[JetSquirrel]

Part of #64 and #65.

Why

A chart can mislead without being wrong about any one number. A line spaced one day and eight days alike, a NULL reading vanished under a line running across it, a pick on SQL NULL also caught the text 'NULL', and a pie drawn from a query cut short looked complete.

The audit specified in docs/dashboard-quality-audit.md is done, with the results in docs/dashboard-quality-audit-findings.md. Codex never ran on this (the repo has no Codex environment), so this was done here instead.

What

Severity	Before	After
P0	Uneven times evenly spaced	line and scatter placed by time, in dashboards and the results chart
P0	NULL reading vanished on a dashboard line	The time stays and the line breaks, with a note
P0	Picking NULL also matched the text 'NULL'	Each selects its own rows
P0	Query cut short, unsaid on most charts	Every dashboard chart says so
P1	A series vanishing recolored the rest	A series keeps its color by name
P1	One NaN made a whole average NaN	Left out, and counted on dashboards
P1	13 series read as "112"	Counts series
P1	open --state gave source rows only	Each plot adds scope: what it draws of its query
P1	Area over uneven times, unsaid	A dashboard area says it's evenly spaced

Decisions to review

• A line over times draws in time order, whatever the query's ORDER BY. A row with a NULL time is left off, with a note.
• A line breaks only at a NULL the query returned. Where a series has no row at all it still crosses, because breaking there would shatter a series that only has rows at some times. The wider spacing shows that gap.
• A pick on NULL narrows further than before. The old match on both was deliberate, and a dashboard relying on it now gets one or the other.
• scope is added beside rows, so the state schema version stays.

Gaps

• Not seen in the window. Nothing here has been looked at on screen; the findings end with a checklist. This is why it's still a draft.
• Area charts stay evenly spaced and don't break at NULL. The kit's area chart can't take time positions, and it draws a series' missing row as zero. That's the largest gap left, and it needs hand-built drawing.
• Smaller ones are in the findings: the results chart's NULL rows, bars over times that aren't whole days, a chart drawing NULL and 'NULL' as one band, and a card's tone shown by color alone.
• No zoom or pan. The Dynatrace comparison is written up and nothing was implemented; zoom on a time axis deserves its own issue.

Testing

• cargo test --locked: 495 unit, 35 CLI, 8 LSP and 4 network-path tests pass, with 1 ignored. Fmt and clippy are clean.
• New tests cover each fix, including a real query over the row budget, every timestamp type DuckDB renders, and a paint of a broken line on a time axis.
• Paint probes show no regression against main, in a debug build. They don't use a time axis or NULL breaks, so those paths have no baseline.

🤖 Generated with Claude Code

[JetSquirrel]

@codex Please implement the Dashboard quality audit in this draft PR, using docs/dashboard-quality-audit.md as the task specification.

You own the inspection, regression tests and focused fixes. Start by reading the existing dashboard/results-chart implementations and tests; do not reimplement features that already work.

1. Audit P0 data correctness first: missing/irregular timestamps and honest time scales, sample/series caps and truncation disclosures, NULL/zero/negative values, format/units, cross-filters and stable series identity.
2. Audit P1 states and interactions: loading vs empty vs error, SQL/source visibility, tooltip/legend/CJK readability, local map behavior, dense-chart performance.
3. Add deterministic regression tests for real findings, and implement only scoped fixes in this PR branch. Do not replace the chart library or add unrelated chart types.
4. Run cargo fmt --check, cargo test --locked and cargo clippy --all-targets --locked where available. Report any missing test environment.
5. Post a findings matrix with code references, P0/P1/P2 severity, evidence, and what was changed. Mark missing-but-large features as follow-up tasks, not surprise scope expansion.

Use Dynatrace Strato's chart guidance as a quality benchmark, not as code to copy. The maintainer will review the changes before merging. Please keep this PR as draft until findings and tests are ready.

[chatgpt-codex-connector]

To use Codex here, create an environment for this repo.