Engineering blog
How the arcade is built.
Stacks, system designs and the trade-offs behind each Fusion Arcade product, written from the code. Where something is planned rather than built, we say so.
How the arcade fits together, today and eventually
Five products, five codebases. What they actually share today (principles and patterns, very little code), the few concrete links between them, and the plan for Bridge and Forge OS to become the platform the others run on.
Read the postChamber: a local-first notebook with a scheduler you can audit
How Chamber keeps notes, flashcards and review history on your device: one write path with an operation log, an in-house FSRS-6 scheduler with a parity test, reversible AI proposals, on-device OCR and speech, and sync that encrypts before upload.
Read the postTesting a spaced-repetition scheduler: FSRS-6 parity and a personal optimiser in Chamber
A deep dive into Chamber's scheduler: the FSRS-6 maths, why we test it against an independent implementation rather than against itself, the three real bugs that caught, and the small deterministic optimiser that tunes it to your memory.
Read the postLifekeep: history that can't be rewritten and a planner that explains itself
Inside Lifekeep's design: effective-dated versions enforced by Postgres, evidence-gated weekly adaptations, an explainable day planner that respects prayer times and your calendar, and an operation language that keeps every AI write approvable and undoable.
Read the postMapSight: building a versioned, multilingual graph of local businesses
How MapSight turns open place data into a clean market map: priced builds on Temporal, append-only observations with provenance, explainable entity resolution across scripts, industry packs as data, immutable dataset versions, drive-time reach with Valhalla and change detection that waits for corroboration.
Read the postOne business, many spellings: cross-script identity and deduplication in MapSight
A deep dive into how MapSight decides two listings are the same business: a conservative, versioned text fold that works across Arabic, Persian, Urdu, Devanagari, Thai, Latin and more; blocking keys; and a noisy-OR match score that refuses to pretend it's a probability.
Read the postBridge: designing a command deck that agents report into
When coding agents write most of the code, the founder's bottleneck becomes decisions. How Bridge is designed around a typed generative-UI contract, an MCP check-in protocol and tiered approvals, and what is actually built so far.
Read the postForge OS: an evidence chain from idea to merge, with agents that can't weaken the tests
The internal toolchain behind the studio: schema-checked AI skills, a change graph that marks downstream work stale, governance that scales with release stage, and a red-then-green coding loop that rejects any run that touches a test file.
Read the post