HomeOverviewHow it worksPricingFAQ
Request access

System Scope — Technical Appendix (Canonical)

Internal-only technical reference. Source-of-truth for implementation and future API / schema design.

System scope

Observed surfaces and continuity signals across connected tools.

SignalsArtifactsDependenciesContinuityAgreementsOutputsIntegrationsIdentityConstraints

Signals captured

  • Message events: create/edit/delete, reactions, mentions, threads, reply chains
  • Decision language cues: commitments, approvals, reversals, blockers, ownership statements
  • Intent/priority cues: urgency markers, deadlines, "next step" statements
  • Context cues: links, attachments, referenced artifacts, quoted text, forward references
  • Temporal cues: when a decision was made vs when work occurred elsewhere

Artifacts tracked

  • Tickets/issues: state transitions, assignee changes, priority changes, comments
  • PRs/commits: references to tickets, reviewers, approvals, merge state, reverted work
  • Docs/pages: edits, comments, headings referenced in discussions
  • Deploy/releases: version markers, environments, timestamps, rollback signals
  • Incidents/alerts: creation, escalation, resolution notes, postmortems

Dependencies

  • Cross-tool entity resolution (same "thing" represented in multiple systems)
  • Link extraction and normalization (ticket ↔ PR ↔ doc ↔ message)
  • Relationship types:
  • "supports / depends on"
  • "supersedes / replaces"
  • "blocked by / waiting on"
  • "decided in / implemented in"
  • Confidence scoring for inferred links vs explicit links
  • Drift detection in the graph (links that go stale, missing edges, conflicting edges)

Continuity and drift detection

  • Dropped-thread detection: decision exists with no downstream execution artifact
  • Orphan work detection: execution exists with no upstream decision trail
  • Context decay: key rationale present in chat but absent in ticket/doc
  • Decision drift: original intent conflicts with later actions or ticket changes
  • Ownership drift: implicit owner vs actual assignee divergence
  • Scope creep signals: expanding requirements without corresponding decision update
  • Repeat-loop signals: recurring questions that indicate missing canonical context

Agreement preservation

  • Canonical "what we agreed" extraction (short, source-linked)
  • Change-of-mind detection (explicit reversals, re-litigation)
  • Conflict surfaces: two simultaneous "truths" across tools, competing priorities without resolution
  • Timestamped agreement snapshots (what was true at time T)

Outputs and delivery

  • Digest outputs (quiet, periodic): unresolved drift, orphan work, missing rationale
  • Event-driven outputs (sparse): a decision is at risk, a dependency edge broke, a thread dropped
  • Per-surface views (optional later): "Decision trail" for a ticket/PR without building a dashboard
  • Deep links back to source-of-truth artifacts (no re-hosting content)
  • Alert semantics:
  • "Needs linking" (missing edges)
  • "Decision not implemented" (agreement with no execution)
  • "Work not justified" (execution with no decision trail)
  • "Conflicting truth" (tool states disagree)
  • "Rationale missing" (implementation changed without updated decision context)
  • Noise controls: rate limits, collapse duplicates, confidence thresholds, user-level suppression rules

Integrations surface (initial + expandable)

  • Chat: Slack / Teams-style message graphs (channels, threads, mentions)
  • Work tracking: Jira / Linear-style issues and state changes
  • Code: GitHub/GitLab-style PR/commit references and approvals
  • Docs: Notion/Confluence-style pages and edits
  • Alerts/ops: incident and monitoring event streams
  • Identity: user mapping across tools (email/SCIM where available)

Identity and access

  • Actor resolution across systems (same person, multiple accounts)
  • Workspace/project boundaries respected (no cross-boundary inference)
  • Role-aware visibility (only show what the viewer can already access in-source)

Constraints and non-goals

  • No new workflow requirement
  • No attention-seeking dashboard by default
  • No "engagement" mechanics, no activity scoring for its own sake
  • No selling or sharing user data; no surveillance monetization
  • Data minimization: store only what is required to preserve continuity and link trails
  • Explainability: every surfaced claim links back to source events/artifacts

Future depth (kept terse)

  • Custom policies: define what counts as a "decision," "done," "drift"
  • Team-specific ontology: map labels/states to meaning
  • Org-wide dependency atlas (opt-in, permissioned)

Scope evolves; principles stay fixed.