01 · The problem
The archive isn't lost. It's forty minutes away.
Your problem was never finding the 2014 presentation. It's that starting the new proposal from memory is faster than the archaeology, so years of accumulated judgment sit permanently just out of reach.
And forty minutes, in practice, means infinitely far. So you write from the top of your head. The work is good, because you are good. It's also a fraction of what you know.
The questions you'd want to put to it are the ones a folder full of files can't answer:
- What have I learned about pricing over fifteen years, and has my position changed?
- Where did the 37% figure in that proposal actually come from?
- What do I currently believe about this, and what's the evidence for it?
- Which of my claims are running on thin evidence?
- Where does my own work contradict itself?
02 · Why the obvious approach fails
Everyone points a model at everything first
It produces a demo that feels remarkable for about a week. Then it starts producing confident garbage.
Flattening
Your flagship framework, an abandoned think-piece, and someone else's deck you saved are three equal documents to a model reading a folder. So it blends them.
Numbers without baselines
The paper says 140 accounts. The workbook says 112. Nobody reconciled the two, and a model will repeat whichever it finds: fluently, in your name, in front of a client.
The agreement machine
A model over your own corpus is a machine for agreeing with you, because you wrote the corpus. The valuable question is the opposite one.
Confidentiality
Your best proof points came from client work, and most can't be repeated as written. You know which is which. Your file system doesn't.
Ask an ungoverned pile: "What have I learned about pricing?" You get a fluent synthesis of forty documents, with no way to tell which ones you still stand behind.
Ask a governed one: "You've used three pricing models since 2011. The current one is value-based, backed by four engagements. Two earlier notes still argue the opposite: flagged, not merged."
03 · How it works
Four steps, about fifteen minutes
Install the plugin
One file, dropped into the Claude desktop app. No terminal, no new account.
Point it at a folder
You choose the folder your vault will live in. Claude sees only what you explicitly hand it.
Answer four questions
What you do, and in what words. That's the whole setup; the vault builds itself around the answers.
It maintains itself
It files what you capture, checks its own numbers, and tells you plainly what it can't verify.
Your archive and the vault stay on your disk, and we never see them. Claude reads what you point it at, the same way it does with anything else you use Claude for. Nothing extra is sent to us or to any third party. It is the reason a governed archive of your own judgment is safe to build at all.
Fig. 02 · Your files and the vault stay on your disk; we never receive them
04 · Where to start
Two ways in
Same discipline, two different doors. Take either one first; they lead to the same place.
Start here
The free Claude plugin
Software you install and run. It builds the governed vault for you, on your own disk, today.
- One quick email, then it's yours
- About fifteen minutes to set up
- Best if you want to see it working first
Or read first
The whitepaper
The thinking behind the system: what breaks when you point AI at your archive, and the discipline that fixes it.
- One email, and the paper opens right away
- A worked example and a five-signal rubric
- Best if you want the reasoning first
05 · The whitepaper
What's in the paper
The plugin runs the discipline. The paper explains it, and makes the case for why it's worth building.
- The claim-and-evidence model we run in live engagements, and the linking rule that stops the two from drifting apart.
- The loop that stops it rotting: why a system's opinion of its own output isn't a usable signal.
- A five-signal rubric for whether any of this is working, from the polite floor to the one signal a demo can't produce.
- A worked example: four items, four formats, fifteen years, and one idea a consultant had been saying since 2009 without writing it down.
Fig. 03 · The paper, open at the five-signal rubric
06 · Fit
Who this is for
Written for
- Independent consultants and advisors whose practice runs on accumulated judgment
- Boutique firm founders whose IP is the asset
- Authors and practitioners with a book, a methodology, or a body of articles
- Anyone who's been asked "where's that number from?" and had to go looking
Probably not for
- Work that doesn't accumulate. If each engagement leaves nothing durable, there's no body of work to govern
- Anyone wanting a five-minute AI demo. This is real software with a real install
- Personal knowledge management. That optimises for what you read; this is about what you produced
07 · Consulting
The plugin is the start, not the offer
It gets you a working vault. What it can't give you is the years of judgment calls that go into keeping one governed as it grows.
That's the part we do directly: what to feed it first, how to resolve the contradictions it finds, and how to turn a governed archive into the proposals and decisions it was built to support.
The Claude plugin
Self-serve. Install it, point it at a folder, it builds and maintains the vault.
Runbooks
An industry's operating procedures, shipped into the vault already written, so the first question is "is this how you do it?" rather than "what do you know?"
Consulting and custom automations
Us, working with you directly on your own system, from first ingest to the decisions it supports, and the automations built on top of it.
08 · Get the plugin
Your Archive Is Not a Memory
Share your email and the download starts right away, plus word when consulting slots open.
The file downloads in this browser the moment you submit. Nothing is emailed to you, and nothing else unless you ask. No list, no sequence, no unsubscribe theatre.