Final calm engineering polish — periodic SV report grouping, readable SCL and verified candidate FAT gates

[masarray]

Latest addition: real per-channel SV freshness

• Full stream STALE: the waveform control binds SelectedStream.IsActive and mutes all voltage/current traces and legends, retaining the last valid oscilloscope evidence (same visual semantics as phasors). Prominent STALE · last valid samples lane header.
• Partial 4I+4V quality fault: only the channel whose verified instMag/q word indicates invalid/old/untrusted data is withheld. Its previous coherent waveform/phasor/RMS is shown as HELD at 38% opacity; if no prior valid samples exist it becomes MISSING. Five healthy phases remain normal. The stream becomes amber LIVE · WARN, not green.
• A channel is not switched to LIVE on the first good ASDU: it needs a full new coherent window. Zero-amplitude valid data remains LIVE; never infer stale from a flat waveform.
• All consumers share ChannelFreshness in the selected-stream analyzer SSOT: waveform, phasor, measurement rows, immutable runtime snapshot and power readout. Three-phase power returns N/A if any required channel is HELD/MISSING; held data is not claimed as a current measurement.
• Same raw decoding and aligned smpCnt window, no additional background timer, packet worker, or channel packet cache. Presentation uses per-trace opacity and existing frozen pens.
• Respect SCL channel-to-element mapping for quality bits when channel ordering differs, and fail conservatively for unverified layouts. Unsupported arbitrary layouts do not get guessed channel freshness.
• Fix an initial quality-window off-by-one regression discovered by CI: exact 2/4/8-cycle PCAP replay remains valid. Test covers one bad Ic, five healthy peers, recovery, missing-quality semantics and valid zero.

Previously implemented in this PR

• Phasor exact-tip callouts for co-linear U/I, stable label collision avoidance, snapshot-driven U/I scaling, and source-relative electrical angles.
• Npcap trusted native DLL selection; no fabricated fallback Ethernet NIC.
• Safe Windows Inno Setup installer with in-place repair and local PDF manuals regenerated from Markdown with SHA256 provenance.
• Public release remains manually gated on main and documented physical target-NIC/long-running injector field acceptance. No auto-merge or automatic stable release.

Validation

Current head automated Windows CI, Runtime Architecture, Runtime Stability, candidate package and CodeQL are the gate. Physical single-channel quality inject test on the target PC and the full 60-minute bench are still pending.

WPF visible-spacing correction — follow-up to physical screenshots (2026-10-20)

The first styling pass was insufficient: it standardized padding, but retained a visibly oversized 16-DIP gap between the 280-DIP explorer, central canvas and 320-DIP inspector, versus only 8 DIP below the capture header and 12 DIP between Phasors / Measurements.

Actual Grid geometry fixed in this follow-up:

• Outer shell horizontal gutters: 280,8,*,8,320, rather than 280,16,*,16,320.
• Analyzer interior canvas/inspector and phasor/measurement split: 8 DIP.
• Diagnostics hero/metric vertical and nine adjacent metric-card horizontal gaps: 8 DIP.
• GOOSE timeline/inspector: 8 DIP.
• Timing/PTP hero / evidence / events and three header cards: 8 DIP.
• SCL hero/body and catalog/inspector: 8 DIP.
• All primary panel sibling gaps now use compact 8 DIP, with independent 12-DIP inside-panel padding.
• Fixed explorer Grid.RowSpan=2 bug that caused the sidebar bottom to overlap the footer gap and fail to align with the main canvas bottom. It now occupies exactly Grid.Row=0, like the central canvas, above the shared 8-DIP footer gutter.
• The updated parsed-XAML CI contract checks root shell geometry, major tab grids, actual left panel row alignment, and legacy metric gaps. No data acquisition/decoder/phasor math changed.

Validation: Windows CI, runtime architecture/stability, CodeQL, dependency review, and Inno installer for current commit are the automated gates. Live Windows screenshots at 100%/125%/150% DPI remain a separate visual acceptance gate.

Physical screenshot verification and build provenance (2026-10-20)

Four same-resolution (1586x893) operator screenshots show a 16 px explorer/main gap (left edge x9 to right edge x288; next panel begins x305), and the SCL hero/body and analyzer right-stack still resemble the old 12 px split. The left explorer also extends ~8 px below the center panel. These are incompatible with the PR's parsed XAML 8-DIP shell and removed RowSpan, so do not claim that those screenshots prove Candidate 251 was installed or launch-path-correct. TabControl template uses zero padding and only a SelectedContent presenter, so a hidden 8-DIP template inset is not supported by source review.

Installer/runtime diagnostic added:

• Remove misleading static About labels Version 1.0.0 / Build 2026.04.
• Include BuildInfo.json with exact Git SHA, version, GitHub run ID, and SHA256 of packaged EXE in both the Inno and portable payloads.
• About shows the assembly version, short commit/CI number, and the actual executable path (Environment.ProcessPath); hover for the full SHA/path.
• Both ZIP validator and Inno install/repair smoke verify provenance manifest and actual EXE hash, making accidental stale/mismatched installation detectable.

Field acceptance still pending: Install the verified new candidate, open top-right Info → About, record the Build commit and actual executable path. Compare against candidate artifact and capture fresh screenshots of workspace gaps. If correct commit shows old 16px gutters at 100% Windows scale, investigate WPF visual-tree geometry/DPI live rather than making speculative spacing edits.

Calm engineering polish — baseline checkpoint and scope

Progress locked first at exact commit 03fc3657f2ba21df2bc570d2198c59bb2f2be2dc, on checkpoint/pre-calm-engineering-polish-20261020, with all six baseline workflows green. All subsequent work remains on this PR.

• Bounded exact-message DiagnosticEventSummarizer, groups repetitions as ×N, earliest/latest time tooltip; original forensic AnalyzerShellState.Events unchanged.
• Incremental display reconciliation prevents clearing/rebuilding up to 200 ListView items on each arrival; no new timer, packet worker, or waveform cache.
• Correct engineering semantics: signed host/Npcap ARRIVAL Δ · HOST vs clock jitter claims; missing smpCnt evidence labeled inferred gaps, not confirmed NIC packet loss.
• Static RX waveform badge separates reception from amber LIVE · WARN integrity. The former pulsing animation was removed.
• Compact U/I scale toolbar, full option tooltips; PT P transport-column widths corrected; calm truthful GOOSE empty states; SCL full technical identifier tooltips.
• Read the acceptance and locked baseline.
• Latest automated CI: WPF 8-DIP geometry PASS, Windows build PASS, 183/183 xUnit PASS, Runtime Architecture PASS, Runtime Stability PASS and Dependency Security PASS. CodeQL and candidate Inno verification remain explicit gates until their latest current-SHA workflows finish.
• Still DRAFT: public release requires final physical field acceptance, NIC/quality injection regression and operator DPI screenshots; no claim of live field certification.

Final five-tab screenshot follow-up — 2026-10-20

The 13:10 operator screenshots show restored premium/compact 8-DIP spacing, readable analyzer waveforms, clear passive PTP claims, and a meaningful SCL mismatch notice. Visual blocker: Diagnostics still shows dozens of periodic SV arrival timing excursions >= 300 us summaries. Root cause: the raw analyzer emits report messages with changing counters/latest-smpCnt values, while the UI initially grouped by full message. This is not proof that prior aggregation worked in the field.

Targeted correction at current candidate head:

• Exact known periodic family grouped by stable severity + svID + title, only if svID is explicit, retaining latest raw evidence, earliest/latest report timestamps and reporting ×N reports (not misleading ×N missing packets).
• Ordinary diagnostics, different svIDs and changing severity remain separate. Raw forensic events unmodified, bounded 200-row projection and incremental ListView reconciliation unchanged.
• New positive/negative regression tests for volatile reports, two publishers, severity differences and missing-source fail-closed behavior.
• Readability: SCL full signal references 12-DIP monospace / appropriate row height; compact P/Q/S/PF measurements raised to 12 DIP. No panel-grid or SSOT signal/phasor changes.
• Preserved checkpoint checkpoint/calm-polish-candidate-273-20261020 at 754ec5d850b1d2b1f7572db8b17f484f103b6681.

Stable public release deliberately BLOCKED until docs/validation/V1.4.1_PUBLIC_RELEASE_ACCEPTANCE.md physical target-PC gates are signed on this exact candidate: actual NIC and 60-minute capture, multi-setting U/I validation, 4000/4800 rates, quality-gap/held/missing tests, 100/125/150% DPI and Npcap/installer lifecycle. CI, screenshots and candidate packaging are not calibrated relay/clock-lock or packet-loss proof. Remain Draft PR; do not merge/publish stable without this field evidence.