BIG Tools
Sign Up

Recipe: Tacit-Context

Five questions about what you already know — written down before any AI tool runs against this idea. Externalizes Unknown Knowns so AI verifies and enhances rather than generating.

What it does

Five questions about what you already know — answered before any AI tool runs against this idea. The gate exists to surface your Unknown Knowns (Žižek): things you tacitly know but haven't said out loud. Once externalized, the responses are prepended to every tool's system prompt as user-stated context, so the AI verifies and enhances rather than inventing.

Why per-idea (not per-session)

Tacit context belongs to the idea, not the work session. One row per insight. You can update it as your understanding shifts; previous responses become the audit trail.

The five questions

  1. What do you already know about this idea that's not yet written down?
  2. What are you quietly assuming that you haven't stated?
  3. What's the failure mode you're already imagining, even if you haven't said it?
  4. Why does this idea matter to you specifically?
  5. What would have to be true for this to fail? Have you said that out loud?

How tools use it

Every gated tool (Alignment, Reframe, Collaborate, Mixtape) prepends your responses to the LLM's system prompt as a <user_context> block. The model is instructed: "Treat user_context as fact stated by the user; do not generate new content for these fields; verify or enhance only."

When to update

  • A new failure mode shows up that you didn't write down originally.
  • You realized an assumption you'd been making was wrong.
  • The idea's "why for me" sharpened.

Source

Wednesday spine §B.2 (audit's classroom form). Refinement expected post-cohort — the question set is stored as jsonb so it can evolve without a schema migration.