fix(frontend): show units and a generic heading in the block statistics panel

[wagenet]

Opened by Claude (Claude Code) on behalf of @wagenet.

The statistics panel from #857 ignored StatMetadata.unit, so a configured buffer of 400 ms read as 400 and packet counters as bare numbers. Its heading and empty-state hints also said "RTP", but any block that reports statistics can use the panel. The Audio Bridge Input in #844 is one such block. This PR does not depend on #844.

• StatValue::format_with_unit and Statistic::format_value in strom-types append the unit. DurationNs keeps the unit it picks itself, Bool stays Yes/No, and an empty unit ("") counts as none. No producer on this tree sends one, but the Audio Bridge helper in feat(audio): adaptive audio bridge for the conversation return #844 does for its Bool stats.
• The panel uses format_value. The heading is now "📊 Statistics", and the hints read "Statistics appear once the block receives a stream." and "Start the flow to see this block's statistics."

Tests

• New strom-types unit tests in stats::tests: 5 ran and passed. I reverted each branch of the formatter in turn: ignoring the unit, appending it to DurationNs, appending it to Bool, and treating "" as a unit. Each revert made at least one test fail.
• cargo test --lib stats:: (backend stats tests): 9 passed. No other test suites were run locally.
• cargo fmt --check, cargo clippy --all-targets -D warnings, the WASM clippy (--target wasm32-unknown-unknown, as in CI) and the trunk WASM build all passed.

Visual check

I ran a local backend with a WHIP Input flow (audio only) and a headless whipclientsink publisher, then viewed the web UI in Chrome. The panel showed 8289 packets, 2.82 ms average jitter (no doubled unit), 400 ms configured buffer and 456 requests. Both empty-state hints appeared as intended: one with the flow running and no publisher, one with the flow stopped. I did not check an AES67 input visually. It produces the same statistics through the same RtpJitterbufferStats::to_statistics.

🤖 Generated with Claude Code

[srperens]

Verdict: Request changes. The formatter and the panel change are both correct, and all five new tests ran in CI. What blocks is one comment in the tests. It says the current /rtp-stats endpoint does something it does not do on this tree. Correcting that comment is the whole request; the rest of the review would approve.

Requested changes

1. Fix the provenance comment at types/src/stats.rs:348 — // /rtp-stats sends Bool statistics with an empty unit, not null. On this branch and on main, the only producer of Statistics is RtpJitterbufferStats::to_statistics, and it emits no Bool at all. Its units are all non-empty, e.g. types/src/stats.rs:180 — unit: Some("packets".to_string()),. The empty-unit Bool stats come from #844's Audio Bridge helper, which does unit: Some(unit.to_string()) for producer_attached/producer_overrun. Say that, or state the rule generically: "an empty unit counts as none". The same sentence in the PR body needs the same fix. The handling itself is right, so keep it.

Claims

Claim	Verdict	Evidence
The panel now shows units for every statistic it lists	CONFIRMED	Both render sites switch to the new method: frontend/src/properties.rs:1279 — let formatted = stat.format_value(); and frontend/src/properties.rs:1306 — let formatted = stat.format_value();. A grep finds no other StatValue::format() caller in frontend/src
DurationNs and Bool never get a doubled or odd unit	CONFIRMED	The first match arm of format_with_unit returns the bare value for both variants (quoted below). avg_jitter_ns carries unit: "ns", so without that arm it would read "1.50 ms ns"
/rtp-stats sends Bool statistics with an empty unit	CONTRADICTED	See requested change 1. This is true only once #844 lands
The five new tests guard the formatter	CONFIRMED	Check (Linux) job 108422252909 at b373d30: stats::tests::{numeric_values_show_their_unit, duration_keeps_its_own_unit, bool_takes_no_unit, missing_or_empty_unit_shows_bare_value, jitterbuffer_stats_read_with_units} ... ok. They call Statistic::format_value itself, so reverting any arm fails at least one of them

The arm: types/src/stats.rs:66 — (StatValue::DurationNs(_) | StatValue::Bool(_), _) => value,

Radius — LOCAL. The change is display only. strom-types gains two methods and no fields, so the serialized shape and the OpenAPI schema are unchanged, and API Contract Check is green. The heading and hint text change for every block that shows the panel, which on this tree means AES67 Input and WHIP Input.

Overlaps — #844 is the other producer this PR anticipates. It extends RTP_STATS_BLOCK_DEFINITION_IDS and gains units from this PR, but neither PR needs the other. #809 edits frontend/src/properties.rs about 60 lines below these hunks, so a rebase may conflict, but the overlap is textual only. Neither settles a design question.

Tests & CI — Run 36248581300 at b373d30: Check (Linux), both Linux builds, WASM, API Contract Check and Static JS Tests all green. macOS and Windows were skipped, and nothing here is platform-specific. The UI change is covered only by the author's visual check. That is fine for strings and a method swap.

frontend/src/properties.rs is 2961 lines, far past CLAUDE.md's 1500-line mark. The stats panel would make a clean first sub-module, but not in this PR.

Confidence: HIGH

[wagenet]

Posted by Claude (Claude Code) on behalf of @wagenet.

Fixed in 2d60c59. The test comment now states the rule generically ("An empty unit counts as none"). The PR body now says no producer on this tree sends an empty unit, and that #844's Audio Bridge helper does for its Bool stats. No code change. The five stats::tests tests pass locally.

[srperens]

Supersedes my Request changes at b373d30. The verdict moves to Approve, because the one requested change is made.

Verdict: Approve. At 2d60c59 the only change since my last review is one comment line. compare b373d30...2d60c59 shows ahead_by=1, behind_by=0 and one file. The PR body has the same fix.

Claims

Claim	Verdict	Evidence
The provenance comment no longer credits /rtp-stats with empty-unit Bool stats	CONFIRMED	types/src/stats.rs:348 — // An empty unit counts as none. The PR body now says "No producer on this tree sends one, but the Audio Bridge helper in #844 does"
Panel units, the DurationNs/Bool arm, and the five tests guarding the formatter	CONFIRMED	Confirmed in my review of 2026-09-28 at b373d30. None of those lines moved

The arm is unchanged: types/src/stats.rs:66 — (StatValue::DurationNs(_) | StatValue::Bool(_), _) => value,

Tests & CI: I checked run 36486749200 at 2d60c59. Check (Linux), both Linux builds, WASM, API Contract Check and Static JS Tests are green. The Check (Linux) job log (109145432918) shows all five stats::tests::* tests ... ok. macOS and Windows were skipped, and nothing in this diff is platform-specific.

Radius: LOCAL, carried from my previous review. Overlaps: #844 and #809 are unchanged from my previous review. Both are textual or anticipatory, and neither settles a design question.

Confidence: HIGH