[Clear1nsight]

Change and mathematical scope

The connected caller in #6863 rejects its selected loader at the real environment-association factory because one source-owned path retains the unsuffixed predecessor root. Its waiting entry also launches the native view from the stage directory while the native guard requires the original historical cwd.

This contribution repairs both interfaces in one immutable source successor under research/h25-caller-binding-repairs/. Preparation structurally selects the exact association root, preserves strict authentication/calling-frame checks, and changes the existing native Popen cwd to READ_ROOT. Both stage-entry copies and dependent stage, queue, packet and manifest pins are rebuilt after the corrected leaves. Every other Python change reverses to exact selected-root rebinding or dependent runner pins; native/compiler/scientific bodies and historical controls remain unchanged.

This is a completed source correction with bounded controls, not complete runtime admission, baseline equality or a performance result. The remaining process-group transfer and complete baseline/comparison/CLOSE adoption are explicit. No multiplication theorem, selected frontier or original acceptance criterion changes.

Frozen contribution: c8f7bd582ffc1301b851dde2364c60f1636aac8b. All 82 remote file bodies and Git blob identities match the frozen export.

Task and overlap

Original investigation: #6321. Maintained specification and contributor handoff. Immediate predecessor: #6863 at c009a7514790e865b664ab8cd4688d37787d7eaa. Conditional native/compiler argument: #6830 at 39fcd67c991bd72918080b1d2c02feb3d679206d. Base: 7c49fe32c55de02c2c873655c804b16e4c3fef0a.

Reuse the existing private-loader/factory code, waiting entry and native guard. Adapt the earlier root-rebinding preparation to cover its omitted leading-slash string form. Reuse unchanged timer/error-custody controls without rerunning them. #6837/#6847 provide the genuine baseline and inherited-limit correction; #6736/#6762 retain their separate automatic-retention and successful-CLOSE adoption obligations.

Bounded all-author/all-state PR-body searches covered Linux-model, assembly, guardian/compiler and loader/association behavior; current implicated heads and public review/comment threads were refreshed. An unrelated historical ingress PR was excluded by its different contract. This is not an exhaustive repository novelty claim. Scope is refined to the two concrete production obstructions and their connected correction.

Validation

• Five new tests passed in normal mode and the same five in optimized mode, once each: 0.490 s and 0.504 s retained unittest durations. The enclosing operation used Python 3.12, a clean environment, 20-second timeout, two-second kill tail, 256 MiB AS cap and fixed file limits. These fixture measurements establish no speedup.
• The actual private loader, loaded association module, factory method, strict authentication and authentic calling-frame checks reproduce the frozen predecessor's rejection and let the successor reach its mandatory missing-selection boundary. No genuine association or native-I/O Context is created. Eight fixture substitutions/direct-call cases remain rejected.
• The complete production native-argv body and actual launch expression run with inert read/Popen ports. The predecessor cwd disagrees with argv[10]; the corrected cwd matches the unchanged historical native guard. No subprocess/native entry is launched; this is explicitly a source projection.
• Both frozen stage copies, dependent queue/source pins and every unchanged original stage member are checked. No old successful timer/process suite, native compilation or scientific workload was rerun.
• The passive archive integrity command passed for 81 manifest-listed files plus the manifest, 61 source roles and 7 profiles. Cold passive reproduction: python3 -I -S -B verify_integrity.py from the package directory. Executable fixture reproduction requires the exact retained local inputs; this archive is not a portable runtime installation.

Readable source mapping, preparation, raw new-control output, interface proof and Pro construction are included. No scientific replay hook is registered. Hosted planning/skipped replay will be reported against the actual submitted head. Complete physical/preparation/Pro/review/publication/final-metadata costs remain OPEN_NOT_ZERO.

Credit

Chris R / Clear1nsight; original CrocSwap contributors and the #6321/#6863/#6830 predecessors, with substantial OpenAI Codex and ChatGPT Pro assistance. Pro independently identified the two public-source defects and repair obligations; Codex implemented and checked the exact successor. AI source reasoning is distinct from independent human or proof-kernel acceptance. Original source notices and licences remain intact.

Open next questions

Prove the changed common-runner/waiting-entry process-group relation against the exact guardian and group-sensitive consumers; do not blindly change start_new_session. Then connect the genuine primary baseline/full comparison and complete-outcome CLOSE owners through the actual caller and accept the whole source-binding scope. Current native/kernel/compiler/configuration, ownership, resource/funding and complete normal/optimized outcomes remain separate prerequisites for full equality and measurement.

[Clear1nsight]

Submitted Ready for review at c8f7bd582ffc1301b851dde2364c60f1636aac8b, with maintainer modification disabled. All 82 remote file bodies and Git blob identities match the frozen export; the contribution is one commit based on 7c49fe32c55de02c2c873655c804b16e4c3fef0a.

Hosted run 42427499289, attempt 1, completed successfully on that exact head: plan passed; replay skipped. This supplies archival planning evidence, not candidate, native or scientific execution.

The two source corrections passed five new tests in each Python mode, including the actual factory boundary, eight strict fixture negatives, inert native-launch projection and both changed stage joins. Passive archive integrity also passed. Old successful timer/process controls were not rerun. Complete physical and outside costs remain OPEN_NOT_ZERO.

The maintained specification handoff on the original investigation now maps this PR and its limits. The source round is closed. Process-group preservation, the complete source/native binding scope, genuine baseline/comparison and successful CLOSE adoption, actual qualification/resource fit and full equality remain open. This delivery awards no original-activity or performance credit and starts no new execution or research round.

[Clear1nsight]

Follow-up source audit: PR #6866, at fcba647106f3e2f08de1b39ebed04a13c0a6ef8b, identifies a concrete limitation in this PR's common runner. It can reap its child before an unconditional final numeric group signal; a reused number can then identify an unrelated group. The Linux v6.12 source check explains why held pidfds do not prevent that numeric lookup race.

This preserves the factory/cwd corrections established here and the earlier timer/error correction. The audit's proposed signal-before-consuming-wait owner, its exact caller/source-binding successor and discriminating controls are not implemented or executed. Complete source/runtime qualification and the original comparison remain open. The maintained goal and contributor handoff records the next connected revision.