Assembly Is a Consultation, Not a Rewrite

"Pippa proposes clusters, an ordering, and prose — surfacing what it could not interpret rather than smoothing over it."

Propose, Don't Overwrite History

Assembly is a Pippa consultation over a selected set of crumbs, not an automatic batch job. The flow is deliberately human-in-the-loop: you select a day in the active journal and invoke Assemble; Pippa reads the crumbs in scope plus relevant prior drafts and history; Pippa proposes clusters, an ordering, and prose; and then you decide — accept it, edit it in place, hand it to Rekindle for deep editing, or discard it entirely. At no point does Pippa silently rewrite your history into polished prose and save it. A proposal you didn't accept changes nothing.

That posture matters most for a memory tool. The whole value of a journal is that it holds what actually happened. A system that quietly 'improves' your record — smoothing a rough note into a nicer sentence, resolving an ambiguity by guessing — is corrupting the evidence in the name of helpfulness. Consultation keeps the evidence sacred: the crumbs stay exactly as written, and only the derived draft is up for negotiation.

Surface what you couldn't interpret; never smooth it over. The honest move when a fragment is ambiguous is to flag it — 'I wasn't sure what this crumb meant' — not to invent a confident reading. For a memory system, a visible gap is a feature; a plausible fabrication is the worst possible bug, because it looks exactly like a real memory.

The Anti-Hallucination Stance

This is where a journal built on an AI has to be most careful. A crumb might read the thing with M. finally made sense. A careless assembler would expand that into a fluent paragraph about a resolved conflict, inventing detail to make the prose flow. Vesta's consultation does the opposite: it surfaces the crumb as something it could not fully interpret and leaves the interpretation to you. The draft is allowed to be interpretive, but it must never pretend to certainty the crumbs don't support. Better a proposal that says 'here's what I could assemble, and here are three fragments I couldn't place' than a beautiful draft quietly stitched from guesses.

Agency Stays With the Author

Every exit from the consultation keeps you in control. Accept and the draft persists with full lineage. Edit and it's your words on top of the proposal. Send to Rekindle and you get a real editor for the deep work. Discard and it's as if the assembly never happened — the crumbs are untouched, ready for a different attempt. Pippa is the consultant who drafts on request; you are the author who decides what becomes part of the record.

Code

A proposal that admits its gaps, instead of faking them·json

{
  "proposed_draft": "A slow morning; the garden finally turned.",
  "source_crumb_ids": ["01J..a", "01J..b"],
  "clusters": [
    { "theme": "morning", "crumbs": ["01J..a"] },
    { "theme": "garden",  "crumbs": ["01J..b"] }
  ],
  "could_not_interpret": [
    { "crumb": "01J..c", "text": "the thing with M. finally made sense",
      "note": "unclear referent — left out of the prose, flagged for you" }
  ]
}
// the ambiguous crumb is surfaced, NOT expanded into invented detail

Exercise

Imagine an AI feature that turns your rough notes into finished writing. List two ways it could be 'helpful' that would actually corrupt your record — a smoothing, a guess, an invented connection. Then rewrite each as a consultation move that surfaces the uncertainty to you instead. Notice how the honest version is less impressive and more trustworthy.

Hint

The corrupting moves all share a shape: they resolve an ambiguity on your behalf and present the result as fact. The consultation fix is always the same — show the ambiguity, name what couldn't be interpreted, and let the human decide. Less magic, more truth.

Assembly Is a Consultation, Not a Rewrite

Propose, Don't Overwrite History

The Anti-Hallucination Stance

Agency Stays With the Author

Code

External links

Exercise

Progress

Comments 0