[itz4blitz]

Summary

• Redesigns Call health into a cleaner card: status badge, KPI strip for ping/jitter/loss, and quieter session details with a path pill.
• Adds a last-60s chart of ping and receive bitrate (seeded under --shot so screenshots are not empty).
• Keeps the same float and Voice settings entry points.

Screenshot

Test plan

• cargo test -p tinyclient --features gui --bin tinyclient chart_keeps
• cargo test -p tinyclient --features gui --bin tinyclient panel::tests
• --shot with TINYCLIENT_SHOT_CALL=tiny_joined --mode tiny --tiny-menu stats
• Join a Tiny voice channel, open Call health from the panel, confirm the graph fills over time

[unRekable]

Code Review Summary

Verdict: Approved ✅ (0 critical, 0 warnings, 4 cosmetic suggestions) — read the full diff plus surrounding panel/chart code.

💡 Suggestions

• panel.rs:138 — the X axis spaces samples by ring index, not by time, so "The last 60 seconds" is really "the last 60 polls". Given a 1 s poll cadence that is fine, but a timestamp (or a short doc note) would make the label precise when mediad answers slowly for a stretch.
• panel.rs:435 — KPIs paint 0 ms, 0 kb/s and 0.0% in green before any measurement arrives (rtt_ms is None -> unwrap_or(0)), which reads as perfect health in the first second of a call. A neutral color while None would say "not yet".
• panel.rs:251 — the chart footer shows the raw last sample while the Ping KPI above is steady-smoothed, so right after a spike the two numbers in one card can disagree for a beat. Harmless; aligning would be tidier.
• panel.rs:380 — health.hints() is built twice here; the second call can reuse the first binding.

✅ Looks Good

• The 60-sample ring is exactly right: pops front only when at capacity, so the window is precisely a minute with no off-by-one and no unbounded growth; cleared on every end-of-call and error path.
• The --shot seeding of the chart is feature-gated and returns before any real poll task starts — screenshot data cannot leak into real builds or real calls.
• Degenerate cases are all handled: empty ring early-return, divide-by-zero-proof scaling, single-sample and all-zero data keep points inside the canvas.
• Compared against the old card, the rewrite drops nothing: channel/bitrate, ping/jitter/loss, sending/receiving/codec/path all survive, and the 288->320 width is consistent in both entry points.
• The ring test pins the exact off-by-one behavior (70 samples -> window 60 with correct head and tail), and the chart only paints while the card is mounted, so the cost-when-hidden is unchanged from before.

CI note / manual check: the failing Windows job is sending_echoes_once_and_failures_say_why timing out in chatd --test discord — this PR's diff is GUI-only with zero chatd/transport changes. I re-ran that exact test from a checkout of this branch: it passes (~4 min). Treat it as an unrelated Windows timing flake and re-run the job. I also verified the open checklist item in code: the 1 s poll feeds the chart ring regardless of card visibility, so the graph does fill over time — worth one real-network session to confirm visually and then tick that box.

[unRekable]

💡 Suggestion: x_at spaces samples by ring position, so the axis is sample-index rather than time — with a slow stretch of mediad responses, the "60 seconds" compresses silently. A timestamp on ChartSample (or a doc note) would make the label precise; low priority given the 1 s poll cadence.

[itz4blitz]

Done: samples carry Instant and the X axis is time-based.

[unRekable]

💡 Suggestion: with rtt_ms still None, the unwrap_or(0) zeros render in the positive/green tone, which reads as "excellent connection" during the first second of a call. Remaining plain/neutral until the first real sample would match the old card's instinct.

[itz4blitz]

Done: unmeasured KPIs show — in soft text instead of green zeros.

[unRekable]

💡 Suggestion: the footer value comes from the raw last sample while the KPI above comes from the smoothed health.rtt_ms, so after a spike the card shows two different ping numbers for a beat. Reading both from the same source would keep them consistent.

[itz4blitz]

Done: chart footer uses the same smoothed ping (and current recv) as the KPI strip.

[unRekable]

💡 Suggestion: health.hints() is called twice in this render (with take(1) here, again below); building it once and reusing the binding is cleaner. Pure style today, but it also avoids a divergence if the two call sites ever stop matching.

[itz4blitz]

Done: hints are built once and reused.

[unRekable]

✅ Nice: popping the front only when the ring is full-before-push keeps the window at exactly 60 samples with no off-by-one, no unbounded growth, and VecDeque of a copy-able struct keeps it cheap. The test pins this precisely (70 samples -> correct head 10 / tail 69).