Split adaptive paper sells within the durable order envelope - #1091

The mover flat-recovery outcome is intended under CC #1091: owned exits are exempt from entry qty/notional caps, oversell stays refused. Pin NativeSellRisk to fixed RTH using the existing FixtureClock; retain real native risk. Test-only correction plus derived hash registries; no production regression or code change.

The mover flat-recovery outcome is intended under CC #1091: owned exits are exempt from entry qty/notional caps, oversell stays refused. Pin NativeSellRisk to fixed RTH using the existing FixtureClock; retain real native risk. Test-only correction plus derived hash registries; no production regression or code change.

The mover flat-recovery outcome is intended under CC #1091: owned exits are exempt from entry qty/notional caps, oversell stays refused. Pin NativeSellRisk to fixed RTH using the existing FixtureClock; retain real native risk. Test-only correction plus derived hash registries; no production regression or code change.

The mover flat-recovery outcome is intended under CC #1091: owned exits are exempt from entry qty/notional caps, oversell stays refused. Pin NativeSellRisk to fixed RTH using the existing FixtureClock; retain real native risk. Test-only correction plus derived hash registries; no production regression or code change.

The mover flat-recovery outcome is intended under CC #1091: owned exits are exempt from entry qty/notional caps, oversell stays refused. Pin NativeSellRisk to fixed RTH using the existing FixtureClock; retain real native risk. Test-only correction plus derived hash registries; no production regression or code change.

The mover flat-recovery outcome is intended under CC #1091: owned exits are exempt from entry qty/notional caps, oversell stays refused. Pin NativeSellRisk to fixed RTH using the existing FixtureClock; retain real native risk. Test-only correction plus derived hash registries; no production regression or code change.

…-0005) and scheduling-supervision (A5-0024)

From the practice gap ledger of 2026-10-15 (rows A3-0090, B2-0005 and A5-0024, each ruled ADOPT and re-checked
AGREE by the co-op's Codex lane). The plan asks for one PR per layer page; the coordinator combined these two small
pages in one commit. This commit is the branch squashed onto the same base under the digest-custody precedent of
#1091, #1148 and #1150: no commit in the branch keeps a private digest prefix.

observation-inference, slot usage-cost-monitoring:
- A3-0090: the template and the collector example set OTEL_METRICS_INCLUDE_ENTRYPOINT=true (off by default;
  env-vars.md:510, monitoring-usage.md:156 and :584 at client 2.1.300), so metrics carry app.entrypoint to split
  interactive, -p and SDK runs. The row records that the collector's transform/privacy metric allowlists (the
  repository profile, and the 2604 profile the live collector runs) do not keep app.entrypoint yet, so the collector
  sums the label away until they do; the host steps go to the command center as a host diff, named without a path.
  The dispositions catalog records the switch as adopt-pending on the template, and the new-WSL client configuration
  record gains a dated projection (404 pieces, one more wired through the otel-collector-contrib slot).
- B2-0005: signed in with the plan, claude -p, Agent SDK and third-party app runs draw the plan's usage limits
  (support article 15036540, its 2026-06-15 and 2026-10-07 updates); the Max and Team monthly API credits cover a
  linked Console organization's API-key usage only and leave the plan's limits unchanged (support article 17154008).

scheduling-supervision, slot scheduled-tasks:
- A5-0024, a correction: --resume and --continue restore CronCreate tasks that have not expired or passed, and a
  backgrounded session keeps them firing (scheduled-tasks.md:11, :228, :230). Measured on this host on 2026-10-15
  and retained in evidence/artifacts/d4-bg-trial-20261015/scheduled-task-carry.json: a one-shot CronCreate task
  fired at 07:36:00Z inside the copy that claude --resume <id> --bg started. The route now has resume and background
  paths run CronList first and cancel stale tasks with CronDelete (scheduled-tasks.md:159-160).

Digest custody (command center, 2026-10-15; the GPT read at 8c03cbe, P1): repository text cites committed files and
public sources only. test_the_rows_cite_no_scratch_path_or_unpublished_digest checks both rows for the scratch roots
and the private findings files' name and holds no private value; a synthetic negative control plants a scratch
reference with a synthetic digest and is caught. The slot manifest's sha256_when_labeled follows slots.json.

tests/test_native_practice_ledger_obs_sched.py: 6 of 6 tests failed before on the base (38 subtests); 7 pass now;
native-practice modules 106 OK; the mechanism modules pass except four CachedArtifactTests in
test_upstream_surface_watch, which fail the same way on the base; validate rc 0.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

…-0005) and scheduling-supervision (A5-0024)

From the practice gap ledger of 2026-10-15 (rows A3-0090, B2-0005 and A5-0024, each ruled ADOPT and re-checked
AGREE by the co-op's Codex lane). The plan asks for one PR per layer page; the coordinator combined these two small
pages in one commit. This commit is the branch squashed onto the same base under the digest-custody precedent of
#1091, #1148 and #1150: no commit in the branch keeps a private digest prefix.

observation-inference, slot usage-cost-monitoring:
- A3-0090: the template and the collector example set OTEL_METRICS_INCLUDE_ENTRYPOINT=true (off by default;
  env-vars.md:510, monitoring-usage.md:156 and :584 at client 2.1.300), so metrics carry app.entrypoint to split
  interactive, -p and SDK runs. The row records that the collector's transform/privacy metric allowlists (the
  repository profile, and the 2604 profile the live collector runs) do not keep app.entrypoint yet, so the collector
  sums the label away until they do; the host steps go to the command center as a host diff, named without a path.
  The dispositions catalog records the switch as adopt-pending on the template, and the new-WSL client configuration
  record gains a dated projection (404 pieces, one more wired through the otel-collector-contrib slot).
- B2-0005: signed in with the plan, claude -p, Agent SDK and third-party app runs draw the plan's usage limits
  (support article 15036540, its 2026-06-15 and 2026-10-07 updates); the Max and Team monthly API credits cover a
  linked Console organization's API-key usage only and leave the plan's limits unchanged (support article 17154008).

scheduling-supervision, slot scheduled-tasks:
- A5-0024, a correction: --resume and --continue restore CronCreate tasks that have not expired or passed, and a
  backgrounded session keeps them firing (scheduled-tasks.md:11, :228, :230). Measured on this host on 2026-10-15
  and retained in evidence/artifacts/d4-bg-trial-20261015/scheduled-task-carry.json: a one-shot CronCreate task
  fired at 07:36:00Z inside the copy that claude --resume <id> --bg started. The route now has resume and background
  paths run CronList first and cancel stale tasks with CronDelete (scheduled-tasks.md:159-160).

Digest custody (command center, 2026-10-15; the GPT read at 8c03cbe, P1): repository text cites committed files and
public sources only. test_the_rows_cite_no_scratch_path_or_unpublished_digest checks both rows for the scratch roots
and the private findings files' name and holds no private value; a synthetic negative control plants a scratch
reference with a synthetic digest and is caught. The slot manifest's sha256_when_labeled follows slots.json.

tests/test_native_practice_ledger_obs_sched.py: 6 of 6 tests failed before on the base (38 subtests); 7 pass now;
native-practice modules 106 OK; the mechanism modules pass except four CachedArtifactTests in
test_upstream_surface_watch, which fail the same way on the base; validate rc 0.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

…-0005) and scheduling-supervision (A5-0024)

From the practice gap ledger of 2026-10-15 (rows A3-0090, B2-0005 and A5-0024, each ruled ADOPT and re-checked
AGREE by the co-op's Codex lane). The plan asks for one PR per layer page; the coordinator combined these two small
pages in one commit. This commit is the branch squashed onto the same base under the digest-custody precedent of
#1091, #1148 and #1150: no commit in the branch keeps a private digest prefix.

observation-inference, slot usage-cost-monitoring:
- A3-0090: the template and the collector example set OTEL_METRICS_INCLUDE_ENTRYPOINT=true (off by default;
  env-vars.md:510, monitoring-usage.md:156 and :584 at client 2.1.300), so metrics carry app.entrypoint to split
  interactive, -p and SDK runs. The row records that the collector's transform/privacy metric allowlists (the
  repository profile, and the 2604 profile the live collector runs) do not keep app.entrypoint yet, so the collector
  sums the label away until they do; the host steps go to the command center as a host diff, named without a path.
  The dispositions catalog records the switch as adopt-pending on the template, and the new-WSL client configuration
  record gains a dated projection (404 pieces, one more wired through the otel-collector-contrib slot).
- B2-0005: signed in with the plan, claude -p, Agent SDK and third-party app runs draw the plan's usage limits
  (support article 15036540, its 2026-06-15 and 2026-10-07 updates); the Max and Team monthly API credits cover a
  linked Console organization's API-key usage only and leave the plan's limits unchanged (support article 17154008).

scheduling-supervision, slot scheduled-tasks:
- A5-0024, a correction: --resume and --continue restore CronCreate tasks that have not expired or passed, and a
  backgrounded session keeps them firing (scheduled-tasks.md:11, :228, :230). Measured on this host on 2026-10-15
  and retained in evidence/artifacts/d4-bg-trial-20261015/scheduled-task-carry.json: a one-shot CronCreate task
  fired at 07:36:00Z inside the copy that claude --resume <id> --bg started. The route now has resume and background
  paths run CronList first and cancel stale tasks with CronDelete (scheduled-tasks.md:159-160).

Digest custody (command center, 2026-10-15; the GPT read at 8c03cbe, P1): repository text cites committed files and
public sources only. test_the_rows_cite_no_scratch_path_or_unpublished_digest checks both rows for the scratch roots
and the private findings files' name and holds no private value; a synthetic negative control plants a scratch
reference with a synthetic digest and is caught. The slot manifest's sha256_when_labeled follows slots.json.

tests/test_native_practice_ledger_obs_sched.py: 6 of 6 tests failed before on the base (38 subtests); 7 pass now;
native-practice modules 106 OK; the mechanism modules pass except four CachedArtifactTests in
test_upstream_surface_watch, which fail the same way on the base; validate rc 0.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

…-0005) and scheduling-supervision (A5-0024) (#1158)

From the practice gap ledger of 2026-10-15 (rows A3-0090, B2-0005 and A5-0024, each ruled ADOPT and re-checked
AGREE by the co-op's Codex lane). The plan asks for one PR per layer page; the coordinator combined these two small
pages in one commit. This commit is the branch squashed onto the same base under the digest-custody precedent of
#1091, #1148 and #1150: no commit in the branch keeps a private digest prefix.

observation-inference, slot usage-cost-monitoring:
- A3-0090: the template and the collector example set OTEL_METRICS_INCLUDE_ENTRYPOINT=true (off by default;
  env-vars.md:510, monitoring-usage.md:156 and :584 at client 2.1.300), so metrics carry app.entrypoint to split
  interactive, -p and SDK runs. The row records that the collector's transform/privacy metric allowlists (the
  repository profile, and the 2604 profile the live collector runs) do not keep app.entrypoint yet, so the collector
  sums the label away until they do; the host steps go to the command center as a host diff, named without a path.
  The dispositions catalog records the switch as adopt-pending on the template, and the new-WSL client configuration
  record gains a dated projection (404 pieces, one more wired through the otel-collector-contrib slot).
- B2-0005: signed in with the plan, claude -p, Agent SDK and third-party app runs draw the plan's usage limits
  (support article 15036540, its 2026-06-15 and 2026-10-07 updates); the Max and Team monthly API credits cover a
  linked Console organization's API-key usage only and leave the plan's limits unchanged (support article 17154008).

scheduling-supervision, slot scheduled-tasks:
- A5-0024, a correction: --resume and --continue restore CronCreate tasks that have not expired or passed, and a
  backgrounded session keeps them firing (scheduled-tasks.md:11, :228, :230). Measured on this host on 2026-10-15
  and retained in evidence/artifacts/d4-bg-trial-20261015/scheduled-task-carry.json: a one-shot CronCreate task
  fired at 07:36:00Z inside the copy that claude --resume <id> --bg started. The route now has resume and background
  paths run CronList first and cancel stale tasks with CronDelete (scheduled-tasks.md:159-160).

Digest custody (command center, 2026-10-15; the GPT read at 8c03cbe, P1): repository text cites committed files and
public sources only. test_the_rows_cite_no_scratch_path_or_unpublished_digest checks both rows for the scratch roots
and the private findings files' name and holds no private value; a synthetic negative control plants a scratch
reference with a synthetic digest and is caught. The slot manifest's sha256_when_labeled follows slots.json.

tests/test_native_practice_ledger_obs_sched.py: 6 of 6 tests failed before on the base (38 subtests); 7 pass now;
native-practice modules 106 OK; the mechanism modules pass except four CachedArtifactTests in
test_upstream_surface_watch, which fail the same way on the base; validate rc 0.

Co-authored-by: Claude Opus 5.5 <[email protected]>