In 2014, I walked into 180Fusion and found thousands of outstanding tasks — possibly tens of thousands — for a book of business of roughly 2,000 clients. The team wasn't just behind. They were falling further behind every month, because the backlog wasn't a pile to work down once. Recurring monthly and quarterly work kept adding to it faster than anyone could clear it.
Within two months, the backlog was gone. SLA compliance went from 19% to 100%. The processes I built cut labor needs by 80%. I didn't fix it by working harder or hiring more people. I fixed it by changing what the system remembered and where it remembered it, so the work could be seen and the process could run without someone holding it all in their head.
The architecture of remembering
Twelve years later, I built the Brain Vault for the same reason: to write the memory down where it can be worked.
It's a local-first Obsidian vault with 1,765 interlinked documents — 869 synthesized, 822 raw imports — organized into 50 concept-pillar documents with 7,075 links between them. It has a routing layer that tells an AI agent where to start, 17 playbooks that define named routines (imports, interlinking, health checks, weekly review), and a set of operating documents: live priorities, operating assumptions held as confidence-scored hypotheses, an energy map for scheduling against actual focus, and an accomplishment ledger built specifically as a deliberate counterweight to undercounting your own output.
This is largely designed as a support for my ADHD. This is an accessibility tool, and a highly effective one.
A notes app is built for retrieval: where did I put that? The Brain Vault is built for connection and action: what does an agent need to know right now to act on real context?
Two layers, on purpose
The vault has a raw layer that holds imported source material and is never edited, and a synthesized layer — concept pillars, notes, timelines — written on top and linking back. An import pipeline splits a batch into one file per item, removes near-duplicates at a 0.80 similarity threshold with every decision logged, then scores each item for signal strength. 96% of documents carry structured metadata in frontmatter; undated material gets an inferred date and a stated confidence level, not a guess dressed as fact.
A standard-library Python script builds the link graph and runs PageRank to score importance across the synthesized layer. The same scores drive a 3D visualization plugin I built — "Cosmos" — that maps domains to stars, concept pillars to planets, and evidence notes to moons. The plugin is built, with 31 tests and one feedback round still open, but not shipped yet. Scene fog was washing the additive sprites to black; no test caught it, I found it by looking.
What stalling looks like from the outside
Separately, the game I'm building in Unreal Engine has a commit history that shows two days of intense scaffolding in July, then a 63-day stretch with zero commits, then 257 commits in two weeks once the blocker — a missing engine capability — cleared.
From outside, that 63-day flatline is indistinguishable from an abandoned project. The only way to tell a stall from a block is knowing precisely what the blocker was. I knew. The vault knew. The git log doesn't.
What carried over
Both the 180Fusion backlog and the 63-day gap look like chaos from the outside. Both were actually a specific, nameable condition — a process that depended on memory, a missing capability — that a system could be built to track.
The Brain Vault is the general-purpose version of what the 180Fusion fix did once, by hand, for one team's workload. It externalizes the structure my brain doesn't hold natively: what's blocked and why, what's stalled versus what's parked, what I've already accomplished (because I undercount it), and what the agent needs to know to act without me having to remember and re-explain it every time.
Thriving in chaos doesn't mean tolerating disorder. It means building the external structure that turns chaos into something ordered and workable — writing the memory down where it can be worked, whether that's a process for 2,000 clients or a second brain that an AI agent can navigate.