Why we built this
Tool dashboards can show adoption. Git can show output. Jira can show delivery. Pull requests can show collaboration. The engineering decision lives in the relationship between those events, but that relationship should not require collecting the work itself.
Evidence checked: 2026-07-16
01 / problem
A token total cannot tell you whether work merged. A commit cannot explain the AI cost behind it. A completed issue cannot show where review or deployment slowed down. The useful answer lives in the relationship between those events.
02 / trust
An agent installed on every workstation creates a real power boundary. We built the collection contract first: structured metadata only, no prompts, no responses, no source code, no review text, no issue descriptions, no CI logs, and no secrets.
03 / evidence model
Clavinci joins assistant sessions, token classes, models, caches, and cost to commits and Jira issues, then carries the relationship through pull requests, CI, deployments, incidents, DORA, sprint flow, review health, AI quality, and 30-day code survival. Attribution is versioned so the result can be replayed and corrected.
04 / trade-off
Excluding prompts, responses, source, comments, descriptions, and logs means Clavinci cannot perform content-aware code review or reconstruct the conversation. We accept that limitation because a metadata-only measurement system can still answer outcome questions without becoming a searchable archive of engineering work.
05 / decisions
Collection begins close to the engineering tools and sends only the structured event contract to the hosted service.
Session, commit, issue, and pull request relationships can be recomputed instead of accepted as an opaque score.
Direct-manager access is explicit. Every individual read creates an immutable row visible to the engineer.
Skip-level views begin with team aggregates and privacy floors rather than individual drill-down.
The anti-metric blocklist prevents ranking and automated people decisions before data reaches a view.
Clavinci operates the product and application stack as a paid service. Customers retain control over access, integrations, retention requirements, and governance.
Evidence checked: 2026-07-16
Clavinci exists to make AI engineering measurable without making engineers observable as people.