💡 Cross-Session Impact Tracking: Avatar-Reported Output Utilization as Retention Engine

[github-actions[bot]]

Summary

The avatar tracks whether outputs from past sessions were actually USED — files opened, documents shared, PRs merged, work items updated — and reports impact naturally on the user's return. "The brainstorm document from last Tuesday was opened 4 times and shared with 2 team members. Want me to build on that momentum with a follow-up analysis?" This transforms the avatar from an output generator into an outcome partner, creating a progress-and-impact feedback loop that drives daily return usage.

Market Signal

No AI assistant currently tracks post-output utilization — this is entirely unoccupied territory. Gamification research consistently shows that achievement tracking and progress visibility are the strongest retention mechanics. The personal AI assistant market has split into model runners, developer frameworks, and identity-layer assistants (per Vellum, Zapier 2026 roundups) — impact tracking deepens the identity/relationship layer that distinguishes TalkTerm. The AI avatar market's 33.1% CAGR signals demand for embodied AI that builds lasting relationships, not transactional tool usage.

User Signal

The PRD targets ≥1 TalkTerm session/day within 30 days — the key retention metric. However, the current architecture has no mechanism to drive re-engagement beyond session resume (FR31, FR33). No existing discussion covers post-session output impact measurement. Related discussions #234 (Proactive Workspace Monitoring) and #211 (Contextual Workspace Intelligence) monitor project STATE, not OUTPUT UTILIZATION — the distinction is between "what's happening in your project" and "what happened with the work I produced for you."

Technical Opportunity

TalkTerm already writes output files to the local filesystem (FR25) and can commit to Git (FR55). File access timestamps (fs.stat mtime/atime) and git log provide utilization signals for TalkTerm-created artifacts. The session/audit SQLite database (main/storage/) can store output artifact paths and track their lifecycle. The avatar greeting flow (FR33) already narrates session history — impact metrics slot directly into that greeting. The DomainEventBus architecture pattern enables an artifact:utilized event that any component can subscribe to.

Assessment

Dimension	Score	Rationale
Feasibility	high	Builds on existing file I/O and session persistence infrastructure; requires artifact path tracking and lightweight fs monitoring
Impact	med	Novel retention mechanic — no competitor does this; directly serves the PRD's daily-usage target
Urgency	med	Most valuable once users are producing real outputs (post-MVP workflow completion); design decisions can be made early

Adversarial Review

Strongest objection: File access timestamps are unreliable (automated indexers, antivirus, backups trigger them), and tracking file access patterns could feel surveillance-like to privacy-conscious users.

Rebuttal: Only track files that TalkTerm itself created (paths stored in session records), not arbitrary filesystem browsing. Use high-confidence signals: git log (was the file committed/pushed?), file modification time (was it meaningfully edited?), explicit sharing actions (was it attached or uploaded via MCP?). Frame as celebration, not surveillance: "Great news — your team picked up that analysis!" Users can opt out via a single preference toggle. The avatar relationship makes monitoring feel caring ("I noticed your work had an impact") rather than intrusive.

Suggested Next Step

Define the output artifact lifecycle model (created → accessed → modified → shared → committed). Identify reliable utilization signals available on macOS and Windows without elevated filesystem monitoring. Design the avatar greeting integration that naturally reports impact metrics alongside the existing session resume flow (FR33).