Skip to content

Programme plan

Written by Claude on 2026-10-10 at the maintainer's request from the spec's plan (section 4), the GitHub milestones and the repository history.

Plan and actuals

Milestone Planned due Exit test (spec) Status on 2026-10-11
M0 18 Oct A PX4 SITL vehicle flies a survey via MAVSDK from a script Engineering done 8 to 9 Oct. Product docs v1 edited by the maintainer; issue 1 can close
M1a 25 Oct Nominal and jamming recordings committed Done 9 Oct, as JSON-lines recordings with ground truth, not ROS 2 bags
M1b 1 Nov NT-01 and NT-02 pass on recordings Done 9 Oct
M2a 8 Nov A three-vehicle survey completes nominally Done 9 Oct, with backend and audit log
M2b 15 Nov Approve and deny visible in the audit log; C2-01 passes Done 9 to 10 Oct
M3 22 Nov Results table generated by one command and checked in CI Done 10 Oct, plus a held-out validation
M4a 29 Nov Usability session, short findings note Round 1 done 11 Oct (2 participants, found a blocker); protocol v2 and a console redesign made; round 2 (2 to 3 participants) planned
M4b 6 Dec Demo video, README polish, write-up, v1.0 tag Repository public, docs site live, README summary, demo video recorded and release notes written; v1.0 tag next

The re-plan, honestly

Nothing slipped. The engineering milestones were finished far ahead of their dates, in about three days of repository history (the maintainer planned over about three weeks beforehand). The remaining schedule risk is people and writing time, not engineering: the usability sessions, the product documents, the video and the write-up.

Scope changes made on the way, with reasons: - No ROS 2. The spec allows falling back to MAVSDK only. Recordings are JSON lines. - PX4 pinned to an unreleased main commit, because GPS failure injection for Gazebo does not exist in v1.17.0 or v1.18.0-rc1. - LAND for a frozen position, after ground truth showed the vehicle flies away (ADR 0003). - A calibrated position bound, after NT-03 turned out not to be met (PR 37). - Added: a held-out validation, ground truth, a requirements trace with a test. - Not done: link_loss scenario, a noise or offset injector for gradual drift, reuse of the approval pattern from personal-ai, mission-completion measurement.

Next

Maintainer: product docs, risk log and weekly status, usability sessions, "How this was built". Agent: the video support and README after the sessions. Sector-completion tracking and the fleet-level mission-completion measurement are done (eval/fleet/results.md).