Six research tracks were told they had no programme. Each had one

pick_track.py matched programmes to tracks by slug prefix or 'track N', so rows naming a glob such as qual-* were missed and six tracks drew blank.

A research session in this repository starts with research/_shared/pick_track.py. It draws one of 25 tracks and, since 2026-11-12, prints the open programme that track already has, so the session advances that programme instead of starting something loose.

It found the programme in one of two ways: the row's "Builds on" column says "track 15", or the programme's slug starts with the track's prefix (sensing-selfcal for the sensing track). Several rows in research/PROGRAMMES.md do neither. llm-coding-inference builds on `qual-*`; verified-dse builds on `eda-*`, `chip-*` and `accel-*`; sim2real-inference builds on `mimic-twin*`. A run drawn to track 19 was told "no open programme names track 19 or starts with qual-: opening one is the default", while the qual programme sat in the table one line down. A run record from today reported exactly that.

What changed

The matcher now also reads the backticked project names and globs in "Builds on" and takes the part before the first - or *. If that is the drawn track's prefix, the programme belongs to the track.

Run against the live table (track_programmes(t) for t = 1..25):

tracks with a programme
before13 of 25
after19 of 25

The six newly mapped tracks are chip (3), eda (7) and accel (8), all to verified-dse; osrobot (9), to the aerolog, marine and hydraulics programmes; mimic (11), to sim2real-inference; and qual (19), to llm-coding-inference. Track 14 (codesign) also gains the three programmes that list `codesign-*`. Tracks 1, 2, 4, 5, 6 and 10 still have no programme, and the draw still says so; that is a true answer now.

A test pins the qual, eda, accel and mimic cases on a fixture table, plus one check against the live table.

All writing · More in Research findings · All tools and research