Add the Console Waterfall page

[JacobStephens2]

Summary

Adds page 4, Waterfall, with exact Command selection, nested Runs/stages/tool calls, and one shared time axis.

Command chip (live / died / ended)
  Run → stage → Claude/Codex tool call
                  match scoped Command-log line
                  unmatched → Time unknown, retained

Tool intervals extend to the next matched Session call or stage boundary and include waiting. The page labels this distinction from measured execution time. Both show --json and logs are registered in CLI parity.

Closes #827

Evidence

• Before: page 4 returned 404; the new HTTP matching test failed.
  After: cargo test --test console passes all 16 tests, covering real-shaped Claude/Codex events, duplicate calls/item updates, scoped Ticket Runs, unmatched calls, midnight rollover, fractional boundaries and CLI data parity.
• Before: each Spec-review reproduction failed: empty timing fields for a same-second Run start; a tool under Setup instead of implement; missing logs parity metadata.
  After: all three exact reviewer commands pass. Portable retained checks: python3 console/reproductions/waterfall-fractional.py, python3 console/reproductions/waterfall-stage.py, python3 console/reproductions/waterfall-parity.py.
• cargo check --all-targets, cargo fmt --check, cargo clippy --all-targets -- -D warnings, and the route parity test pass.
• cd console && npm test: 14 passed. Checks nested bar alignment/scale, Command chips, Session links, escaping, keys 4/f and 390 px layout. The final run also corrected an existing navigation test's stale expectation that key 4 was unavailable.
• cargo test: full Rust suite passed (repository-configured ignored tests remain ignored).

Merge Danger

Door: two-way

Revert the commits to remove the page and its reader.

Blast Radius: Console

Reads retained local logs when loading a selected Command. Tool durations represent recorded intervals; unmatched evidence remains visible without a time.

Test seams

• Built CLI (show --json, logs) and Console HTTP over isolated fixture records and real-shaped Session/Command logs.
• Console in Chromium for selection, nested timeline scale, escaping and 390 px layout.

Unaddressed findings

Standards

• Optional Repeated Switches / Shotgun Surgery at matching.rs:119: retain parser-local event filtering and raw-path matching. Plan decision: Interpretation::condense:192 returns mixed progress strings, so the reader needs to identify tools; matching.rs:119 reuses Harness formatting and line 147 reads raw paths because Codex does not retain cwd. A new shared typed-evidence interface would expand the contract for a single new consumer. The reviewer explicitly classified this as an optional design judgement, with no demonstrated behavioral failure. This implementation decision does not require a Spec or Day shift ruling.

Spec

None. All three findings were reproduced as written, fixed, and kept as passing regression checks.

Review coverage

Standards and Spec independently reviewed against fetched issue-776 commit 56938c1f8fe1afb36be72bfac96d825a988dbfb2, including one coverage follow-up each. Both read 15 of 16 changed files; no generated-output exclusions.

The review left console/tests/inbox.test.cjs unread on both axes. Its late test-only correction adds Waterfall to the key/page and jump-letter loops and removes the unavailable-key assertion. The primary agent inspected this diff, ran its targeted navigation check, and ran all 14 browser tests successfully.

Final findings, dispositions, each axis's complete files-read list and unread disclosure are retained locally in .thirdshift-review-xVlImL/standards.md and .thirdshift-review-xVlImL/spec.md.

Built with codex · gpt-6.1-sol · high