Raw Upload Content Scale — Scope Note — v0_1

Version. 0.1 Date. 2026-07-02 Status. Open finding, not yet scheduled for build.

The problem

When a file is uploaded to an engagement, its content is stored raw — not interpreted, not chunked, not summarized. This shows up two ways:

1. In the conversation. A readable file's full text is pasted inline into the chat as a single block.
2. In Memory. The file becomes one held assertion containing the entire raw content.

New evidence (2026-07-02)

Operator uploaded 3 files (the Loomworks seed and manifest, ~156K and ~54K characters). Result:

• The chat showed the full raw text of each file inline.
• Held 19 in Memory is the entire seed document as one assertion — not summarized, not broken into meaningful pieces.

Operator's point: this does not scale. At 100 files in a folder, this would flood both the chat and Memory with raw, unstructured dumps.

Why this matters

The seed says Memory accumulates knowledge "with provenance." It does not specify how content is extracted or structured on the way in. This is specification-level, not a seed conflict — but the current behavior (raw dump) will make Memory unusable at any real scale of file upload.

What's needed

Interpretation on upload: extract meaningful content, chunk appropriately, summarize where useful — rather than storing and displaying raw file contents verbatim. The evidence confirms this is not a quality-polish issue — it's a scale-blocking issue.

Status

Not yet scheduled. Recorded so it is not lost. Should be considered before any workflow that involves uploading many files or large documents.