what-an-extraction-may-do-without-a-person

Set the threshold above which extracted values are accepted unreviewed from measured accuracy on real documents, after establishing that the results can be written anywhere at all.

Procedure

  1. start start. Goes next to s1
  2. decision Can the results be written at all?. Goes if no_route to s2; if default to s3
  3. action Redesign around the route that exists, RO (read-only). Goes on_fail to s9; next to __end_aborted
  4. check Who owns the credentials and the entitlement. Goes on_fail to s9; next to s4
  5. end aborted. Terminal
  6. action Run the bad documents too, RO (read-only). Goes on_fail to s9; next to s5
  7. action Set each threshold from that measurement, RO (read-only). Goes on_fail to s9; next to s6
  8. action Review everything, for now, RW (reversible-write). Goes on_fail to s9; next to s7
  9. decision Has a field earned it yet?. Goes if material to s8; if not_yet to s11; if default to s10
  10. action Get the decision signed, or keep reviewing, RO (read-only), requires approval. Goes on_fail to s9; next to s10
  11. wait Keep reviewing, and measure again later. Goes retry to s7; next to s9
  12. action Accept that field unreviewed, and watch it, RW (reversible-write). Goes on_fail to s9; next to __end_success
  13. escalate Escalate to whoever owns the target system. Terminal
  14. end success. Terminal

Open the graph on its own page — larger, printable, and an address you can send.

Walk it without running it — you answer what your environment would answer, and the path, the capabilities, the approvals and the distance back to an undo come out of the document. Nothing executes; there is no Run button here and there will not be one.

Trust

T2 · attested 2026-10-11 · runbooks.directory

Nobody has attested this record. That is the ordinary case: T2 is what the catalog can check on its own, and T3 needs somebody to run the procedure and sign for it.

  • Link liveness has not been checked: it needs a network and the catalog builds without one. Run pnpm gates --online where there is one.

Steps

  1. s1 decision — Can the results be written at all?
  2. s2 action — Redesign around the route that exists (read-only)
  3. s3 check — Who owns the credentials and the entitlement
  4. s4 action — Run the bad documents too (read-only)
  5. s5 action — Set each threshold from that measurement (read-only)
  6. s6 action — Review everything, for now (reversible-write)
  7. s7 decision — Has a field earned it yet?
  8. s8 action — Get the decision signed, or keep reviewing (read-only) — requires approval
  9. s10 action — Accept that field unreviewed, and watch it (reversible-write)
  10. s11 wait — Keep reviewing, and measure again later
  11. s9 escalate — Escalate to whoever owns the target system

Capabilities

This is a privilege request. Check it against your own allowlist before running.

    Provenance

    Source
    https://github.com/microsoft/ai-agent-runbooks/blob/083f33eec889/02-patterns/Intelligent-Document-Processing-Pipeline/Intelligent-Document-Processing-Pipeline-Runbook.md
    Fetched
    2026-09-04
    Source last checked
    not since it was captured on 2026-09-04 — recorded as current
    Licence
    MIT
    How it got here
    Rewritten here from the source above, under its licence. The words and the steps are this catalog's; the procedure it describes is not this catalog's invention, and the source is where to check it.
    Highest risk in this procedure
    reversible-write

    Take it with you

    Run it under supervision

    Runs here, on your machine, under supervision: every step is proposed before it happens, an approval gate stops for you, and a failed expectation goes where the procedure says rather than on to the next step. Nothing is sent anywhere. To supervise an agent that is doing the work instead, runbooks hook sits in its tool-call path.

    npx @runbooks/cli run adapted/what-an-extraction-may-do-without-a-person@1.0.0 --as "$(whoami)" \
  --input target_system=<where the extracted values are supposed to end up>