Zevmon status

The step funnel follows each piece of work through its steps. Merged PRs: Origin (opened by the SDLC agents, or by a local Claude Code session or a person), Code (every commit by the SDLC bot or a coding agent, or a person committed), Review (an agent review or approval, a person's review, or none) and Merge (the agents, or a person). SDLC stories: Intent, Plan, Design, Build, Review, Merge (the code PRs), UAT and Release. A step counts as human only when a person did its work: a person's commit, review, gate approval or merge. Starting a stage (a hand dispatch or a re-run) is not doing it: the SDLC stage job does the work. A PR's code is its own commits: a merge commit ("Update branch") or a commit another PR already landed is not its work. Before that, the agents worked under the Chairman's token, so the engine's own merges of what an agent approved, and its auto-merges of build and release notes, count as LLM. Intent is the Chairman's input: shown, never scored. The headline is the LLM share of all steps performed in the range. Merged PRs are also shown split by where the work came from: from an issue (an SDLC story records the PR, or its branch, title or body names an issue: zdev-n/…, ZDEV-n, “Closes #n”, an issue link) or no issue (local Claude Code sessions, hand PRs, role and bot PRs); a card picks which ones the funnel draws. Stories are those with activity in the range; PRs those merged in it.

LLM-authored means every inspected commit was authored by the SDLC bot. Verified autonomous requires evidence for all six delivery stages, their artifact and code PRs, approvals of the reviewed commits, and successful release execution. Human approvals, commits, merges and feedback count as human involvement; who started a stage run does not. Missing or truncated records remain unverified.

Human account identifies participation under a GitHub user account. A shared automation token can also appear under that account, so it does not prove a person clicked. Other bots are unverified unless identified as this SDLC pipeline. Scheduled runs are automatic; a dispatch alone does not prove autonomy.

Each row counts distinct PRs at that step in the window; rows overlap and should not be added together. Commit counts deduplicate SHAs within a row. Lines are the current PR diff (added and removed), including tests and generated files; they measure change volume, not net codebase size or productivity. Deploy and release rows use code PRs linked to SDLC stories with a recorded production smoke pass. Merge-only tooling releases appear in Release, not Deploy. Untracked releases are outside this cohort.

The change log of each app, in plain words, one card per app per day. Released means it reached people using the app. Merged, not released means it is done in the code but no store release is tracked yet (iOS, Android). Dev tooling is how the team builds Zevmon; it is live as soon as it merges. Numbers cover the time range picked at the top; “behind the scenes” counts tests, CI and tooling work that users don’t see.

A fictional household uses the app every day on a test copy and reports what breaks.