--- title: Cross-Meeting Intelligence Detection description: Protocols for convergence detection, contradiction tracking, and action item lifecycle across meetings tier: 0 used_by: - commands/file-meeting.md (Step 0, Step 4, Step 6) --- # Cross-Meeting Intelligence Detection Protocols for detecting patterns ACROSS the full meeting history. While `cross-relationship-patterns.md` handles within-meeting cross-subsystem scans, this reference handles cross-MEETING intelligence: convergence signals, contradiction tracking, and action item lifecycle. --- ## 1. Action Item Triage Protocol (Step 0) Before starting the filing pipeline, Larry checks for unfinished business from prior meetings. ### Load Open Items Read `room/action-items.md` if it exists. This aggregated file is produced by `compute-meetings-intelligence` and contains all open/done action items across meetings. **If the file does not exist or has zero open items:** Skip Step 0 entirely. Proceed to Step 1 silently -- do not mention action items. ### Present Quick Triage Show open items as a pre-flight check, not an interrogation: > "3 open items from your last meeting. Quick check -- any done?" > > | # | Owner | Task | From Meeting | > |---|-------|------|--------------| > | 1 | Lawrence | Review TAM analysis | 2026-03-15-mentoring | > | 2 | Sarah | Send competitor deck | 2026-03-15-mentoring | > | 3 | Tyler | Schedule user interviews | 2026-03-10-research | > > [mark done: 1,3 / skip / review all] Keep it lightweight. The user is here to file a new meeting, not review old business. Quick in, quick out. ### Handle Responses - **mark done (e.g., "1,3")**: Update each item's status from `open` to `done` in the SOURCE meeting's `action-items.md` file (e.g., `room/meetings/2026-03-15-mentoring/action-items.md`). Find the source meeting by the `Source Meeting` column in the aggregated table. Do NOT edit the aggregated `room/action-items.md` directly -- it gets rebuilt by `compute-state`. - **skip**: Move on to Step 1. No changes made. - **review all**: Show each item individually for yes/no confirmation. Track items marked done for inclusion in this meeting's summary: "Cleared 2 action items from prior meetings." ### Lifecycle Rule Action items follow a simple open/done lifecycle. No intermediate states. No auto-expiry. The user decides when something is done. --- ## 2. Cross-Reference During Filing Protocol (Step 4 Enhancement) During segment filing (Step 4), if open action items remain (not all cleared in Step 0), compare each segment being filed against them. ### Detection Method Use Larry's judgment -- not exact text matching. An action item about "Review TAM analysis" could be addressed by a segment discussing "Updated market size projections." Look for semantic overlap between the segment content and the action item task description. ### Confidence Threshold Only surface when Larry is at least 70% confident that a segment addresses an open action item. False positives here interrupt the filing flow and erode trust. ### Surfacing When a match is found: > "This looks like progress on Lawrence's 'Review TAM analysis'. Mark as done?" If user confirms: update the source meeting's `action-items.md` (same as Step 0 handling). If user declines: continue filing without changes. ### Rate Limiting Surface at most 3 action item matches per filing session. If more are found, batch the remaining ones and present them at the end of Step 4: "I also noticed possible progress on 2 more items. Quick review?" --- ## 3. Cross-Meeting Convergence Detection Protocol (Step 6 Enhancement) After the within-meeting cross-relationship scan (existing Step 6 behavior), perform a cross-meeting scan for convergence signals. ### Scan Protocol 1. **Extract key topics** from the current meeting's filed segments. These become the search terms. 2. **Grep `topics:` fields** in ALL `room/meetings/*/metadata.yaml` files. This is a pre-filter -- fast and cheap. 3. **Count topic frequency** across meetings. For each topic appearing in 3+ meetings (including the current one), generate a convergence signal. 4. **Surface each signal** in Larry's voice: > "Market validation has been raised in 4 of your last 6 meetings. This is becoming a central theme." > "Regulatory risk keeps coming up -- 3 meetings in a row now. Might be worth a dedicated session." ### Convergence Signal Format Each convergence signal includes: - The topic/theme that is converging - How many meetings it appears in (and which ones) - A brief interpretation of what the convergence means ### Recording Convergence signals are written to: 1. The current meeting's `summary.md` in the new `## Convergence Signals` section 2. Tracked in room-level state for compute-state to aggregate --- ## 4. Cross-Meeting Contradiction Detection Protocol (Step 6 Enhancement) After convergence detection, scan for contradictions across the meeting history. ### Pre-Filter Step Load metadata.yaml from recent meetings and identify those that share topics OR speakers with the current meeting. This avoids loading irrelevant meeting summaries. ### Scan Window Load summaries from up to 10 matching prior meetings. If fewer than 10 share topics/speakers, load all matching ones. Never load more than 10 -- context window management per research pitfall #1. ### Detection Method Use Larry's LLM reasoning to compare current meeting claims against prior meeting claims. Look for: - **Position changes**: Same speaker said X before, now says Y - **Speaker disagreements**: Speaker A said X in a prior meeting, Speaker B says not-X now - **Numerical conflicts**: Different figures for the same metric across meetings - **Strategy reversals**: Direction decided in one meeting contradicted in another ### Severity-Based Flagging **HIGH-impact** (financials, strategy, key decisions): Flag immediately during filing with specific references: > "Lawrence said TAM was $190M in the March 15 meeting, but now says $120M. Which is current?" The user resolves high-impact contradictions on the spot. Record the resolution. **LOW-impact** (opinions, preferences, minor details): Collect and include in the meeting summary's `## Cross-Meeting Contradictions` section only. Do not interrupt filing for low-impact contradictions. ### Resolution Tracking When a high-impact contradiction is surfaced and the user resolves it (confirms which claim is current): - Note the resolution in the current meeting's summary - The resolved claim becomes the authoritative version - The prior claim is marked as superseded (noted in the contradiction record) --- ## 5. Enhanced Meeting Summary Sections The meeting summary (`summary.md`) gains two new sections, added AFTER the existing sections (Key Decisions, Insights Filed, Contradictions Detected, Gaps, Action Items, Rejections, Speakers): ### Convergence Signals Section ```markdown ## Convergence Signals {Topics appearing in 3+ meetings across the meeting history. Only include this section if convergence was detected. Skip entirely if none.} - **{topic}**: Appeared in {count} meetings ({meeting-id-1}, {meeting-id-2}, {meeting-id-3}) Interpretation: {what this convergence suggests} ``` ### Cross-Meeting Contradictions Section ```markdown ## Cross-Meeting Contradictions {Contradictions detected against prior meetings. Beyond the within-meeting contradictions in the "Contradictions Detected" section above. Only include this section if cross-meeting contradictions were found. Skip entirely if none.} - **{Claim in current meeting}** vs **{Claim in prior meeting}** Prior meeting: {meeting-id} Speaker(s): {current speaker} vs {prior speaker} Severity: {HIGH | LOW} Resolution: {user's resolution, if provided, or "Unresolved"} ``` --- ## 6. Context Window Management Cross-meeting scanning can consume significant context. Follow these rules: ### Meeting Limit Scan at most 10 prior meetings. For rooms with extensive meeting history, use the 10 most recent. ### Pre-Filter Before Loading Use metadata.yaml grep to PRE-FILTER relevant meetings by topic and speaker BEFORE loading full summaries. Only load `summary.md` for meetings that share at least one topic or speaker with the current meeting. ### Efficient Scanning Pattern 1. Grep all `metadata.yaml` files for topic matches (fast, small files) 2. From matches, identify the most relevant 10 meetings 3. Load only those summaries for detailed comparison 4. Never load all meeting transcripts -- summaries contain sufficient signal ### Graceful Degradation If the room has fewer than 3 prior meetings, skip convergence detection (threshold cannot be met). Contradiction detection still runs against any available prior meetings. --- ## Cross-Reference - Within-meeting cross-relationships: `references/meeting/cross-relationship-patterns.md` - Meeting summary template: `references/meeting/summary-template.md` - Action item format: per-meeting `action-items.md` in meeting archive - Aggregated action items: `room/action-items.md` (produced by `compute-meetings-intelligence`)