Behaviour spec notes

One file per behaviour, hand-authored and implementation-blind. Each was distilled from the upstream ExcelJS backlog during the harvest (ADR-0007): a real scenario someone hit, a statement of what the library should do about it, and the questions that were still open when the note was written. They describe behaviour, never a design, which is why they outlived the codebase they were extracted from.

A note is not a corpus case. A case is executable and pins the library down; a note is the evidence a case was written from, and a note whose questions are still open is work that has not been done. Read one to learn what Excel actually does in a corner; read test/corpus/ to learn what this library is held to.

Most notes carry the same three headings:

Heading	What it holds
Scenario	The situation as the person who reported it met it
Desired behaviour	What should happen, stated without reference to any implementation
Open questions	What was undecided when the note was written

A Cluster: line near the top groups a note with its neighbours (streaming, styles, types, images and about thirty more). It is a reading aid, not a schema.

When a decision needs grounding in the format itself rather than in a report, these notes are the third tier, after the vendored ECMA-376 schema graph and the Microsoft Learn search: see the agent correctness playbook.