[JacobStephens2]

Summary

prototypes/console-split/
├── prepare.py       # pinned Base → core library + Console/CLI binary
├── benchmark.py     # validation and three sequential samples per layout
├── test_prepare.py  # export contract checks
├── test_report.py   # reported test totals match the recorded suites
└── results.json     # host, commands and individual measurements
docs/research/console-core-split.md

On the recorded four-vCPU host, Console edits rebuild 62.37% faster and
check 71.55% faster. Core rebuild and cold medians are essentially unchanged;
core checks cost another 1.19–4.67%. The report includes every sample,
method, host details, limitations and adoption costs. Production sources,
Cargo configuration, CI and ADR-0001 stay unchanged.

Closes #1889

Evidence

Execution evidence

• Before: python3 -m unittest discover -s prototypes/console-split -p 'test_*.py'
  failed with AssertionError: None != 'https://github.com/JacobStephens2/thirdshift'.
  After: the same command passes all three tests. The split now preserves
  CARGO_PKG_REPOSITORY, which the Eval replay importer uses to construct URLs.
• cargo build --locked and cargo clippy --all-targets -- -D warnings pass
  on the generated split.
• cargo test --test evals replay_import_completes_trial_inputs_and_preserves_local_edits -- --exact
  and cargo test --test packaging non_linux_compilation_reports_only_the_platform_error -- --exact
  both pass. The export preserves package metadata and Linux guards.
• cargo test --no-fail-fast on the split passes: 3927 passed, 0 failed,
  91 ignored, 2352.408 s, including remaining compilation.
• python3 prototypes/console-split/benchmark.py --samples 3 --jobs 4 --output /tmp/console-split-results-1889-v3
  passed: 48 measured samples, all 95 commands exited 0. Raw data is in
  prototypes/console-split/results.json.
• Spec review reproduction failed with report says 3930; top-level suites report 3927; the same command now passes and is retained in test_report.py.
  Prototype discovery passes 3 tests; existing Python tooling passes 41 tests.
• The Standards refactor preserves all 1396 generated files/links byte for byte.
• /usr/bin/time -p scripts/affected-tests main passed the full Rust suite via
  cargo nextest run: 3927 passed, 91 skipped, 2203.52 s, including compilation.

Test seams

• The prepare.py CLI's exported Cargo workspace: dependency direction, parser
  ownership, package metadata and byte-for-byte integration-test preservation.
• The prepare.py CLI's refusal to overwrite a nonempty destination.
• Unsupported-platform diagnostics from the exported binary and library roots.
• The research report's suite count against the recorded top-level Cargo results.
• Existing CLI and core/Console tests, including Eval replay import URLs.

Unaddressed findings

Standards

None. The one finding (duplicated archive export) was addressed.

Spec

None. The one finding (nested fixture results inflated the suite total) was addressed.

Security

None.

Security review outcome

No unaddressed introduced findings.

Review coverage

Both axes read all seven changed files, including every measurement record
and the new regression test. Generated-output exclusions: none. Unread changed
files: none. Review totals: Standards 1, Spec 1, both addressed.

Merge Danger

Door: two-way

The prototype and report can be removed without changing the production layout.

Blast Radius: prototype

The split runs only in exported temporary trees.

Built with codex codex-cli 0.164.0 · gpt-6.1-sol · xhigh