Company
How we build
Eight rules for shipping the WB-1: edge decisions, worst-case numbers, deterministic conflicts, rules as data, and evidence before brochure copy.
This is not a values mural. These are choices that cost schedule, board area, or convenience and survived anyway. Each principle points to the surface where a visitor can check it.
Principle 1 — Decisions happen at the edge
A building should not lose its reflexes because an uplink is slow or a vendor’s status page is red. The Cortex-M7 owns the deterministic path; cloud services may analyze history, never hold the critical loop. The split is visible in the architecture.
Principle 2 — Worst case or it does not count
A median hides the exact event that ruins a cue or delays a hold. We quote the sub-8 ms path as a worst-case measurement under stated load, keep the layer budgets separate, and publish the conditions beside the number. The specifications carry the contract.
Principle 3 — Conflict order must be boring
Two rules will eventually demand opposite outputs in the same cycle. Priority, then freshness, then specificity decides; both contenders and the deciding key enter the audit record. There is no random scheduler victory to explain after the fact. Read the engine’s conflict model.
Principle 4 — Rules are data, not site firmware
A customer’s room plan belongs in reviewed YAML, not a fork no one remembers how to build. Rules validate before activation, hot-reload without a reboot, and carry a revision that can move across a fleet or roll back. The engine proves the boundary; the DSL reference makes it usable.
Principle 5 — Measure production hardware
A component datasheet is an input, not a product result. We measure assembled units at stated temperature and channel load, quote the nearest limit, and keep the ugly trace when it explains a redesign. The field journal shows that work before it becomes a launch claim.
Principle 6 — Dirty RF before general availability
A quiet bench cannot reproduce fourteen courts, VFD cabinets, moving metal walls, or a saturated event network. Design partners get early units because those rooms reveal antenna placement, retry, and operator problems we cannot invent in the lab. The field journal and case studies preserve what changed.
Principle 7 — Cloud is an option, not life support
Cross-site analytics and long retention are useful. Neither should be required for a local decision, a rule edit, or an audit export. A disconnected mesh keeps running and syncs evidence later; the topology guide shows which state crosses each boundary.
Principle 8 — Design for the returns pile
We are six people and one product. Signed A/B updates, brown-out recovery, watchdogs, calibrated radios, and replaceable mounting are cheaper than arguing with a failed unit on someone else’s loading dock. The failure assumptions live in the specifications and the release trail lives in the field journal.
The short version
Make the local path deterministic. Make the external surface documented. Make the failure leave evidence. Measure the unit that ships. Then write the brochure — in that order.