Software development studio/Manila / Remote
We map the domain, name the system, and build the smallest architecture that will still be true in five years.
01 — Manifesto
Named objects. Sharp interfaces. Constraints that outlast the sprint. This is how we develop software — and how we refuse to.
A feature is a request. A structure is a decision. We refuse to pile screens onto an unnamed model.
If the business cannot point to it, the code should not either. Language is the first architecture.
Latency, empty states, and failure modes are part of the product — not leftovers for a later sprint.
Until it is the cheapest way to keep a promise. We write less, then we make what remains durable.
The next engineer should be able to change it without a séance. Observability is a courtesy to the future.
Constraints, grain, and the grain of change. Software that holds its shape can still move.
02 — Systems
We do not start with a stack. We start with the planes a product actually lives on — then we choose tools that keep those planes honest.
What people actually touch
Copy, density, empty states, and the time between intent and result. We treat the surface as a contract with a human, not a skin on a schema.
03 — Practice
Engagements run twelve to thirty-six weeks. We ship a vertical slice early so the model can be proven against reality — then we grow the system along the grain we found.
01
Artifacts over opinions. We sit with the real workflow — not the org chart — and write down what the system already believes.
02
Names, invariants, events. If we cannot draw it on one page, we do not understand it yet. The model is the product.
03
Thin vertical slices, production from week two. Interfaces stay small. The architecture earns complexity; it is not gifted it.
04
SLOs, on-call, the next year of change. We leave a system that a new hire can hold — and a written grain of how it should grow.
04 — Workbench
Type help in the kernel. Filter the stack by plane. This is how we keep the work inspectable — including the marketing site.
kernel
v0.9
lattice 0.9 — studio kernel
type `help` to list commands
Stack by plane
05 — Selected work
Three engagements. Same grain: name the domain, keep the surface small, leave a system someone else can hold.
2025
Freight operations
Replaced a twelve-year spreadsheet spine with a domain model: vessels, berths, windows, exceptions. Dispatchers went from tribal knowledge to a system a new hire can run.
Exceptions closed 40% faster in the first quarter.
Case closed
2024
Treasury
A mid-market bank needed a single picture of cash. We modeled positions, not reports. The reports fell out of the model — and stopped disagreeing with each other.
One source of position truth across four desks.
Case closed
2025
Scientific capture
Field biologists were losing observations to camera rolls and group chats. An offline-first capture client, structured at the point of seeing, that syncs when the canopy allows.
Observations reusable the week they are taken.
Case closed
06 — Brief
Typical engagements run twelve to thirty-six weeks. We reply with a written read of the domain — not a deck of capabilities. If it is not a fit, we will say so.