Skip to content

The Steward Guide

Stewards are the humans of last resort in the clarification loop: when routing finds no one, when authorized people disagree, or when a correction needs an accountable override. If you hold the steward role, this is your job description inside Telha.

What lands on your desk, and why

You receive Because Your move
A redacted question (document link, no text) No candidate passed the permission gate, so nobody could be shown the passage Fix the underlying gap (grant access, correct ownership) or answer from the source directly
A conflict notice Two authorized people gave different answers; the fact is frozen uncorrected Review both answers and the source, then bind an override
A digest of dead-lettered questions Every ranked candidate was exhausted (unanswered, unreachable, or skipped) Scan the skip reasons; they tell you exactly what to repair
A quorum stall A critical fact has one confirmation and needs a second Confirm it yourself if appropriate, or nudge the second candidate

The daily loop

telha clarify queue                    # your attention queue, worst first
telha clarify show <id>                # full story: candidates, deliveries, skips, answers
telha clarify review <id> --bind <n>   # steward override: bind candidate n on a conflict

Everything you do is recorded append-only. An override does not erase the conflicting answers; it sits on top of them with your name on it, and the full disagreement remains visible in the fact's evidence forever. That is by design: your signature is worth something precisely because it cannot quietly rewrite the past.

Reading skip reasons (your best diagnostic)

Skip codes are how the system tells you what is broken upstream:

  • NO_READ_PERMISSION: the best-placed person cannot see the document. Either the ACL is wrong at the source or the workspace role is. Fix whichever is lying.
  • NO_DELIVERY_ALIAS: the person exists but has no reachable chat identity. Usually the Teams app is not installed for them, or their Slack email does not match their corporate identity. The audit hint says which.
  • ALIAS_AMBIGUOUS: two people match a delivery identity. Identity hygiene issue; check the person records.
  • DAILY_CAP_EXCEEDED / DIGESTED: not a problem, just pacing. The question queues.
  • RESPONDER_NOT_AUTHORISED: someone answered from an unverified identity. Their answer was kept as a non-binding attestation; if it is right, confirm it yourself with telha clarify answer.

A recurring pattern of the same code is an infrastructure fix, not a per-question fix: install the app org-wide, run the identity sync, correct the source ACL.

Judgment calls, with house positions

  • Attestations from viewers. A viewer's answer binds nothing until an editor or steward confirms it. If the viewer is clearly the person who knows (they wrote the document), confirm; the record will show both the attestation and your confirmation.
  • Post-binding conflicts. If someone disputes a fact after it bound, the correction stands, the dispute is recorded, and you are notified. If the dispute convinces you, bind a new correction; history keeps both.
  • When NOT to override. If neither answer is obviously right, the honest state is conflicted. A frozen pending fact with a visible caveat is better than a confident wrong one. Escalate to the document owner outside Telha if needed; answer when you actually know.

Boundaries of the role

Stewards resolve conflicts and bind overrides. Stewards do NOT: see content their own permissions do not allow (redacted questions exist for exactly this), edit history, bypass quorum silently (an override is itself a recorded, attributed act), or manage retention/erasure (that is the operator's dual-approval domain, deliberately separated from yours).