@@ -10,6 +10,8 @@ Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "ProcessBus.Iec61850.Raw", "
 EndProject
 Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "ProcessBus.Tests", "tests\ProcessBus.Tests\ProcessBus.Tests.csproj", "{2F4B9C6E-7A31-4E8D-9B5A-6C1D82E4A0F7}"
 EndProject
+Project("{FAE04EC0-301F-11D3-BF4B-00C04F79EFBC}") = "ProcessBus.InteropCli", "tools\ProcessBus.InteropCli\ProcessBus.InteropCli.csproj", "{6F21A42B-A4EE-4901-8336-05A85E40D15F}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Any CPU = Debug|Any CPU
@@ -50,6 +52,14 @@ Global
 		{2F4B9C6E-7A31-4E8D-9B5A-6C1D82E4A0F7}.Release|Any CPU.Build.0 = Release|Any CPU
 		{2F4B9C6E-7A31-4E8D-9B5A-6C1D82E4A0F7}.Release|x64.ActiveCfg = Release|x64
 		{2F4B9C6E-7A31-4E8D-9B5A-6C1D82E4A0F7}.Release|x64.Build.0 = Release|x64
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Debug|Any CPU.ActiveCfg = Debug|Any CPU
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Debug|Any CPU.Build.0 = Debug|Any CPU
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Debug|x64.ActiveCfg = Debug|x64
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Debug|x64.Build.0 = Debug|x64
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Release|Any CPU.ActiveCfg = Release|Any CPU
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Release|Any CPU.Build.0 = Release|Any CPU
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Release|x64.ActiveCfg = Release|x64
+		{6F21A42B-A4EE-4901-8336-05A85E40D15F}.Release|x64.Build.0 = Release|x64
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE
 
@@ -0,0 +1,78 @@
+# M1/M2 receiver hardening — 17 October 2026
+
+Scope: Process Bus Insight / DigSubAnalyzer receive-only application. This record
+describes the bounded software tranche proposed in PR #29. Physical interoperability
+and timestamp accuracy require separate evidence.
+
+## Non-negotiable IEC 61850 boundaries
+
+- The ARStack injector's canonical SCL-derived multicast/VLAN/APPID/svID identity and
+  broadcast `AR_DIAG_SV1` mirror are distinct streams. A mirror observation cannot
+  establish canonical multicast/VLAN acceptance.
+- One frame may contain multiple SV ASDUs; each ASDU contributes to sample sequence,
+  but the frame has only one host receive timestamp. Do not invent intra-frame jitter.
+- `SmpRate/SmpMod` are sampling metadata, not universal sample-counter modulo rules.
+  Counter wrap conclusions remain bounded by observed and profile-supported evidence.
+- The eight (INT32 instMag, UINT32 q) pairs are a 64-byte 4I+4V *profile-specific*
+  decoding authority. Never apply their quality interpretation to arbitrary SCL
+  layouts solely because the payload is 64 bytes.
+- Quality with unusable validity/detail flags immediately invalidates displayed
+  engineering measurements; wait for a new complete coherent window before resuming.
+  A TEST or operator-blocked quality remains visible with its truthful context.
+- PTP observations are scoped to domain/sourcePortIdentity/transport; observation
+  does not imply clock lock, grandmaster traceability, power-profile compliance,
+  or reliable software receive timing.
+
+## Data plane and UI rules
+
+- One dedicated capture pump preserves input ordering. Do not introduce independent,
+  unordered parallel SV workers or silent lossy queues merely to increase thread count.
+- Npcap owns raw frame memory only until its next API read. The source transfers
+  a newly allocated owned frame to the analyzer. The latest payload is retained for
+  debug-on-demand; indiscriminate ArrayPool return would corrupt this evidence.
+- Per-ASDU projections use direct bounded INT32 reads and stack-local eight-channel
+  values. SCL eligibility and explanatory labels refresh when mapping/sequence state
+  changes, not for each SV frame.
+- Rolling jitter maximum uses a bounded monotonic ring rather than scanning the
+  entire window for each sample. Retire the unused counter dictionary/linear scan.
+- The active per-stream coherent waveform buffer is the single snapshot authority.
+  RMS/phasor/waveform projections of the same generation reuse that window; an
+  invalid quality state must invalidate prior measurements, never show stale values
+  as live.
+- User-visible SCL comparisons may be coalesced to 500 ms, with explicit user SCL
+  reload forcing immediate recomputation.
+- Stream history has a 256-identity guard. Reclaim only stale unselected state;
+  when at active capacity, report rejection rather than silently evicting a stream
+  currently being inspected.
+- PTP source history is bounded and stale sources are labeled STALE instead of
+  green OBSERVED; the interface should emphasize the fault and next diagnostic
+  action rather than flooding raw details.
+
+## Automated acceptance
+
+`dotnet build ProcessBusSuite.sln -c Release` and `dotnet test
+ProcessBusSuite.sln -c Release` plus repeated RuntimeStability acceptance must pass.
+
+New regressions include:
+
+- O(1) rolling jitter reference equivalence under a 30,000-sample run;
+- invalid-quality suppression, complete coherent recovery, and TEST labeling;
+- multi-ASDU continuity with a single frame timestamp and independent frame count;
+- per-domain PTP source isolation and stale Announce;
+- strict PTPv2 length/transportSpecific parsing;
+- bounded stream identity churn and preserved selected stream;
+- rejection of `SmpRate` as an unverified counter modulus.
+
+## Evidence still required before field/product claims
+
+1. Trusted independent canonical multicast/VLAN capture from ARStack ESP32-P4.
+2. Same-run canonical SV (4000/4800 samples/s) plus Announce, Sync, Follow_Up,
+   Pdelay and `smpSynch` transition stimuli.
+3. Npcap/adapter drop counters, host CPU/GC/memory, sustained capture throughput,
+   and 60+ minute bench soak with no cross-stream contamination.
+4. Hardware timestamp validation for any measurement-grade jitter/offset statement.
+5. Multi-vendor profile/SCL fixtures, arbitrary `nofASDU` and vendor DataSet layouts,
+   and receiver/IED behavior with real devices.
+
+The software tranche avoids claiming those field results. A green CI result means
+the bounded software regressions passed, not that an IED accepted the traffic.
@@ -0,0 +1,94 @@
+# M3 operator workspace and M1/M2 closeout — 17 October 2026
+
+This engineering handoff describes the implemented changes on PR #29, and
+distinguishes a successful software acceptance run from hardware interoperability.
+
+## Operating contract
+
+Process Bus Insight remains a **receive-only** IEC 61850 SV/GOOSE/PTP analyzer.
+The operator workflow is:
+
+1. Select the physical Ethernet/Npcap adapter.
+2. Press **Start** to observe traffic.
+3. Choose an SV publisher in the explorer to inspect waveform, RMS, phasors,
+   stream identity, quality and counter continuity.
+4. Read the one-line context-aware status for the current situation.
+5. Open **Diagnostics** for capture-driver drops vs SV sequence evidence, and
+   **Timing/PTP** for passive source-specific PTP evidence.
+6. Open **SCL/Validation** only to compare engineering-file expectations with
+   actual wire-observed identities; **Advanced** holds low-level raw evidence.
+
+No value inferred from the diagnostic broadcast mirror is presented as proof
+of canonical multicast, APPID, VLAN, PCP or SCL profile acceptance.
+
+## M3 implementation
+
+- Main capture-status strip displays one concise contextual message instead of
+  five overlapping status chips. The persistent footer is short and conditional:
+  it surfaces diagnostic-mirror, invalid-quality and stale-stream concerns.
+- Real SV measurements are never tweened/interpolated between capture snapshots.
+  Dead 30 FPS composition-loop/interpolation code was removed from the waveform
+  control: WPF paints the latest decoded, coherent snapshot atomically.
+- `LiveRenderCoalescer` compares **every numeric waveform sample** plus
+  series/shape/status metadata and channel RMS/angle/instant/units. It suppresses
+  equivalent redraw notifications while preserving tiny real transients, shape
+  changes, quality invalidations, stream selections and cleared values.
+- Phasor lists and per-stream UI workspace objects are reused while unchanged.
+  Periodic SelectedStream/IsRunning re-notifications were eliminated to prevent
+  selection churn.
+- Diagnostic navigator refresh is capped to the passive UI cadence;
+  managed-memory sampling runs at 1 Hz; SCL binding joins are coalesced.
+- Engineering tables use recycling/row virtualization and shorter minimum row
+  heights. Static flow instructions were removed from the main content.
+- PTP two-step sequence correlation is passive and displayed only in the Timing
+  workspace; there is **no** claim of measured PTP lock or UTC traceability.
+
+## M1/M2 additional correctness/efficiency
+
+- PTP packet-rate tracking uses fixed 5-second buckets, not an unbounded list
+  of packet arrival timestamps per source.
+- Announce/Sync/Follow_Up evidence is isolated by source/domain/transport;
+  Sync+Follow_Up pairing additionally requires matching sequence and bounded
+  freshness. Source history and pending correlators have hard capacity limits.
+- Diagnostic/event collection indexes persist across UI refreshes, and remain
+  bounded at 200 visible rows; reset removes both rows and indexes.
+- Stable SV explorer rows are retained rather than replaced on every update.
+- Npcap native drop statistics are polled at 1 Hz off the capture hot path.
+  Driver drop counters are displayed as **capture-path evidence**, and are never
+  folded into IEC 61850 sample-counter packet-loss inferences.
+- Full IEEE 61850 profile compliance, PTP servo discipline, calibrated host
+  jitter, and physical IED subscription are not asserted here.
+
+## Automated regression gates
+
+Windows .NET 8 full build and complete xUnit suite must pass together with
+repeated RuntimeStability, CodeQL, and Dependency Security Review workflows on
+the **same final commit**.
+
+New coverage includes render-coalescing correctness, multi-ASDU timing,
+quality invalidation/recovery, rate-window boundedness, cross-source two-step
+correlation, stream catalog pressure and reset-safe UI event deduplication.
+Do not merge if final-commit CI is red or incomplete.
+
+## Physical and measured acceptance still necessary
+
+The following must be collected on real hardware before an
+**interoperable/field-ready** release claim:
+
+- independent preserved-VLAN TAP/SPAN capture of ARStack canonical multicast
+  alongside the separately labeled diagnostic mirror;
+- 4000/4800 SV samples/s and concurrent L2 PTP source / receiving profile;
+- verified observed APPID, VLAN, MAC, confRev, svID, quality, smpCnt rollover,
+  two-step matching, source expiry and smpSynch transition evidence;
+- actual Npcap drop counters, process CPU, allocation rate, Gen0/Gen1/Gen2 GC,
+  heap/working set, UI frame responsiveness during sustained mixed traffic;
+- 60+ minute hardware soak and multi-vendor IED capture/acceptance matrix.
+
+**Receiver-worker decision:** maintain one ordered dedicated capture worker until
+independent profiling demonstrates its insufficiency. Unbounded independent
+workers or lossy frame queues would violate sample continuity. A multi-worker
+dispatcher requires bounded queues, source-preserving ordering and explicit
+drop accounting and remains a future measured-design gate, not a hidden
+"optimization" in this PR.
+
+The software milestone can be reviewed independently of that bench evidence.
@@ -0,0 +1,146 @@
+# M4 — ARStack Injector ↔ DigSubAnalyzer interoperability acceptance
+
+Date: 17 October 2026  
+Repositories: `masarray/DigSubAnalyzer` (consumer, **receive-only**) and
+`masarray/arstack61850` (ESP32-P4 SMV Injector).
+
+## Three different evidence levels
+
+1. **Software replay gate (automated):** anonymous synthetic PCAPs and SCL-intent
+   fixtures exercise the **same production parser** used by live reception.
+   This demonstrates decoder behavior but not a physical process bus.
+2. **Independent observed-wire gate (operator bench):** capture on an independent
+   TAP or correctly configured SPAN port that preserves multicast and 802.1Q
+   tags. Compare observed frames with the actually deployed SCL profile.
+3. **Device/IED gate (external equipment):** verify subscriber/IED acquisition,
+   quality, synchronization and sustained operation with equipment-approved
+   tools. DigSubAnalyzer alone cannot certify relay subscription, functional
+   safety, protection action or hardware PTP synchronization.
+
+The M4 software utility intentionally reports **WIRE_FIELDS_MATCH** instead
+of “compliant”, “certified”, “locked” or “accepted by IED”.
+
+## Run the offline acceptance inspector
+
+Windows with .NET 8 SDK installed:
+
+```powershell
+dotnet build .\ProcessBusSuite.sln -c Release
+dotnet test .\ProcessBusSuite.sln -c Release --no-build --filter Category=M4Interop
+dotnet run --project .\tools\ProcessBus.InteropCli -c Release -- --pcap C:\lab\arstack-canonical.pcap --scl C:\lab\current-profile.scd --max-frames 20000000 --json C:\lab\interop-report.json
+```
+
+The tool supports classic Ethernet PCAP 2.4 (microsecond/nanosecond). If the
+capture application creates PCAPNG, export/copy the packet recording as
+**classic PCAP** with full Ethernet headers before inspecting it.
+
+`--scl` is optional for wire fingerprint only; mandatory for automated
+expected-vs-observed verdicts. The CLI prints a short operator summary,
+keeps raw details in optional local JSON, and returns:
+
+- **0**: screening executed and no specified SCL comparison failed; **not**
+  certification and **not** proof of capture-path visibility. For a capture
+  without SCL, there are no configured fields to validate.
+- **3**: observed/configured mismatch, unresolved SCL identity or expected
+  stream not observed; this includes **MIRROR_ONLY**.
+- **2**: malformed arguments, unreadable/unsupported PCAP, or no useful
+  successfully decoded traffic.
+
+Bench files, identities, MAC addresses, PCAP, JSON logs and original SCL stay
+on the authorized **local machine**; do not push them to GitHub.
+
+### Verdicts
+
+| Verdict | Interpretation | Operator action |
+| --- | --- | --- |
+| `WIRE_FIELDS_MATCH` | Visible fields match configured SCL stream | Verify TAP/SPAN provenance and independent IED behavior |
+| `MIRROR_ONLY` | Only ARStack diagnostic broadcast identified | Fix capture visibility; do not infer canonical multicast/VLAN |
+| `NOT_OBSERVED` | Configured svID absent | Check VLAN membership, multicast filtering, SPAN/TAP and NIC |
+| `CONFLICT` | APPID/MAC/VID/PCP/confRev/nofASDU differs | Inspect deployed SCL and exact wire identity; do not normalize away |
+| `INCOMPLETE_INTENT` | Some observed fields match but SCL lacks authority | Load the deployed full SCL/profile; do not auto-fill defaults |
+| `UNRESOLVED` | No trustworthy SCL svID | Correct/resolve engineering intent before claiming match |
+
+Capture **VLAN absence on a Windows USB-NIC is not proof of untagged wire TX**.
+A mirrored `AR_DIAG_SV1`, APPID `0x4F01`, broadcast/untagged stream is diagnostic
+evidence only, always separate from the canonical SCL stream.
+
+## Physical bench procedure
+
+1. Record ARStack Injector firmware build/manifest, selected immutable profile,
+   source identity, configured DataSet order, 4I+4V payload authority, APPID,
+   multicast MAC, VLAN ID/PCP, confRev, nofASDU, sample cadence, smpCnt modulus
+   policy and intended `smpSynch`. The SCL is **configured intent**, not
+   automatically the deployed firmware identity.
+2. Connect a trusted independent TAP or SPAN that demonstrably retains VLAN
+   tags, **without trusting a diagnostic mirror as a canonical substitute**.
+   Capture timestamp source, precision, NIC/driver version and driver drops.
+3. With an ARStack `LAB_SOURCE` release-candidate profile, observe L2 PTPv2,
+   domain 0, Announce ~1/s, two-step Sync ~4/s, matching Follow_Up and peer
+   delay response **when requested**. clockClass 248 and clockAccuracy Unknown
+   do **not** constitute external UTC traceability; AUTO `smpSynch=0` remains
+   truthful for undisciplined source mode.
+4. With the same capture path, observe independent canonical SV and diagnostic
+   mirror identities and intentionally verify **different** APPID, destination
+   MAC, VLAN visibility and counters. Check `nofASDU`, channel layout,
+   instMag/q quality and `smpCnt` continuity for the selected profile.
+5. Test nominal **4000 and 4800 samples/s**, including appropriate
+   profile-validated counter rollover, TEST/bad-quality stimuli, stream restart
+   and multi-source isolation. A PCAP's host-arrival jitter is
+   screening only, not on-wire timestamp quality.
+6. Run mixed SV/PTP for at least **60 minutes** per acceptance scenario.
+   A 4000 sample/s stream produces ~14.4m SV frames in 60 minutes if
+   `nofASDU=1`; canonical plus mirror doubles the SV frames. Ensure
+   `--max-frames` covers all traffic, or split recordings into explicitly
+   indexed windows and inspect each.
+7. Record capture-driver counters alongside publisher firmware telemetry
+   (TX failure, missed slots and timestamp quality). A zero sequence-gap result
+   on the reader does **not** imply an independent NIC dropped zero frames.
+8. Independently verify receiving IED/subscriber behavior. A matched stream
+   in DigSubAnalyzer is not proof the IED subscribed or acted on samples.
+
+## Performance and UX acceptance
+
+While the live analyzer displays mixed traffic, record every 5 seconds:
+
+```powershell
+.\\scripts\\measure-m4-bench.ps1 -DurationMinutes 60 -IntervalSeconds 5 -OutputPath C:\\lab\\m4-local-metrics.csv
+```
+
+This is a local process CPU/working-set sampler, **not** a substitute for
+`dotnet-counters`, native Npcap driver statistics or hardware wire timestamps.
+
+
+- CPU utilization and process memory/working set;
+- .NET allocation rate and GC collections (`dotnet-counters` or equivalent);
+- NIC driver/interface drops where supported;
+- SV sample/counter anomalies, rejected frames and identity counts;
+- selection-to-coherent-snapshot latency and scope redraw responsiveness.
+
+Inspect the main workspace for these user-facing invariants:
+
+- changing selected SV publisher never carries waveform, RMS, phasor or
+  engineering scaling from the previous publisher;
+- TEST and invalid quality are labeled; invalid data never appears as
+  valid calibrated measurements;
+- canonical and mirror traffic remain distinct;
+- PTP observation, pairing, domain and freshness are not misrepresented as
+  servo LOCKED;
+- the one-line status remains calm, and detailed raw evidence lives under
+  Timing/Diagnostics/Advanced, not in an overwhelming initial screen.
+
+There is no synthetic CPU/heap reduction percentage quoted here. Measure
+**before/after** on the same hardware/NIC/driver/traffic scenario and record
+absolute values and drop counters. Without such measurements, resource
+optimization is implementation-level, not performance-certified.
+
+## Release gate and signed-off evidence
+
+The engineering reviewer should retain, locally or in an authorized evidence
+store: sanitizer-approved capture checksum, SCL version, firmware manifest,
+analyzer commit/build, report JSON, port/TAP diagram and VLAN settings,
+NIC/drop counters, 60-minute performance measurements and IED independent
+results. Publish only sanitized summaries.
+
+M4 cannot be called **physically complete** without this independently
+collected evidence. The automated replay suite is suitable for a software
+checkpoint and for detecting compatibility regressions.