01 — CONTEXT: Post-Session Production
Input definitions, validation rules, and what each step requires.
Mode 1 Inputs
| Input | Required | Source | Used For |
|---|---|---|---|
| Session transcript (.txt) | Yes — blocking | Client repo sessions/ or Google Drive AMeeting TranscriptsRaw/ | Ground truth for all content |
| Reference data file | Yes — blocking | Client repo reference-data/ | Proper nouns, name spellings, transcript overrides |
| Prior advisory email | Yes | Client repo archive/ | Voice and format reference |
| Constraint brief or CPM | Recommended | Client repo constraints/ | Context for what the engagement is working on |
| Project plan(s) | Recommended | Client repo projects/[initiative]/project-plan/ | Build statuses, what's in progress |
| Master plan | Recommended | Client repo master-plan/ | Strategic framing |
| Session JSON | Optional | Client repo sessions/ | Speaker timing, metadata for ambiguous attribution |
| Relay.app recap | Do not use as source | Google Drive BMeeting RecapsFull/ | Cross-reference only — never as primary source |
Validation Rules — Mode 1
- Transcript must exist. If only a recap or JSON is available, stop and ask the advisor. Do not proceed with a recap as the source.
- Reference data must be read first. Before reading the transcript. Before reading anything else. Every proper noun in the output must match reference data.
- Prior advisory email must exist. If this is the first email to a new client, flag it — the advisor needs to establish the voice baseline.
- Transcript overrides in reference data take precedence. When the transcript renders a name incorrectly (speech-to-text artifacts), the reference data correction wins.
- Do not use the relay.app recap as a content source. It is one generation removed from what was said. Using it makes the output two generations removed. Cross-reference only — to check for topics you may have missed, never as the basis for factual claims.
Mode 2 Inputs (Improve)
| Input | Required | Source | Used For |
|---|---|---|---|
| The email as sent (after advisor edits) | Yes | Client repo archive/ | Updated golden example |
| Advisor feedback | Yes | Conversation | What to fix in the kit |
| QC results from the failed run | If applicable | QC output | What the checklist missed |
Validation Rules — Mode 2
- Compare the sent email to the drafted email. Every difference is a signal.
- If the advisor changed the voice — update the golden example and the voice rules in the build skill.
- If the advisor added or removed content — determine whether the build skill should have caught it.
- If QC missed something — add the check to file 04.
Input Priority Hierarchy
When inputs conflict, this is the order of authority:
- Reference data — wins on all proper nouns, always
- Transcript — wins on what was said, what was committed to, what happened
- Advisor (Kathryn) in conversation — wins on framing, priority, strategic intent
- Prior advisory email — wins on voice and format
- Project plan / CPM / master plan — wins on engagement context
- Session JSON — supplementary, clarifies ambiguous attribution
- Relay.app recap — cross-reference only, never authoritative
Content Filtering: What Goes In vs. What Stays Out
The advisory email transforms a full session transcript into strategic client communication. Not everything discussed belongs in the email.
In the Email
- Decisions made or confirmed during the session
- Wins and progress — factual, not editorialized
- Topics covered — compressed to one line each
- Action items — specific, grouped by owner, tied to builds or engagement work
- Next session date and proposed agenda
- Items the client needs to review, decide on, or bring to the next session
NOT in the Email
- Client's own business operations (their delivery work, their client management) — unless it became a constraint
- Internal advisor observations about the client's behavior or growth
- Coaching language, motivational framing, or narration of behavior
- Detailed sub-bullet recaps that mirror the transcript
- Topics that were just check-ins with no substance
- Items that require more context than the email can provide (those go on the agenda)
- Anything the client already knows because they were there — the email advances the engagement, it doesn't replay it
The Agenda Test
For every item you consider including in the email body, ask:
- Is it actionable before the next session? If no → agenda, not email.
- Does it require more context than the email can provide? If yes → agenda, not email.
Gap Protocol
A gap in this kit is different from an SOP gap. Here, a gap means: information needed to produce the email that isn't in the transcript.
Common gaps:
- A commitment was discussed but not confirmed (did the client actually agree, or did they hedge?)
- A date was mentioned but ambiguous ("next week" with no specific day)
- An action item's owner is unclear
- A name or organization referenced in the session isn't in the reference data
The rule: Flag the gap to the advisor. Do not guess. Do not fill in what seems likely. A wrong date, a misattributed action item, or a misspelled name erodes trust faster than a missing item.