Rewritten 2026-09-11, immediately after v4.0.0 shipped. Entry point for everything here.
The v2-era planning corpus reached 62 files and ~36,700 lines, written almost entirely in one
month and largely wrong within two releases — not through carelessness, but because it described a
server that was then built differently. ../v3/RE-MEASURE.md keeps the score: nine consecutive
releases opened by measuring the plan against the running server, and every one found the plan
wrong. v4.0.0 made it ten, in the sharpest way yet — the plan said do not cut this release, and the
measurement said the plan had left out the only trigger that could ever have fired.
So this directory is shaped against that failure:
| One measured plan | v4.x-plan.md — evidence first, and its central finding reorders the charters |
| Nine thin charters | charters/v4.1.0.md … v4.9.0.md — intent, candidate scope, measurable trigger, kill criteria |
| Phase checklists | to-dos/ — execution structure, filled in when a charter is actually taken |
Every charter carries a trigger (the observation that would make it real) and kill criteria (the observation that would make it dead). A charter with neither is a wish, and wishes are what produced the 36,700 lines.
Two things, and both are structural rather than cosmetic:
- The numbers are real now. Under the old regime a "v4.2.0 charter" shipped as a v3 minor,
because
4.2.0could not precede a4.0.0the plan said not to cut. v4.0.0 exists, sov4.1.0is simply the next release. Every ordering banner incharters/is retired, and the two charters whose content already shipped were rewritten rather than renumbered — see below. - A major no longer needs the specification's permission. Triggers A–D mean the protocol forced one; T-13 means this project has surface to retire. v4.0.0 was cut on T-13. The v5.0.0 question at the end of this line is therefore answerable without waiting on anyone.
- No charter may be executed without the
../v3/RE-MEASURE.mdpass first. That ritual is the project's, not v3's, and it stays where it is rather than being copied here, because two copies of a ritual diverge. - A charter that is measured and found dead gets archived with a dated banner naming what replaced it — never deleted. The reasoning is the useful part, and a reader who finds no trace of a plan assumes it was never considered.
- Order is not commitment. v3 executed its charters out of order twice and was right to. What is committed here is one dependency: the tool programme does not start before the surface can carry it. That is measured, not preference — see the plan's section 1.
- A release is cut when work is done, not when a number comes up. v2.x ran to eleven minors and v3.x to eleven; both were better for it.
Grouped by what blocks what, not by number:
| Charters | Depends on | |
|---|---|---|
| Unblock | v4.1.0 surface navigationv4.2.0 dispatch consolidation |
nothing — measured, local, ready |
| Build | v4.3.0–v4.5.0 the tool programme, in slices |
unblocked — v4.1.0 landed |
| Watch | v4.6.0–v4.9.0 spec-dependent, and the v5 question |
triggers that have not fired |
The Watch group is deliberately the back half. Its triggers were re-measured on 2026-09-11 and none had moved. Those charters stay thin on purpose, and any that never fires gets archived rather than executed to fill a number.
v4.x-plan.md the measured plan for the line, and the finding that reorders it
v4.0.0-plan.md HISTORICAL -- why v4.0.0 was not scheduled, and what that got wrong
charters/
v4.1.0.md registry tools get a navigation path <- unblocks everything below
v4.2.0.md one place to add a tool
v4.3.0.md the tool programme, slice 1
v4.4.0.md the tool programme, slice 2
v4.5.0.md magic, and scoring that is measured rather than asserted
v4.6.0.md arm64, from reporting to gating
v4.7.0.md tool result contracts, if they standardise
v4.8.0.md auth metadata, audit context, and the rest of the spec watch
v4.9.0.md consolidation, and the v5.0.0 question
to-dos/
00-PHASE-0-TRIGGER-WATCH.md the standing watch, including T-13
01-PHASE-1-SURFACE.md v4.1.0-v4.2.0 -- sprint breakdown
02-PHASE-2-TOOLS.md v4.3.0-v4.5.0 -- the tool programme
03-PHASE-3-MEASUREMENT.md v4.6.0
04-PHASE-4-PROTOCOL.md v4.7.0-v4.8.0
05-PHASE-5-CONSOLIDATION.md v4.9.0 and the v5.0.0 question
DECISION-GATES.md every kill criterion in one table
archive/
v4.1.0-entry-point.md shipped as v3.10.0
v4.2.0-ext-proj-int-phase-1.md first slice shipped as v3.11.0; programme moved to v4.3.0
Because nine was asked for originally, and that is worth stating plainly rather than implying the number came from analysis. What does come from analysis is their content and their order: the first two are measured defects with numbers attached, the middle three are a programme that now has a costed prerequisite, and the last four are watches on triggers that have not fired.
If a charter here looks speculative, it has failed its own standard and should be archived rather than executed.