Lanes and windows, in plain terms¶
DESTINED FOR: Tim, and any cold-started window. Kind: explanation; not an instruction. Written 2026-09-03 by the advisory seat. Its repo home is beside the Client Spine (docs/docs/ops/) at the next canon push; until then it lives here.
What a lane is¶
A lane is a series of briefs about one part of the data, with a letter as its filing prefix. It is not a person, a room or a permanent structure. A lane's artefacts live under outputs/lane-<x>/ and scripts/lane-<x>-* forever, because they are the record. The window that does a lane's work is temporary: a Claude Code terminal Tim opens, pastes a brief into, and closes when the lane's current brief is finished.
Lanes exist for one reason: extra hands to make up lost time. The main window was the only place work happened until 3 September. Read-only or write-bounded arcs that never touch main's files can run beside it, so Tim opened extra windows for them. When the parsing and the database writes are done, the lanes end and everything folds back into one main window.
The lanes, as at 3 September¶
| letter | what it is about | what it has done | window state |
|---|---|---|---|
main (-01, -02) |
the NRT content — units of competency and their sections: elements, performance criteria, evidence, modification history, prerequisites, descriptions; the store tga-nrt-content |
-01 closed (four families); -02 at Gate 4 — the first write to the store of record landed today, 2.5 M rows | open, in its own desk rtopacks-project-main |
| F | the companion volumes — the PDF guides that accompany each training package; an index of their headings against TPOF's vocabulary | F1–F2 closed; F3 waits for the 14 September fetch | closed; reopens for F3 |
| Q | qualifications — packaging rules (the unit list of a qualification), and TGA's derived unit grid as the cross-check | Q1 (shape census) and Q2 (the labelled grid walk, 12,902 releases) closed; Q3 (the packaging parser) is next | open, in its own desk rtopacks-project-lane-q, waiting for Q3 |
| Z | the RTO side — registered training organisations, their history, and scope (which RTO delivers what); the store tga-rto |
Z1 (shape census) closed; Z2 (the column-lift into a proper store) at Gate 3 | open, in the shared tree; moves to its own desk at Z2's close |
Not lanes: Lane A / the content spine was the name for main's earlier work; -02b is main's list of remaining section families; Q4 (skill-set re-key) and Z3 (the history timeline) are future briefs in their lanes.
Desks (worktrees), and why they appeared today¶
All windows used to work in one folder on one copy of the repository. Git has one "staging area" per folder, so when four windows staged files at once, one window could sweep another's half-finished files into its commit. That nearly happened three times today and was caught once as a real breach. The cure is a desk per window: a sibling folder (rtopacks-project-lane-q, rtopacks-project-main, …) holding the same repository on a branch named for the window (window/lane-q). Each window commits on its own branch and, when a brief closes, hands the work back to main and pushes. The rule is RULE-WINDOW-BRANCHES-01-INTERIM-2026-09-03.md.
Nothing moved out of the project. There is one repository, one main, one GitHub. The desks are copies; when a lane ends, its desk is removed with one command and the project looks as it always did. The record was never anywhere but main.
What Tim checks, if he wants to¶
In the project folder: git worktree list shows the desks that exist. git log origin/main --oneline | head shows the work that has landed. The second is the only thing a gate ever reads.
What Tim does¶
Paste the Time Machine at the start of a session. Paste each brief into the window it names in its first line. Paste each window's report back. That is the whole of the management; everything else is a file.
— Advisory seat, 2026-09-03.