diff --git a/workplans/SECRETS-WP-0009-glas-claude-native-delivery.md b/workplans/SECRETS-WP-0009-glas-claude-native-delivery.md
index 9016b3b..59239e5 100644
--- a/workplans/SECRETS-WP-0009-glas-claude-native-delivery.md
+++ b/workplans/SECRETS-WP-0009-glas-claude-native-delivery.md
@@ -346,3 +346,102 @@ Intermediate `95c9edb1` is installed but refused: a sandboxer bwrap info-fd
 partial-read race, fixed in `221126c`. Receipt: prj-helixforge-factory
 `evidence/2026-09-22-runtime-placement.json`. The exec_owner argv will use
 `2cfc7b5f`. Next: B3 fixture repo, repository grant, ActivityDefinition, worker id.
+
+### 2026-09-23 B3 scope correction and model choice
+
+Operator chose `claude-sonnet-5` for MessagesOwner and a throwaway fixture repo
+with no remote. Reading the owner contracts corrected the B3 framing above:
+
+- The Glas acceptance run (`glas-harness/docs/local-profile-acceptance.md`,
+  `scripts/prove-local-profile.py`) creates its own disposable fixture repo
+  (one `PROOF.md`, one local commit, no push remote). No repo is pre-created.
+- That path runs with `--no-metrics` and carries no repository grant, so it
+  needs no grant, ActivityDefinition or queue worker id. Those belong to the
+  Activity Core queue loop (HFACT-WP-0001-T05), not to this first run.
+- The candidate `harness.agent-dev-local@1.1.0` pins `claude-sonnet-4-6`.
+  Moving to `claude-sonnet-5` is a glas-harness profile change, reviewed as a
+  new compatibility set, and must be checked against the pinned Claude CLI
+  2.1.263 in runtime `2cfc7b5f` before the attended run.
+
+No repo, grant, definition, credential or approval was created.
+
+### 2026-09-23 Sonnet 5 compatibility check
+
+Correction: runtime `2cfc7b5f` ships Claude CLI **2.1.266** (binary
+`19842705e989…`, per `prj-helixforge-factory/evidence/2026-09-22-runtime-placement.json`),
+not 2.1.263 (that was the earlier `5cf9a16c` candidate). The same binary in the
+local `5371156d` store matches that digest. Its built-in model registry lists
+`claude-sonnet-5` as a first-class entry (`{"name": "claude-sonnet-5",
+"supports1m": true}`), and it is the CLI's own example model for hooks. No
+deprecation marker references it. No runtime rebuild is needed. This was a static
+check of the binary, not a provider request. The first real call is still
+the attended exec. Remaining model work: a glas-harness profile change from
+`claude-sonnet-4-6` to `claude-sonnet-5`, reviewed as a new compatibility set.
+
+### 2026-09-23 Sonnet 5 runtime placed
+
+Operator chose a 1.1.1 bump. glas-harness `30be818` adds
+`harness.agent-dev-local@1.1.1` (`claude-sonnet-5`) and blocks 1.1.0. rein-aharness
+`fdc07ee` pins it in the runtime contract lock (llm-connect `37436bb`, workplan-only).
+Owner runtime `b6e4e8a429393d68831c996a65c0489664205df969ac9882e581983ec2da4969`
+(Claude CLI 2.1.266) is installed on railiance01 beside `2cfc7b5f` and `5371156d`
+(rollback). Synthetic metered proof passes locally and on railiance01 (one fake
+provider request, zero on refusal, artifact unchanged). Receipt:
+`prj-helixforge-factory/evidence/2026-09-23-runtime-placement.json`. The
+`exec_owner` argv and the owner config `runtime` block will use `b6e4e8a4`, and
+`messages_policy.model` will be `claude-sonnet-5`. Next: private SpendPolicy, ledger
+`spend init` and owner config on railiance01 (HFACT-WP-0001-T01/T04).
+No credential, approval, profile activation or worker change.
+
+### 2026-09-23 correction: the paid run is a queue claim
+
+The earlier B3 correction was wrong for this task. It read the Glas acceptance
+runner, which makes its own fixture. The T03 exec path is different:
+`rein-aharness metered-once` runs one Activity Core claim cycle
+(`docs/owner-bootstrap.md`), and the SpendPolicy binds `worker_id`,
+`activity_definition_id`, `target_repo`, `project` and `repository_grant_id`
+(`docs/spend-admission.md`). The policy is immutable once `spend init` runs, so
+all of these must be fixed first:
+
+- target repo: a throwaway git repo on railiance01, no remote (operator-approved);
+- repository grant v1: `PROOF.md` only, exactly one commit, `publish: false`;
+- a production Activity Core definition carrying that grant and
+  `harness.agent-dev-local@1.1.1`, left disabled until the attended run;
+- a worker identity and worker token for the metered owner. The existing
+  `rein-aharness-claim-loop.service` on railiance01 claims by label, so the run
+  needs a label it does not claim, or it could take the paid run.
+
+No SpendPolicy, ledger, owner config, repo or definition was created.
+
+### 2026-09-23 option 1 preparation (partial)
+
+Operator chose: request multi-worker identities from activity-core (message
+`914853d9`), and run now under `rein-aharness@railiance01` isolated by label.
+
+- activity-core `6002d5c`: disabled definition `hfact-glas-metered-proof`
+  (id `5bae5505-77f1-5ba3-8dfa-2e10ed15531e`), label `hfact-metered`, profile
+  1.1.1, grant `PROOF.md` / one commit / no publish. The parser accepts it; the
+  live ConfigMap diff is additive only. **Not applied:** the harness classifier
+  blocked the ConfigMap apply and sync Job on railiance01.
+- Fixture repo on railiance01: `/home/tegwick/hfact/targets/hfact-glas-proof`
+  (mode 0700, no remote, base `679d23e0`).
+- SpendPolicy inputs: profile_sha256 `10e423fe…83ca65`, descriptor_sha256
+  `4f151744…3a9b`, repository_grant_id `656911f7dff83e871434b415f0cee685`,
+  max_turns 30. No `project` value is on record yet.
+
+No SpendPolicy, ledger or owner config was written. No credential was read.
+
+### 2026-09-23 definition live; owner provisioning prepared
+
+The operator ran the ConfigMap apply and sync Job. The sync log shows
+`hfact-glas-metered-proof` upserted (29 definitions). Operator chose project
+`prj-helixforge-factory` and list-price bounds for `claude-sonnet-5`
+(claude-api reference, cached 2026-06-24: USD 2 / 10 per MTok). The owner
+MessagesPolicy uses an input bound of 4 micro-USD per token (covers the 1h cache
+write), output 10, context 200000, output 32000, and the betas the CLI passed in
+the synthetic proof. The worst-case hold per request is USD 1.12.
+
+A provisioning script (SpendPolicy, `spend init`, request tables, mode-0600
+owner config under `~/hfact/owner`) validates in memory. Staging it on
+railiance01 was blocked by the harness classifier as a real-world transaction.
+It has not been run. No policy, ledger or owner config exists yet.
diff --git a/workplans/SECRETS-WP-0011-multi-lane-exec-owner-delivery.md b/workplans/SECRETS-WP-0011-multi-lane-exec-owner-delivery.md
new file mode 100644
index 0000000..c32b4f0
--- /dev/null
+++ b/workplans/SECRETS-WP-0011-multi-lane-exec-owner-delivery.md
@@ -0,0 +1,100 @@
+---
+id: SECRETS-WP-0011
+type: workplan
+title: "Multi-lane exec-owner delivery"
+domain: infotech
+repo: secrets-engine
+status: ready
+flavor: implementation
+owner: claude-code
+topic_slug: netkingdom
+created: "2026-09-23"
+updated: "2026-09-23"
+related_workplans:
+  - SECRETS-WP-0009
+  - HFACT-WP-0001
+---
+
+Demand: SECRETS-WP-0009-T03. A configured `exec_owner` child today receives
+only the catalog's fixed non-secret environment plus one delivered field
+(`exec_delivery.py:233`). The Glas metered owner (`rein-aharness metered-once`)
+also needs the Activity Core worker token to claim its queue row. Operator
+decision 2026-09-23: generalize secrets-engine delivery rather than add a
+rein-side token-file reader.
+
+Operator decisions 2026-09-23: one decision chain per lane (no flex-auth
+package change), and the worker-token lane uses ordinary lane approval, not
+human control, since it only lets a worker claim jobs.
+
+Scope: one exec may deliver fields from more than one cataloged lane into one
+catalog-bound owner. The primary lane stays the one named by `--catalog`;
+the others are companions declared on the primary's `exec_owner`.
+
+Invariants carried over unchanged:
+
+- Every delivered value comes from a cataloged lane through that lane's own
+  AppRole, in memory only, with self-revoking sessions and redaction.
+- Each lane's read is its own protected action. No lane's approval, decision
+  or consume covers another lane.
+- The binding stays exact. Companion lanes, fields and env names are part of
+  the `exec_owner` digest, and the command is refused if any differ.
+- Fail closed as a set. If any lane is unapproved, missing, denied or fails,
+  no child is started and no value is kept.
+
+## Define the companion contract
+
+```task
+id: SECRETS-WP-0011-T01
+status: todo
+priority: high
+```
+
+Extend `exec_owner` with `companions: [{catalog, field, env}]`. Validate at
+catalog load: each companion is an existing `kv` lane with the same stage, a
+declared field, an env name that doesn't collide with the fixed environment or
+another delivery, and `exec-env` in its delivery modes. Companions are
+refused on pending owners. Document the contract in `docs/` and the SCOPE
+exec-time delivery section.
+
+## Gate and fetch each lane separately
+
+```task
+id: SECRETS-WP-0011-T02
+status: todo
+priority: high
+```
+
+Run the existing approval-claim / CheckRequest / consume chain once per lane,
+with each lane's own action and resource, before any value is fetched. Then
+fetch every value, re-check the binding digest, and spawn once with all values
+injected and redacted. Evidence records each lane's decision reference and
+session cleanup. Any failure starts no child and reports which lane refused.
+
+## Tests and throwaway OpenBao proof
+
+```task
+id: SECRETS-WP-0011-T03
+status: todo
+priority: high
+```
+
+Unit tests: validation, digest coverage, collision, stage mismatch, partial
+approval, one lane denied after another is fetched, redaction of every value,
+cleanup of every session. Integration: two KV lanes delivered to one owner on
+a throwaway OpenBao dev server.
+
+## Catalog the Activity Core worker token lane
+
+```task
+id: SECRETS-WP-0011-T04
+status: wait
+priority: high
+blocking_reason: "Needs activity-core owner assent to source ACTIVITY_CORE_WORKER_TOKEN from OpenBao (ESO) instead of a hand-generated Kubernetes Secret."
+```
+
+Request activity-core to move the worker token into OpenBao custody, synced to
+`actcore-runtime-secret` by their existing `openbao-activity-core` store. Then
+catalog a read lane for the metered owner and add it as the Glas lane's
+companion. This coordinates with activity-core's multi-worker request (hub
+message `914853d9`): a dedicated metered worker token would use the same lane
+shape.