Your Signal Library Belongs to You: Here Is the Export Button

Most of the value in a marketing operating system is not the software. It is the graded corpus — the 40 documents you pulled from a client's Drive, the ones you tiered as signal, the ones you sent to trash, the three you kept as glints because they will make good content later even though they say nothing about the ICP. That grading is your judgment, applied over hours you mostly did not bill.

If that judgment lives only inside a vendor's database, you have a problem. Not a hypothetical one. A fractional practice with four clients has already rebuilt its stack four times; rebuilding a fifth time because a tool sunsets, gets acquired, or raises price 3x is a cost you can put a number on — roughly [add hours per client re-onboarding] × your unbillable rate, per client, per switch.

So here is the position: you should be able to leave T2D3 OS in one click, and take the intelligence with you in a format any LLM can read. That is what the Signal Pack export does.

A vendor that tells you exactly how to leave is making the only trust claim that costs it anything

Every platform in this category says "your data is yours." Almost none of them say what shape it comes out in. A CSV of document titles is not your corpus. An API you need a developer to call is not portability for a one-person practice.

The claim that costs something is this one: we will hand you the graded, annotated, human-reviewed artifact — the thing that took you the hours — in a format that works in ChatGPT, Claude, Gemini, or whatever ships next quarter, without us in the loop.

That is a real reduction in our switching-cost moat. We think the moat should come from the methodology being genuinely better at multi-client context switching, not from your evidence being trapped. Lock-in is a moat you rent. Usefulness is a moat you own.

What comes out of the export button

The export is not a database dump. It carries the human layer.

• Per-document markdown with YAML frontmatter — the source document, plus its tier, its votes, who graded it, the notes left on it, and its status (signal, noise, glint). The grading travels with the document instead of sitting in a table somewhere.
• JSONL — the same corpus in one machine-readable line-per-record file, for anyone who wants to pipe it into their own tooling or embed it.
• CONTEXT.md — a paste-ready context file. Open a fresh LLM session, paste it in, and the model knows this client's ICP, personas, value props, and which evidence backs them.
• Filtered exports — you don't have to take everything. Filter to messaging documents, or one persona, or one engagement, and export only that slice.

The extracted insights come too: the ICP, the personas, the value-prop pillars, and the documents each one was drawn from. That last part matters more than it sounds. An unsourced persona is an opinion. A persona with the six interview excerpts underneath it is a defensible position in a room with a skeptical CEO.

What a portable marketing evidence format should actually contain

Take this as a spec, and hold every vendor in your stack to it — including us.

1. The source text, not a link to it. Links rot and Drive permissions change.
2. The grade and who gave it. Signal, noise, or held-for-later — plus the votes and the reviewer.
3. The human notes. The sentence you wrote at 11pm explaining why this deck matters is the highest-value token in the file.
4. The derived assets and their provenance. ICP, personas, value props, each tied to the documents that produced them.
5. A prose context file. Because in practice you will paste, not integrate.
6. Open, boring formats. Markdown and JSONL. No proprietary container, no required runtime.

If a vendor can't produce items 2, 3, and 4, they are holding your judgment hostage and calling it a feature.

The honest limits

This is beta, and I would rather say it here than have you discover it. The export reflects what you graded — if the corpus is thin, the pack is thin. One prospect put it plainly: "there's only three documents really that apply to the persona, so you may want to have more before you start." The export does not fix that. It just makes sure the work you did do is yours.

And a pasted CONTEXT.md is not a live system. It has no versioning, no lock, no single source every channel reads from. That is the part you would lose by leaving — which is exactly where we think the moat belongs.

Try the export before you trust the platform

If you are evaluating a fractional CMO operating system, do this on day one, not month twelve: load one client's real documents, grade them, then hit export. Read what comes out. Paste the context file into a fresh LLM session and see whether it actually knows your client.

If it does, you have a system you can leave. That is the only reason to stay in one.