[JacobStephens2]

Summary

prototypes/typescript/
├── Ticket → Claim → Worktree + embedded Factory skills → streaming Harness
│             └── Delivery → CI/conflict Repair → ready PR or draft salvage
├── Pickup → origin-ref reservation + Ready recheck → Ticket → owner release
│             └── bounded BFS: both safety properties + mutation counterexamples
├── shared schema → CLI JSON + HTTP API + OpenAPI + typed Console + Playwright
└── benchmark → fresh-worktree timings + binary/source sizes + runtime peak RSS
docs/research/typescript-vs-rust.{md,json} → measured comparison and limitations

Implements the whole Spec across #1836–#1840 as an isolated Bun-compiled Linux
executable. Prototype-only edits select TypeScript checks; ordinary Rust edits
do not select the prototype. The production Rust executable and ADR-0001 stay
unchanged. The comparison records narrower prototype coverage and distinct
model graphs; it supports a Day shift decision rather than asserting rewrite
equivalence. Console state is separate from Ticket records, with one local
token and a single writer. Continuation, Spec runs and Self-merge remain outside
the implemented prototype slice.

Run, Claim, Delivery, CI and Pickup transitions use tagged states with explicit
exhaustiveness checks. Pickup applies the Claim before dispatch; its model
boundary explains the resulting difference from Rust's known pre-Claim
dispatch counterexample. The compiled executable embeds Factory skills and
Console assets while using Git, gh and the representative Harness from PATH.

The comparison report
retains cold/repeat timings, source inventories, binary sizes, model timings and
architectural tradeoffs. This review adds startup/status peak RSS samples:
25.16 MiB for TypeScript and 6.75 MiB for Rust, with explicit workload limits.

Closes #1831

Evidence

Execution evidence

• Before: PATH="$HOME/.bun/bin:$PATH" bun test --cwd prototypes/typescript tests/benchmark.test.ts -t 'benchmark CLI'
  failed the added RSS regression: TypeError: undefined is not an object (evaluating 'report.runtimeMemory[language]'); 0 pass, 1 fail.
  After: the same command passed: 1 pass, 0 fail, 21 expect() calls.
• The Spec review's exact read-only JSON check failed with
  AssertionError: Spec #1831 requires memory footprint, but report contains no executable RSS/runtime-memory measurement.
  The same check passed after adding actual measurements for both standalone executables.
• The benchmark now runs python3 prototypes/typescript/memory.py <binary> status --json
  after each build. Linux child peak RSS is recorded in bytes with stdout,
  stderr, command and exit status, using isolated HOME/XDG state. The report
  preserves the original timing observations and separately identifies the
  supplemental memory measurement's source commit and toolchain.
• PATH="$PWD/.affected-tests/python-bin:$HOME/.bun/bin:$PATH" python3 -m unittest discover -s scripts -p 'test_*.py':
  Ran 37 tests in 16.255s, OK. The PATH selects Python 3.12; an initial
  attempt let a subprocess select Python 3.9 and failed on missing tomllib.
• PATH="$HOME/.bun/bin:$PATH" bun test --cwd prototypes/typescript tests/console.test.ts:
  4 pass, 0 fail, 61 expect() calls, including compiled CLI/API/OpenAPI parity,
  authorization and desktop/mobile Playwright Dismissal, reload and Undo.
• CARGO_BUILD_JOBS=4 cargo build --release --locked --bin thirdshift:
  passed in 5m 26s; this executable supplied the Rust RSS sample.
• Final PATH="$PWD/.affected-tests/python-bin:$HOME/.bun/bin:$PATH" CARGO_BUILD_JOBS=4 CARGO_PROFILE_DEV_DEBUG=0 CARGO_PROFILE_TEST_DEBUG=0 RUST_TEST_THREADS=3 scripts/affected-tests main:
  typecheck passed in 3.73s; all 56 prototype tests passed (457 assertions,
  99.24s). Local Cargo compilation used
  four jobs, three test threads and disabled debug symbols to limit disk use.
• The selector's Rust fallback, cargo test, failed after 1102.28s
  (compilation included). All 1,657 unit tests passed (3 ignored); integration
  checks passed through the earlier suites, then Console reported 235 passed, 1 failed, 1 ignored. The failure was the existing
  idle::unchanged_requests_reuse_retained_records_but_replacements_are_visible:
  unchanged request reread 156073 bytes (limit 48 KiB). Later integration
  binaries were not run after Cargo stopped.
• Isolated confirmation:
  PATH="$PWD/.affected-tests/python-bin:$HOME/.bun/bin:$PATH" CARGO_BUILD_JOBS=4 CARGO_PROFILE_DEV_DEBUG=0 CARGO_PROFILE_TEST_DEBUG=0 RUST_TEST_THREADS=3 cargo test --locked --test console idle::unchanged_requests_reuse_retained_records_but_replacements_are_visible -- --exact --nocapture
  failed with unchanged request reread 156041 bytes, matching the pre-review
  failure already documented in docs/research/typescript-vs-rust.md.
  git diff --exit-code main...HEAD -- src tests Cargo.toml Cargo.lock skills
  passed with an empty diff. This is a pre-existing production validation limit;
  no production change is included in this experiment.

Test seams: the benchmark CLI's structured results, execution and Worktree
cleanup with external compiler fixtures; the persisted empirical comparison;
the standalone executables' status --json workload. The existing Spec tests
also exercise temporary Git origins, fake GitHub/Harness processes, process-group
escalation, Delivery/Repair/salvage, reservation races and rendered Console flows.

Visual evidence

Implementation captures from #1844,
whose Console source/assets are unchanged by the Spec review. Before shows no
prototype Console listener; After shows seeded Inbox rows after Dismissal,
reload and Undo. This review reran the matching Playwright flow successfully.

Desktop · 1440×900

Mobile · 390×844

Review coverage

Independent Standards and Spec reviewers compared the whole Spec against
main (305923a8), including Ticket integration. Standards raised 0 findings;
Spec raised 1 finding, the missing runtime-memory comparison, reproduced and
fixed in 1600a772 with retained regression tests. Reports and each axis's final files-read
list are in .thirdshift-review-YojVXH/standards.md and spec.md; summary.json
records {"standards":0,"spec":1} before disposition.

Both axes' final passes read all 38 changed files, including the added memory
probe and amended results. No changed file was left unread. No generated
output was excluded: the Bun lock and structured benchmark results were inspected.

Unaddressed findings

Standards

None.

Spec

None. The sole finding was reproduced and fixed.

Security

None. The introduced Console development-page disclosure was reproduced and fixed.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Revert the isolated prototype, comparison and selector changes to remove the
experiment. No production migration or ADR change is included.

Blast Radius: prototype

Prototype validation requires Bun, Chromium and Python 3. The executable uses
real Git/GitHub/Harness commands when invoked on a repository; its narrower
contracts and recovery limits are documented in its README. Ordinary Rust
behavior is unchanged.

Built with codex codex-cli 0.160.0 · gpt-6.1-sol · high

Tickets

• TypeScript prototype: isolated bun build --compile binary, embedded skills, and Ticket run through Harness session #1836 landed with Add isolated TypeScript Ticket run prototype #1842
• TypeScript prototype: Ticket run Delivery, CI Repair loop, and discriminated-union state machines #1837 landed with TypeScript prototype: Ticket Delivery and bounded CI Repair #1846
• TypeScript prototype: Pickup run (ADR-0017 reservation) and in-test BFS state-space model checker #1838 landed with Prototype: TypeScript Pickup and bounded BFS verification #1845
• TypeScript prototype: full-stack shared-schema Console, /api/v1/*, OpenAPI, CLI --json, and Playwright slice #1839 landed with Add shared-schema TypeScript Console, API and CLI slice #1844
• TypeScript prototype: reproducible TypeScript vs. Rust benchmark runner and comparison report #1840 landed with Benchmark TypeScript prototype against Rust #1852