Operator field guide · Change-impact map

Map each Help Center revision to the evidence it can actually support.

Renames, merges, factual corrections, and policy clarifications leave different trails. Use this map to choose the stable identifiers, time boundary, and human review needed for each change—without searching customer message text.

Use this when
You are scoping the impact of one or more Help Center changes across exported AI-support conversations.
Leave with
A reproducible source map, an honest coverage statement, and a queue that keeps exact evidence separate from unresolved source history.
Data boundary
IDs and source references only. Remove message bodies, names, emails, attachments, secrets, and internal notes before import.

The workflow

A reviewable incident path, not an automated customer action.

  1. 01

    Classify the revision

    Separate a factual correction from a clarification, policy replacement, URL rename, article merge, or deletion. The type of change shapes the review, but never proves a prior answer was wrong.

    Operator checkWrite one sentence describing the changed fact or boundary without copying private article content into the export.

  2. 02

    Bind stable and historical identifiers

    Use the stable article ID and current canonical URL. For a known rename or merge, list each verified historical URL as an explicit alias to one article.

    Operator checkReject an alias that is unknown, shared by multiple articles, or based only on a similar title.

  3. 03

    Set the exposure interval

    Use the conversation's first-message time and the article's effective change time to identify pre-change exposure candidates.

    Operator checkConfirm timezone normalization and the inclusive or exclusive boundary before running the queue.

  4. 04

    Measure exported coverage

    Count exact rows, review-required rows, post-change conversations, test exclusions, missing-source records, unmatched references, and invalid records separately.

    Operator checkCoverage is limited to the files supplied. A zero queue is not proof that no affected conversation exists.

  5. 05

    Hand off a reproducible receipt

    Preserve the minimized input hashes, settings, aliases, interval, counts, and canonical queue so a reviewer can repeat the same run.

    Operator checkKeep downloaded artifacts under the buyer's control and do not attach customer content to the receipt.

Decision boundary

Match the evidence method to the kind of change.

Observed signalQueue treatmentHuman action
Factual correction at the same canonical URLMatch that URL before the effective timestampReview the original cited answer against the corrected fact.
Verified article rename with one historical URLMatch canonical URL plus explicit aliasConfirm the alias history and review both evidence groups together.
Several old articles merged into one replacementOne explicit alias per verified source; conflicts stay REVIEW_REQUIREDReview each source lineage instead of assuming every redirect is equivalent.
Policy clarification with no factual reversalQueue evidence normally; do not label it an errorLet the policy owner decide whether any prior answer needs follow-up.
No source reference or incomplete date coverageCoverage gap, not a matchExpand the export only through an approved minimized-data workflow.