Generated from smart-base/skills/content/authoring-who-smart-guidelines/transcription-verbatim.md — do not edit here.

✎ Edit this page’s source

transcription-verbatim

A step of recommendation-extraction, for a source whose recommendations are printed only as images. When recommendation-candidates finds no normative sentence in the entry’s text layer, the L1 is authored by hand (*.l1.yaml, the measles pattern) from the inspected image narratives and transcriptions of the rendered page. This step is the check on that authoring. recommendation-extraction’s section An authored L1 from pictures holds the judgement; this one is the mechanical gate.

Input and output

• Input: input.schema.json
• Output: output.schema.json

In: the directory of authored *.l1.yaml, each naming its library entry in graph.source.entry. Out: a report: one summary line per file, one FINDING line per failure. Exit 0 clean, 1 findings, 2 nothing to check (which is not a pass).

Steps

1. Author the *.l1.yaml from the entry’s images.json narratives and its transcription blocks; never from the picture read by eye with no transcription recorded.
2. Run the Tool. A finding (a statement not verbatim, a transcription word the text layer does not corroborate and inspectionOnly does not list, a duplicate key) goes back to step 1.
3. Hand the result to recommendation-scenarios, which reads an authored L1 YAML as its l1.

That the transcription reads the picture correctly is the reviewer’s call (bun run cat narratives), not this check’s.

Tool

l1-transcription-verbatim (smart-base/tools/l1-transcription-verbatim/l1-transcription-verbatim.ts):

bun run smart-base/scripts/l1-transcription-verbatim.ts --dir <dir of *.l1.yaml> [--root <repo>]