Preparation record

At base 64d091e5ad3f277cce77b88ae0de6eea4c49b4df, the first invocation of nu audit/2026-10-17-hemera-feature-performance/plan-gate.nu . /Users/master/.cache/cyber/nox/hemera-feature-performance/plan-thumb exited 1 before Cargo execution: a multiline predicate without enclosing parentheses parsed and as an external command. The exact initial runner is retained as plan-gate-initial.nu.txt; the correction adds only those parentheses. This is a capture-runner preparation failure, with no source compilation result.

The corrected runner then stopped before Cargo with complete unchanged source inventory: it selected the hygiene intermediate sources-final.json, whose trace.rs predicate predates the accepted rfind correction. The retained plan-gate-intermediate-inventory.nu.txt identifies that invocation. The runner now reads the actual accepted sources-reviewed.json, as the hygiene README and root review require. No compiler input was changed to satisfy this bridge.

The actual first thumb invocation is retained in plan-thumb.json and raw streams. At the same base it exited 101 with error[E0463]: can't find crate for core. Its source inventory exactly matches the accepted hygiene compiler inputs. Root authorized installing the exact target component. The captured rustup target add --toolchain 1.95.0 thumbv7em-none-eabi command reports that component already up to date; installed-targets.json also lists it before installation. The initial inherited PATH placed Homebrew tools before rustup. The first invocation did not capture the child compiler command, so its exact compiler selection cannot be reconstructed from the version string alone.

plan-gate-inherited-path.nu.txt retains that runner. The current runner puts the absolute rustup 1.95.0 toolchain directory first in an allowlisted PATH, clears compiler overrides, and captures executable hashes. Its one distinct rerun, nu audit/2026-10-17-hemera-feature-performance/plan-gate.nu . /Users/master/.cache/cyber/nox/hemera-feature-performance/plan-pinned-thumb plan-pinned, passes the named thumb gate with no warnings and identical source inputs (plan-pinned-thumb.json). The original red remains evidence. Broad hygiene checks were not rerun: their exact accepted compiler-input inventory is bridged, and only the named thumb gate was compiled in this plan unit.

original-replay.json retains the original portable Nu verifier invocation and successful result at the same base: 921 logical payloads, 103 execution records, and the initial 5/9 and repeat 6/9 threshold crossings. Historical Ruby was never executed. Root additionally requires absolute RUSTC/RUSTDOC pins for all forthcoming cell builds and timing, beyond the clean PATH used by the plan gate. No benchmark builds or timed invocations have occurred at this preparation point.

The subsequent source and registry exports completed unchanged. Initial build runner parsing rejected a local parameter named env because Nushell reserves that variable; build-reserved-var.nu.txt and control-reserved-var.nu.txt retain that pre-execution runner. Renaming it to run_env allowed execution.

The A build completed successfully once, but postprocessing expected every -vv --message-format=json stdout line to be JSON. Cargo also emitted ordinary build-script lines, beginning with [proc-macro2 1.0.107] cargo:rustc-check-cfg. build.nu, its raw A receipt, and original tool/script freeze remain unchanged. finish-builds.nu resumes by reading the existing successful A receipt, parses only JSON-object lines while retaining all raw output, and builds each remaining cell once. This correction causes no repeated benchmark build or measured run.

The inspection runner initially failed Nushell parsing because a closure captured a mutable local accumulator. inspect-mutable-capture.nu.txt preserves that pre-execution source. Assigning the completed accumulator to an immutable binding allowed inspection; neither a benchmark build nor a measured run was repeated. Both Rust audit helpers compiled with -Dwarnings. A measure smoke invocation wrapped only /usr/bin/true; its before/after telemetry and process receipt are retained separately from the finite benchmark schedule. No benchmark binary was executed by that smoke check.

The portable replay initially stopped while remapping source paths: Nushell parsed update cells as its built-in cell-update subcommand, which requires a closure. portable-initial.nu.txt retains that source. Using upsert cells selects the intended record-field operation. The failed replay removed only its own temporary extraction; retained capsules and all compiler inputs stayed intact.