What is built,
what is held,
and what it waits on.
The delivery corpus decides gates; this page tracks work. Phase 0 has run and has not exited — eleven of fourteen spikes are stage-closed on paper because they close on a running system and none exists. Exactly one limb of one exit criterion is open enough to write code against. That limb is section 01.
Implementation — unblocked now
—
This is the whole of the authorised build. It discharges the third limb of
Phase 0 exit criterion 3: a command written against absent code is the
same TODO in a new form, so the criterion closes at the first commit that lands a
scaffold those commands run against. It permitted nothing else — every real-world gate in
section 05 stays shut.
A second wave followed it. Four Phase 0 spikes — SPK-04, SPK-05, SPK-06 and
SPK-07 — now run as working prototypes under /spikes, because the spike
register's rule is that no spike closes without a running system and the repository
previously had no way to satisfy it. None of the four closed, and each says so in its own
evidence document. Their results are in section 03 and the defects they found are in
docs/plan/implementation-findings.md. Three spikes could not be attempted at
all: SPK-01 needs a probe fleet at other cloud providers, SPK-02 needs S3 Object Lock,
SPK-03 needs induced failure against real storage. That is the largest single constraint on
exit criterion 1, and it is procurement rather than engineering.
Five rows read Blocked and are not idle. A3 to A7 are built, running and verified — 4,013 matrix assertions enumerated across the four suites, and 159 unit tests over the verdict rules themselves, all green. They are blocked at closure rather than at start, because this register's own rule is that a task is done when its verification passes against real code, and a suite that passes because there is nothing to test is a failure. Each waits on its subject: the operator surface and audit store at phase 8, the credit ledger at phase 6, invoice records at phase 7, and a probe fleet that by D-08 §6.1 must sit outside this repository altogether. The status vocabulary has no term for complete as far as this phase can take it, so this paragraph is the term.
:shadow lane is itself a failed Release. test:nfr-gov-001
is the specific case: a Release with no operator surface must assert the absence, and a skip
is recorded as a failure.
Schedule
D-07 estimate · week 0 = 19 Aug 2026
Every figure here is D-07's own estimate, not a specification figure and not a
commitment. §37's week figures are disclaimed at source. D-07 is itself drafted, not
delivered — converting effort to calendar needs a team size and a stack, and §38 staffs
nothing. The dates below anchor D-07's week offsets to a week 0 of 19 August 2026 and
carry both correlated ends: optimistic at T = 5.6 scope points per week, pessimistic at
T = 4.5. T is borrowed from §37, not measured.
| Milestone | Week | Optimistic | Pessimistic | What it releases |
|---|
Phase 0 — run, not exited
14 spikes · 8 deliverablesAll four exit criteria assessed as at 19 August 2026. No spike has met all of its closing conditions. Eleven close on a running system; three closed on simulated external evidence, which briefs the real engagement rather than substituting for it. A closure record is not a discharge of the obligation it records.
| Exit criterion | State | What is missing |
|---|---|---|
| 1 — Spikes closed on evidence, D-01…D-08 delivered | Not met | 11 stage-closed on paper, 3 on simulated evidence. D-02 held, D-03 open, D-04 not adopted, D-07 drafted, D-08 provisional |
| 2 — Standing verification per binding NFR | Not met | All four suites now written, running and correctly RED — 4013 assertions enumerated, 159 harness unit tests green. Not met: a suite is standing verification only once it has a subject to verify |
| 3 — D-08 records the stack; CLAUDE.md gains the commands | Partial | Limb 3 discharged 19 Aug 2026 — the commands run against a real scaffold and pass or fail. Decision still provisional; orchestration still undecided |
| Blocking rule — "an open input blocks the item" | Replaced 20 Aug 2026 | Six registers carried a sentence that, read literally, blocked every item in every phase permanently — including the work that produces the evidence the spikes wait for. Replaced by a Release-gate/build-gate split. The money, contract and Release gates are unchanged |
| 4 — §39.1 re-checked, no epic orphaned | Not met | D-07 records four epics with no acceptance criterion to verify |
The fourteen spikes
| Spike | Subject | Disposition | Evidence | Unmet |
|---|
The eight deliverables
| Deliverable | Subject | Disposition |
|---|
The forward track
29 epics · 254 Must/MVP requirements
The numbering is not a sequence. Phases 0–4 are the spine, 5 is ongoing, and
6–9 interleave with 1–4 rather than succeeding 5. Three of the four hold Must/MVP
scope the phase 4 pilot cannot clear without. Sequencing authority is
delivery-interlocks.md; the registers cite it rather than re-derive it.
| Phase | Scope | Epics | SI | D-07 | §37 | Δ |
|---|
Held gates
—None of these moves on internal work. Each closes on a named external party or a named deliverable, and each forbids a specific commercial act until it does.
Update protocol
How this file stays true
This page renders from a single TRACK_A array at the top of its own source. Marking a
task complete is a one-token status edit plus a done date; the counters, the meter and
the filters all re-derive. It is updated at the completion of each task, and
redeployed in the same step, so the deployed page and the repository never disagree.
done when its verification runs and
passes against real code — not when it is written. The three binding-NFR suites are the case that
matters: a suite that passes because there is nothing to test is a failure, and must be recorded as
one. Where a task is closed ahead of its evidence, the row says so rather than reading as clean.