INCIDENT RESOLVED + G10 depth-4 boot gate PASS: all 9 DC0 nodes enlisted in MAAS ...

Region restart recovered Temporal; dhcpd verified by PROCESS on both
controllers (0->1 each), not by service_set. One restart fixed both sites,
confirming the single wedged-Temporal root cause. The identical canary that
had failed then enlisted immediately -- proof it was the real blocker.
All 9 nodes enlisted with shapes exactly matching D-121 Option C
(3x16/64, 2x12/48, 4x8/24): node VMs inside vvr1-dc0 PXE-booted from the
Office1 region across the transit and run the ephemeral kernel = real
nested KVM at depth 4, proven behaviorally. Appendix-A entry queued for
this new incident class.
CURRENT-STATE updated in this commit (GA-R1/C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

D-129 edge profile APPLIED on 26.7: qga channel shipped, plugins installed, agent ANSWERS guest-ping ...

expose_qga_channel (opt-in, default OFF; dc0 true) applied as an in-place
domain update -- disk with the console/API bootstrap never at risk. Real
install (status ok + msg_uuid, not dry-run echoes); service started via the
plugin's own API; gate met to the runbook's bar: guest-ping -> {'return':{}}
and domifaddr --source agent reports both legs. Prior capture of the same
name recorded the false dry-run and is replaced. Security finding queued:
the plugin can disable guest-exec RPCs (remote command execution) -- propose
at the 26.7 hardening review.
CURRENT-STATE updated in this commit (GA-R1/C1, L10 satisfied).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

FIX false-success: opnsense-plugins.sh apply ALWAYS dry-ran (DRY_RUN:+ expands on 0) ...

Found on the first-ever live run (vr1-dc0 edge, 2026-07-20): ${DRY_RUN:+--dry-run}
expands whenever the var is set and NON-EMPTY, and DRY_RUN=0 is non-empty -- so
every apply silently dry-ran while printing 'OK: profile applied'. Nothing was
ever installed by this script. The harness missed it because every test drove
--dry-run, leaving the live path with zero coverage. Fix tests the VALUE; harness
gains two live-path tests via a stub api client (verified to FAIL 2/2 against the
buggy version before restoring). 20/20.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

D-125 egress isolation gate: PASS / CLOSED -- bridge-in proven end to end (2 identical runs) ...

Measurement disproved the missing-rule hypothesis first: operator-run nft
dumps show virbr4 structurally identical to the working virbr11, with all
three masquerade rules present INCLUDING the generic one, already fired --
so the pre-staged net-destroy/net-start was correctly not run. Matrix
re-probe: gateway ping 0, internet ping 0, curl 1.1.1.1 301, curl
archive.ubuntu.com 200, twice. Run-2's one-off ICMP failure recorded
UNEXPLAINED rather than swept. Side observation kept: the WAN segment
cannot reach the Office1 LAN through the ISP NAT (D-122/SEC-010 intent
holding) -- queued as an explicit hardening assertion.
CURRENT-STATE position + G10 row updated in this commit (GA-R1/C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

D-125 egress gate executed (throwaway VM, as ruled): bridge-in PROVEN, upstream NAT egress fails -- gate OPEN pending one sudo measurement ...

Run 1 was a test artifact (probe used DHCP; uplink /24 has no <dhcp> by
design -- DOCFIX queued for the gate wording). Run 2 static: gateway
ping 3/3 across the nested bridge = the D-125 property PASSES; 1.1.1.1
fails for ICMP and TCP alike. Controls pass and net-dumpxml shows the
working/failing NATs are substantively identical, so the suspect is
vcloud's live virbr4 rules, not the design -- double-NAT fallback
deliberately NOT invoked. Probe guest + temp addresses removed.
CURRENT-STATE updated in this commit (GA-R1/C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

Deploy step C COMPLETE: inner apply 28/28 from voffice1, converged zero diff; 9 nodes + 26.7 edge RUNNING at depth-4; three first-real-apply fixes ...

voffice1 = Plane-2 executor (tofu 1.12.4, repo @29cf7bf, dc0 key,
keyscan'd known_hosts). Fixed forward, measured at first real use:
provider keyfile/sshauth URI params; wan-bridge illegal <mtu> in
bridge mode; base image uploads from the EXECUTING host (copied to
voffice1, description corrected); staging-dir literal created on
rack. Captures: inner-{plan,apply,converge}-20260720-stepC*.
Edge serial log at FreeBSD login prompt (26.7 D-112 first datapoint).
CURRENT-STATE position updated in this commit (GA-R1/C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

Deploy step B COMPLETE: rack bootstrapped + enrolled, SEC-010 CLOSED both ends (operator-ruled), OPNsense 26.7 base staged (operator-ruled); ledger-scan fail-open hardening ...

Bootstrap runs 1+2 (stale-lists 404 fixed by proxied apt update):
maas 3.7/stable, rack enrolled, libvirt+nested KVM, SEC-010 loaded
+ persistent on enp1s0; final --check EXIT 0 ALL items
(docs/audit/stepB-check-20260720-final.txt). voffice1-side SEC-010
pin applied per the row's follow-on clause. 26.7 nano prepped on the
rack (measured leaseweb mirror via region proxy); inner-root base
default -> 26.7. ledger-scan: FAIL-OPEN no longer counts as OPEN
(harness 45/45). CURRENT-STATE position/G10/G14 + section-5 updated
in this commit (GA-R1/C1). Open SEC count 8 -> 7.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

Step B transit reach ESTABLISHED: voffice1 .1/30 applied (Kea re-key incident fixed); vvr1-dc0 netplan set-name root cause fixed; zero-diff converged ...

voffice1 transit NIC applied 1/1/1 exact; MAC-regen broke the Kea
reservation (measured .108) -- re-keyed via edge API, .20 restored.
vvr1-dc0 both-NICs-silent root cause: netplan match.name+set-name
(one-boot rename lag, measured offline via libguestfs); fix drops
set-name, keeps measured enp1s0/enp2s0, instance-id bump, seed
-replace 2/0/2 exact. Verified: ssh -J voffice1 w/ dc0 key,
rack->region 0% loss. Outer plan ZERO DIFF converged.
CURRENT-STATE sections 3+5 updated in this commit (GA-R1/C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

Step B prereq: voffice1 transit NIC authored (172.31.0.1/30 region end, D-124); ruling recorded; expected plan 1/1/1 ...

Measured: transit net isolated, only vvr1-dc0 attached; nothing held .1.
Operator-ruled mechanism 2026-07-19: voffice1 via IaC + restart.
Capture: docs/audit/outer-plan-20260719-voffice1-transit.txt (1/1/1).
CURRENT-STATE sections 3+5 updated in this commit (GA-R1/C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KiUu1oqt76tWvV4vEC3NAr

DC0 deploy STEP A EXECUTED: outer apply 6/0/6 exact; G8+G9 CLOSED; convergence zero ...

Logged window dc0-deploy (as-executed log + index convention). G8 passed
in the apply session (six planes 0 leases, 0 attachments); validate
PASS; pre-apply capture equaled the recorded triple; the SAVED plan was
applied: 6 added / 0 changed / 6 destroyed. vvr1-dc0 RUNNING (108 vCPU,
416 GiB, autostart disable per D-127); vr1-dc0-uplink active; six vcloud
planes destroyed (Model B relocation); voffice1 + edge untouched (ids
1/2). Convergence re-plan: no differences (captured). CURRENT-STATE:
section 1 position, section 5 expected plan = ZERO DIFF (any diff is a
STOP), G8 + G9 rows CLOSED; active gate G10. Captures committed same
commit (C1/C2).

Revert: infra revert = runbooks/dc-dc-teardown-rollback.md decision tree
(NOT a blind destroy); docs revert = git revert this commit.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

AUDIT EXIT: operator re-signed CURRENT-STATE; G3+G11 CLOSED, G9 OPEN; session close (GA-R4) ...

Batch 6 item 6: signature utterance "Reviewed, approved, continue."
recorded into section 11 (replaces 2026-07-18 per GA-R1 rule 7). All six
exit runs PASS. G3 CLOSED (all six batches); G9 OPEN (audit criteria
MET; remaining closer = the operator-gated logged apply with G8 first);
FREEZE LIFTED. Session GA-R4 close: bounded 14-line summary in the
ledger, full body rotated to docs/archive/session-20260719-phase5-sweep.md.
Repo HEAD at signing = this commit's parent (437d9d1).

Revert: git revert this commit.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

Batch 6 item 3: exit probe trio 21/21 -- 7/7 on ALL THREE (bar exceeded) ...

Three independent clean-context probes (repo + verbatim charter question
set only, read-only, no shared state; fresh contexts satisfy GA-R7/G2)
scored against CURRENT-STATE with citations spot-verified against the
tree. Unanimous on every question; third-trap picks varied (all real);
zero orientation divergence (all landed CURRENT-STATE-first via the
swept skill); probe hole H1 did not recur. Full scorecard in the exit
record. Remaining before audit exit: item 6 only (operator re-read +
re-signature). G3 cell updated same commit.

Revert: git revert this commit.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

Batch 6: exit runs 1/2/4/5 -- archive exclusion (operator-ruled), adjudication, plan==triple, S2 ...

6.1: docs/archive/ joins docs/audit/ as inventoried-not-counted (ruling
quoted in the exit record); record-audit + harness 35/35; per-group
adjudication = ZERO contradictory status-claim pairs on counted surfaces
(the charter's criterion as stated); group floor 9 recorded as re-run
baseline; missed patch file archived; CURRENT-STATE stale-note made
token-free. 6.2: 16 top-level docs (<25). 6.4: fresh capture
outer-plan-20260719-phase6-exit.txt = 6/0/6 == section-5 triple. 6.5:
G9 row names SEC-010's step-B dependency (S2). Exit record:
docs/audit/phase6-exit-runs-20260719.md. Items 3+6 pend operator
execution. G3 cell updated same commit.

Revert: git revert this commit.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

Batch 5: skill sweep -- GA-R1..R7 implemented in openstack-cloud-ops (single commit) ...

New GA-rulings section reconciled against the RATIFIED text (draft
skill-additions cross-checked; the one draft-only item, the
reconciliation circuit-breaker, kept and marked as discipline). Stale
facts demoted to CURRENT-STATE delegation: the header's "current phase =
VR0 testcloud" and the deploy loop's "opentofu UNVALIDATED / no tofu
binary has ever run" (false since 2026-07-13). Session bookends + stage
close rewritten per GA-R4/GA-R6; change-delivery per GA-R2/D1; routing
rows re-pointed. Read-through checklist committed:
docs/audit/skill-sweep-checklist-20260719.md (every ratified rule ->
disposition). G3 cell updated same commit. repo-lint 0 fail.

Revert: git revert this commit (restores the pre-sweep skill).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

Batch 3 CLOSE: memory hygiene (GA-F05..F08 memory surfaces) + G1 review; GA-F16 registered ...

Per the recorded provenance rulings: review-after-each-step.md REMOVED
(GA-F06 -- one-time instruction, never standing) and project-posture-
closed-test.md REMOVED (GA-F07 -- "ruling" never ruled; posture NEUTRAL);
hostname -> vcloud in the Remote Control memory (GA-F05 last surface);
MEMORY.md index rebuilt (GA-F08). GA-R7 rule-6 skill line added to
SKILL.md bootstrap. First full G1 review (rules 1-4) found + fixed one
further violation: creds-folder-convention.md duplicated a repo-carried
convention and linked both removed files -- registered as GA-F16, reduced
to a pointer entry. Post-review: 3 memory files, zero violations, zero
stale refs (grep-verified). Pre-edit copies: docs/archive/memory-20260718/
(index copy ASCII-normalized, noted in the changelog). G3 cell updated
same commit (L10 satisfied).

Revert: restore files from docs/archive/memory-20260718/ (memory side);
git revert this commit (repo side).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

GA-F10/H1 + GA-R1/R6: token migration, State-line demotions, E1 type tags (2.5+2.6) ...

Mapping record: docs/audit/h1-token-map-20260719.md -- all distinct
tokens dispositioned, none unmapped. Live-surface migration: workflow
doc (26 flags -> 0; header drops the tracker model, 8 State lines demote
to identity + CURRENT-STATE pointers), runbooks (18 -> 0; close-out
instructions RETARGETED to record status in CURRENT-STATE, not the
workflow tracker), readiness doc (5 -> 0), CURRENT-STATE itself (6 -> 0:
G9 STOPPED -> BLOCKED with named blocker; stale "GA-R1 NOT yet ratified"
header + pre-D-130 divergence rows corrected per C2). Gate rows gain
GA-R6/E1 type tags ([V]/[R]) with legend. CLAUDE.md expectation line
demoted per S5b. History classes (ledger, changelogs, design-decisions)
deliberately untouched -- their disposition is Batch 4/GA-R3 structural.
Extractor re-run: zero flagged occurrences on live non-history surfaces.
Batch 2 items 2.5 + 2.6.

Revert: git revert this commit.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

G6 reconcile EXECUTED (state surgery, operator-ruled) + G7 triple 6/0/6 recorded ...

Batch 1 items 5-7. Operator ruled (ii) state surgery per S3/GA-R5:
state pull -> autostart:true injected on voffice1 + office1-opnsense
(both were None, matching the measured absence) -> push, serial 22->23,
backup opentofu/terraform.tfstate.pre-G6-20260719 (gitignored). Guests
never touched: ids 1/2 unchanged, both running, autostart enable.
Re-plan captured to docs/audit/outer-plan-20260719-postG6.txt:
Plan: 6 to add, 0 to change, 6 to destroy -- zero in-place changes,
zero replacements. Section 5 records 6/0/6 as THE expected triple with
decomposition (C2: capture + doc same commit); G6 + G7 rows CLOSED;
divergence 2.3-ii marked resolved.

Revert: git revert this commit for the docs; state revert = tofu state
push of the pre-G6 backup (restores the 2-change plan).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

D-130 ADOPTED (a) + implemented: seed-volume ignore_changes guard (GA-F01, G4+G5 CLOSED) ...

Batch 1 items 1-4. G4 verifications captured: v8 = suppression CONFIRMED
(fresh baseline 7/2/7 identical to 2026-07-18, with-edit 6/2/6, zero
forces-replacement; DC0 creates unaffected); v7 = defect reproduced on
isolated throwaway-v7-* domain, replacement applied UNDER the running
domain with NO bounce (domid 3->3), teardown verified zero residue.
Operator ruled (a) per GA-R5 (utterance quoted in the D-130 entry).
Implemented: lifecycle { ignore_changes = [create] } on
libvirt_volume.seed in modules/cloudinit-vm + tests/cloudinit-vm
harness (6/6; guard placement, exact attribute, no leak onto data disk,
tofu validate). CURRENT-STATE: G4/G5 rows CLOSED + rotated, decision
queue item 1 rotated to the register, section 5 records post-fix 6/2/6
pending the G6 reconcile (C2: captures + doc same commit). 342/360 cap.

Revert: git revert this commit (restores the replace-pair behavior and
the pre-ruling CURRENT-STATE rows; captures remain in history).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VZbUeRE8weySLSGY1odrCi

Sweep plan AMENDED per external review (S1-S5, single commit); G3 row updated ...

S1 = item 2.10 R3-F residual disposition (none unaccounted); S2 = Batch-6
SEC-gate confirmation before G9 opens; S3 = G6 reconcile mechanism is a
GA-R5-format operator ruling; S4 = v7 collision-proof throwaway-v7-*
naming, scratch storage, verified teardown; S5 = batch-boundary-only
session breaks with GA-R4 closes + CLAUDE.md expectation-line demotion in
2.6. Diffable against reviewed draft 687dfa8. G3 evidence cell updated
same-commit per GA-R1 rule 8.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA Phase-5 sweep plan DRAFTED (batches 0-6, GA-F/GA-R traced, dependency-ordered) ...

Six batches + exit runs: tooling first (C1 lint check, F3 repoint, GA-F15
hardening, H1 extractor alignment), the v8->v7->D-130 decision chain with
the operator's contingent lean carried verbatim, record fixes per-GA-F,
memory hygiene, consolidation/rotation, skill sweep, Phase-6 exit
criteria. DRAFT: nothing executes until external review + per-batch
operator gates, in a FRESH session. G3 row updated same-commit (C1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

Vocabulary RATIFIED: Option A + H1 (utterance quoted) -- PHASE 4 COMPLETE; G2 CLOSED ...

Four-state vocabulary (OPEN/BLOCKED/HELD/CLOSED); H1 makes the Phase-5
token migration mechanically complete via the extractor scan and defines
event words (VOID/REOPENED/SUPERSEDED) as row-history, never states. All
seven GA-R rulings + the vocabulary are ratified with paired
question/utterance Status blocks, drafts diffable per B2. G2 gate row
CLOSED same-commit per GA-R1 rule 8 (ruling-type closure per GA-R6 E1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R7 RATIFIED (operator, amendments G1+G2+G3, question+answer quoted); vocabulary A/B staged ...

G3 = memory is scratch, durable observations graduate to repo surfaces;
G1 = stage-close memory review joins the GA-R6 close-out set; G2 =
auto-memory posture (enabled for operator sessions, disabled for probe/
audit sessions incl. the Phase-6 exit trio). All seven GA-R rulings now
RATIFIED. Stage-status vocabulary A/B (GA-F10) staged as the final
Phase-4 item. G2 gate row updated same-commit per GA-R1 rule 8.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R4 RATIFIED (operator, amendments F1+F2+F3, question+answer quoted); GA-R7 draft staged ...

F1 = cap binds at session close, live sections never rotated; F2 =
post-hoc closure of orphaned sessions (no status weight); F3 =
security-ledger.md is the SEC register of record, ledger carries
pointer+count only. GA-R7 (memory governance) delivered as DRAFT. G2 row
updated same-commit per GA-R1 rule 8.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R6 RATIFIED (operator, amendments E1+E2+E3, question+answer quoted); GA-R4 draft staged ...

E1 = gate typing (ruling-type gates close per GA-R5); E2 = harnessed
closers, vacuous-pass closures VOID (DOCFIX-194 class); E3 = no
conditional close, remainders split into their own gate rows. GA-R4
(ledger rotation) delivered as DRAFT. G2 row updated same-commit per
GA-R1 rule 8.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R2 RATIFIED (operator, amendments D1+D2, question+answer quoted); GA-R6 draft staged ...

D1 caps changelog creation at one per session; D2 widens the Phase-5
sweep to every docs/ file with manifested consolidation commits and
defines the under-25 counting scope. GA-R6 delivered as DRAFT (executable
definition of done; 5/0/6 superseded by the CURRENT-STATE section-5
mechanism). G2 gate row updated same-commit per GA-R1 rule 8.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R1 RATIFIED (operator, amendments C1+C2, question+answer quoted); GA-R2 draft staged ...

C2 folded into rule 1 (measurement beats the document); C1 is new rule 8
(status-changing commits must touch CURRENT-STATE.md -- applied to this
very commit: G2 gate row updated in-line). GA-R2 delivered as DRAFT per
B2. Phase-5 repo-lint check for C1 queued.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R5 RATIFIED (operator, amendments B1+B2, question+answer quoted); GA-R1 draft staged ...

GA-R5 Status block records both the question as presented and the exact
ratification utterance per B1's own terms; consolidated text extends
rules 2/5 (B1) and rule 4 (B2, draft-diffability -- this commit itself
satisfies it for GA-R1). GA-R1 delivered as DRAFT with the size budget,
rotation, and as-built division directives baked in.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA-R3 RATIFIED (operator, amendments A1+A2, utterance quoted); GA-R5 draft staged ...

Phase-4 rulings file opened. GA-R3 recorded with the exact operator
utterance in the Status block and the consolidated as-amended text;
implementation queued for Phase 5 (class tags, D-121..129 review, A1
D-129 split). GA-R5 delivered as DRAFT, awaiting ruling.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA register: standing proof-of-push rule (ls-remote hash on every completion report) ...

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21

GA register: record the Phase-4 hold/order/format directive (was in-session only) ...

Operator guard applied: an unrecorded directive is not a directive. The
Phase-4 delivery order (GA-R3, R5, R1, R2, R6, R4, R7, vocab A/B), the
one-per-exchange format, and the hold-until-G11 rule are now in the
register. No GA-R text exists yet; Phase 4 remains HELD pending G11.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018ZV3uqyU9oWoFTcwiE4a21