[Essoz]

PRs #140–#148 were merged into their feature-branch parents after #150 had already landed in main. Those later merges did not propagate to main, so main still lacked the reviewed tracing fixes, inference optimizations, and validation harness.

This PR targets main directly and consolidates the six current parent-branch tips that contain the reviewed work. All nine original GitHub merge commits for #140–#148 are ancestors of this branch; six new signed merge commits preserve that history. No feature implementation is changed.

Included work:

• Bound example reservoirs across construction and inference merges #140/Index trace contexts and call records during invariant inference #141/Cache process and thread identities during precondition inference #144/Index function-call time windows while preserving pandas query semantics #147: bounded example reservoirs, context/call indexes, process/thread identity caching, and function-call time-window indexing.
• Fix recorder re-entry and atomic trace buffer delivery #142/Preserve TorchScript builtin identity when wrapping PyTorch APIs #143/Handle dynamic non-string type-name metadata #146: recorder re-entry and duplicate-delivery fixes, TorchScript builtin identity preservation, and dynamic type-name handling.
• Isolate pruning state between independent precondition solves #145: independent precondition pruning state.
• Add fresh-run validation and reproduce object-ID false alarms #148: typed fresh-run invariant validation, explicit coverage/error states, tests, and compact experiment evidence.

Validation at candidate caabdd7c39090de934618e1512be2354867bd09f, based on main 7c72273e182f2c2ec3473ecfed775f048052465c:

• The entire resulting Git tree is byte-for-byte identical to the earlier validated full composition: 2177e2fa826bb4623c3c9cdbb3bdcb61fafbedab (source commit c0d3e9e). This covers all files and file modes, not normalized outputs.
• Fresh local rerun: 140 production + 28 harness tests pass, under Python 3.10.16, PyTorch 2.2.2, pandas 2.2.3, one native thread.
• Pinned Mypy 1.9 passes all 77 production files; stricter harness/test checks pass over 9 files.
• Configured Ruff/Mypy/Black/isort hooks pass on all 22 production/tool Python files changed relative to main. The three previously documented repository-wide formatting failures are outside this diff.
• All nine GitHub merge commits are independently verified as ancestors; all six new commit signatures verify; diff checks pass and the worktree is clean.
• Remote CI remains intentionally disregarded at the user's request because the runner is unavailable.

Existing performance evidence is carried forward unchanged: #144 reported 1,385.656 → 98.551 seconds for one baseline-first full-trace pair; #147 reported median 101.796 → 76.748 seconds across three paired repetitions (24.61% lower total wall time). These are the original experiments with their original revisions and limitations, not new measurements of this consolidation. Detailed provenance remains in the included measurement reports.

Use Create a merge commit to retain the reviewed history. This PR's target is main; it needs no parent-feature-branch landing sequence. No PR has been automatically merged.

Reproduction commands, exact refs, ancestry audit, and fresh test/check logs are retained in outputs/main-landing-20261026/ in the experiment checkout. No new traces, credentials, or machine-specific files are committed.