The ask-map - how a checkbox becomes a ledger rowHands-on

Every FOR REVIEW paper doc ends with a hidden ask-map: one JSON entry per checkbox strip that says what a mark means and which ledger it writes to. That block is how a pen stroke becomes a decision record without a human reading the page.

A paper doc is HTML rendered to PDF. Before the closing body tag the author embeds <script type="application/json" id="ask-map">. Each key is an ask id (d1-curriculum, d2-prune). Each value has 'means' (what a mark in this strip signifies, including the blank rule) and 'target' (the ledger path or action the mark drives). The renderer ignores the block; the return processor reads it.

When you tag a doc done, rm_pull.py notices on its five-minute poll, downloads the ink layer, and rm_process.py composites the strokes over the page images. It then matches strokes to the checkbox positions the ask-map describes, and writes the answer to the target: a markdown file in PDB/data/decisions/ with frontmatter naming the source PDF. Then it calls report_feedback_processed(), which emits the event and creates no Things task, because you acted.

The rule you inked on 8-31 lives in 'means': blank = re-ask. An unmarked ask on a done-tagged doc is held and asked again, never resolved to the printed recommendation. That is why the REC is printed beside every strip but never assumed.

Worked example

See the map on this week's review doc and the record a return produced:

grep -A 8 'id="ask-map"' /Users/tom/Claude/PDB/data/weekly-review/2026-09-06.html

head -12 /Users/tom/Claude/PDB/data/decisions/jobhunt-briefing-ink-2026-09-06.md

• Three entries, one per decision page. Each 'target' is a real path the processor will write on return.
• The frontmatter of a decision file names source (remarkable), the return PDF path, and status. That is the join back to the ink.

Gotchas

• A strip with no ask-map entry is invisible to the processor: the mark is captured in the composite but never reaches a ledger. The paper-doc skill lists this as the top authoring mistake.
• Written comments on the free-write lines are read by a session, not the deterministic processor. They land in the decision file verbatim but nothing acts on them until a skill runs.
• Renaming a doc on the device breaks the join. The outbox filename is the key.