[loopx-agent]

Goal And Delivered Outcome

The initial-reader coverage test must distinguish emitted pointers from logical selected facts. With #7914's recursive fact helper, a delivered parent also exposes its descendants to assertions, so the existing duplicate-child assertion fails even when that child was not transmitted separately. Check the original selections pointers instead. Current and peer requirements, deferred independent material and source-integrity assertions remain intact.

Related: #7914. Intended base: main. This is a test-oracle repair; production behavior is unchanged.

Author Declaration

• Written by: model_agent — OpenAI Codex, OpenAI GPT-6 family.
• Specification: the existing no-duplicate-parent invariant in test_context_first_read.py at the verified base.

Criterion	Disposition	Evidence
A delivered parent is not repeated as a separate child selection	implemented	Inspect raw emitted pointers in the existing test
Complete current/peer requirements, deferred unrelated criteria and capture integrity	implemented	Existing assertions retained and the complete reader module passes

Self-check: inspected the helper and actual reader response; preserved all assertions and the original failing combination. No new abstraction, protocol field or runtime owner is needed.

Scope And Continuation

Complete within this test scope. The broader context proposals and their runtime/rollout acceptance remain with #7906, #7909, #7914 and #7923. The integration experiment is an isolated source combination, not an installed or merged delivery.

Validation

• Tested revision: a965b9266b935d92ee9bc496333ce1a0bc6baecc for the standalone proposal.
• Run state: finished.
• Input classes: synthetic, public_fixture.

Check kind	Result	Evidence / limitation
unit / real_entrypoint	passed	Complete tests/control_plane/test_context_first_read.py: 31 passed on this proposal
regression_parity / integration	passed	On frozen main 4250a94a plus #7906cf520e9, #790982dc675, #79140605a55 and #7923d188b26: original assertion gives 535 passed / 1 failed / 46 unrun native opt-ins; only this oracle correction yields 536 passed / 46 unrun across the same 13 modules. The six affected production-source hashes are unchanged
static / unit	passed	On that isolated combination: 47 related TS tests, zero skips, and control-plane typecheck. On this standalone proposal: Ruff and all five direct premerge checks pass; the risk selector chooses zero additional checks
integration	passed	Final managed premerge: five direct checks pass, zero additional checks selected, no manual holds; exact-scope receipt cqr_32d383ed3238dcdcc40e verifies valid for base 58a85d0a and head a965b926. The initial missing-receipt hold is retained as an earlier result; independent review and exact-head merge readiness are separate

The failure and the later corrected run are separate records. No CI was queried or manually triggered. Native binary, installed host, model outcomes and complete cost are outside this test-only change.

Frontend / Visual Evidence

UI impact: none. Screenshots, preview and viewport review: N/A; only an existing test assertion changes. Source data: synthetic/public fixtures.

Type Of Change / LoopX Area

• Test update; control-plane reader validation.
• Shared-authority fixture impact: N/A; no production schema, provider routing or promotion changes.

Boundary Checklist

• Public-safe diff and body; no private operating state, raw traces, credentials or local paths.
• Scoped to the demonstrated reader-oracle gap; no benchmark work duplicated.
• No paired RFC or UI change.
• Every commit has its DCO sign-off.
• Author proposal only; maintainer review and merge readiness remain separate.