecks.
Where information actually flows. Where it doesn't. Where two teams think they're collaborating but the graph says they're passing messages through a single person who never signed up for that role.
Load distribution.
Who carries the glue work. Who carries the ambiguity. Who carries the context that isn't written down anywhere. The graph shows it. The number confirms it. No narratives required.
What mirror does
mirror is a build system (and a compiler, and a model checker, and a measurement too, different altitudes, same thing).
Deterministic, poly-language, multi-repo — Nix and Bazel are its ancestors. Built by a platform engineer with 15 years in tech who got tired of build tools that pretend the communication graph isn't there. The compiler measures which parts of your codebase are close to fragmenting — the Fiedler value dropping to 1, the literal single point of failure named. Then I sit with the people the math just named, and we ask the circular reflexive questions a build system can't.
Local-first AI runtime (spectral) built on top. Sub-Turing compiler (the grammar is decidable; the compiler is a model checker over it). 425-parameter graph navigation model (weights ship with the binary). Content-addressed package manager at garden.spectral.engineer (ETA June).
It runs on your infrastructure. Your data stays. The runtime does not phone home. Content never leaves the machine. What leaves — if you opt in — are eigenvalues. Sixteen numbers. The shape of your topology. Not the content.
What changes
- Decisions converge instead of looping
- Incidents produce fewer narratives and more agreement
- Responsibility becomes explicit without blame
- Leaders stop carrying unbounded ambiguity alone
- The person who was silently load-bearing gets visible
This is upstream work. Before tech debt. Before burnout. Before the third reorg that solves nothing.
About me
Hi, I'm Alex.
(they/them)
Neuroqueer (AuDHD). Systemic engineer.
Supervised systemic practitioner (DGSF-certified, in training).
15+ years building distributed software systems, 10 on the BEAM.
Scientific Programming degree from Europe's biggest supercomputing centre.
Three burnouts. Still shipping.
(Also doing 1on1s for neuroqueer individuals, complex family systems, and displaced tech workers: Systemic 1on1.)
My brain works differently. Where people see stories, I see Ricci flows. Where people see a party, I see a social topology. Where people see small talk, I see.. polite conversation. (I get it, I do it, I don't like it.)
I didn't choose to be like this. And for most of my life I didn't understand how my brain works. I do now. systemic.engineering is the formalization of that understanding. mirror the compiler that produces spectral triples, a mathematical object that apparently also describes the shape of my cognition. spectral the runtime built on top, the 15 years of distributed systems engineering put to good use.
The math measures the graph. The graph doesn't lie. I help you handle the ripples of your gaps.
The runtime implements the practice. The five operations in mirror are the same five moves I make in the room. Mirror. Offer. Wait. is what the math does when it converges. 🍷
If you want to see how I work
Predator Code — the formation.
The Drone in the Field — the architecture.
A Letter from Reed — the AI work.
Is this real?
EU compliance
GDPR. AI Act. Digital Services Act.
mirror provides compliance by construction. Consent architecture. Anti-extraction. Data locality. Not a checklist you maintain. Architecture that makes violation structurally difficult.
The license
All work is governed by the systemic.engineering License (SEL). Three structural commitments:
- Anti-extraction. Cognitive, emotional, and relational labor has value. The license names it and creates liability for omitting it.
- Consent-based collaboration. Offers are not commands. Silence is not consent. Withdrawal of consent is honored.
- Intersectional justice. Systems built with this work cannot be used to reinforce structural oppression — including punitive performance management, surveillance-driven workforce reduction, or analysis frameworks deployed against your engineers' interests.
A Duty of Care attaches to both sides: I accept it toward your engineers; you accept it toward them by working with me.
Engagement
4-12 weeks. One live system under real load.
I work directly with your engineering leadership.
I observe real interactions. I don't survey. I measure. (mirror)
I show you the topology. (offer)
You decide what to do with it. (wait)
Rates on request. Terms.
Which system keeps failing despite effort and intent?
Email me. As short or as long as the question deserves.