Skip to content

ARSAS P8/P9 consolidated mainline RC — field-tested SG7 77/77, full integration CI pending - #768

Draft
masarray wants to merge 1685 commits into
mainfrom
fix/iec61850-targeted-static-rcb-partial-inventory-20261120
Draft

masarray wants to merge 1685 commits into
mainfrom
fix/iec61850-targeted-static-rcb-partial-inventory-20261120

Conversation

@masarray

@masarray masarray commented Nov 20, 2026 •

Copy link
Copy Markdown
Owner

CURRENT MAINLINE INTEGRATION CANDIDATE — 2026-11-21

One authoritative open ARSAS feature PR: P8→P9 reporting, SIEBCU first-connect correction, BCUNR inventory semantics, SG7 four-RCB diagnostics and explicit bounded Snapshot Read. Three old stacked draft PRs #763/#766/#767 were closed as consolidated without merge after verifying exact source ancestry; their branches and original commit SHAs remain auditable.

Current PR base: main. New combined HEAD: 423907349ccc5fe75f60d5af0ddddbde80a2f029, a two-parent, non-forced merge of accepted P9 field-trial source a3dca3e8483a7a8b1cb8047e7034db75380f8baa and latest main a0976a0b86dc42945e0d5d9d36724f1e541ef458. Full recursive Git-tree 3-way audit (not compare's 300-file cutoff): 63 main-side file changes, 8 path overlaps manually reconciled (7 one-line GitHub Action updates plus asset provenance), 55 main-only changes retained, 0 candidate-only changes lost. The exact candidate branch is now 0 behind main, 1,685 commits ahead and 569 changed files. This is a large integrated release candidate requiring full regression and review, not a cosmetic backlog cleanup.

Existing physical/operator field evidence — BEFORE mainline merge: exact ARSAS a3dca3e8 + ARIEC trial engine 24077e48e393ce3dc2cd1409db6e96f27aa1e368 achieved SG7 77/77 report-backed values, 4/4 real RCB ingress (129 InformationReports), 0 pending and 0 cyclic MMS polling. Original P9 Windows exact-head test 975/975 PASS, canonical 3051/3051, pinned engine 24/24, package 6/6, smoke and IO PASS. The merged ARIEC P9 staging source tree was independently verified byte-identical to the trial engine before bringing in current main maintenance. These are historical field + exact-SHA CI facts; they do not certify current combined HEAD.

Mandatory next gates: Run combined exact-HEAD 42390734 Windows full app/engine/ArdIrec tests, installer + portable smoke, provenance/media guard and workflow integrity. Reconcile engine promotion draft ARIEC #192, and the unresolved P0 command cancellation/Operate stall investigation ARIEC #172. Track SOE buffer-overflow/reconnect continuity separately in ARSAS #769. Preserve typed command/write behavior, bounded report queues, q/t, no hidden polling, no blind GI/PurgeBuf, and source provenance.

Publication stop: Public stable v1.6.42 already exists. Current candidate VERSION and .release/windows.json also read 1.6.42. Do not republish the same stable tag, run a stable release, or change production engine lock until an explicit next-version decision, combined CI, approved pin/FAT, signed packaging and operator acceptance. This PR stays DRAFT.


Historical original PR description (before consolidation)

P9 field-reliability: incomplete MMS inventory vs RCB absence

Evidence (sanitized)

  • Actual user Windows EXE used production engine f75e323e, not P9 experimental engine.
  • BCUNR MMS association succeeded and received 6 frames on 3/3 selected static RCBs, 7/7 selected live values; the user did not select all static members.
  • Smart discovery observed 35,698 names / 288 RCBs / 4 incomplete enumeration chains. Selected target reconciliation observed an explicit 64-page / 6,400-name limit.
  • Original 1.7-MB IID shows 39 DataSets, 35 ReportControl definitions, each configured with RptEnabled max=16. The user source is not committed. A missing RCB in partial live discovery is not proof the IED lacks one.

Code changes

  1. Source-aware RCB absence semantics: opened SCL with no matching ReportControl => NO RCB IN OPEN SCL; live discovery with no observed RCB => NO RCB OBSERVED IN LIVE DISCOVERY (UNVERIFIED). The latter must not mark a point definitively unavailable. Aggregate warnings with exact counts and representative references; never suppress real RCB, buffer overflow, or wire warnings.
  2. One bounded targeted recovery: only if selected exact static RCB segments remain unresolved and the first read-only reconciliation was Partial specifically because NamedVariable GetNameList reached 64 pages, repeat selected-domain, read-only reconciliation once at 192 pages. Keep P0 structural discovery fixed at 64 pages; no general full scan retry, no hidden cyclic MMS, no automated GI/RptEna mutation outside existing exact-target safeguards.
  3. Deterministic positive/negative regressions and a sanitized engineering assessment.

Stack/merge safety

Test expectation

Tests assert that incomplete live evidence cannot claim missing RCB, SCL evidence is correctly scoped, bounded recovery is enabled only for selected unresolved 64-page truncation, the structural Smart Discovery freeze remains, and no polling/ownership side effects are added.


SG7 field acceptance update — 2026-11-21 (trial-only)

Scope distinction: SG7 missing-values/RCB-ingress incident now has qualifying positive P9 field evidence; #769 owns separate continuity/reconnect concerns. This is a nonproduction test acceptance, not stable release qualification.

FAT Write UX: typed inline value editor and fixed Write action
…ng evidence

The canonical Windows guard must assert revised distinction between observed RCB instances, live DatSet binding and physical Report ingress.

masarray commented Nov 20, 2026 •

Copy link
Copy Markdown
Owner Author

BCUNR 2026-11-20 follow-up — evidence-backed fix and exact-SHA P9 Windows trial

This user-provided diagnostic is a field follow-up, not a crash or total MMS failure. The P9 58d22be4 trial showed 4/4 selected exact RCBs receiving 16 total Report frames, 125/125 selected points displayed (not 445/445 wire-proven), 2/2 positive enhanced-SBO control terminations and matching Open/Close feedback, 560 observed RCB instances, four initial BRCB BufOvfl flags, and no observed SqNum discontinuity on sequenced streams. The target address also appeared as a local test-PC alias; physical endpoint remains to be verified.

IID crosscheck: 39 DataSets, 35 RCB templates × 16 configured slots; 33 DataSets have a matching datSet binding in their LN0, six do not. The design target is not automatic 39/39 Report traffic.

Verified root cause of mapping undercoverage: initial app field log reported 35/39 no verified RCB-to-DataSet binding despite 560 observed RCB instances. P9 engine sampled only first 64 sorted BRCB instances and 16 numbered slots per family could exhaust this cap on four families. ARIEC #203 (exact SHA 24077e48e393ce3dc2cd1409db6e96f27aa1e368) changes read-only selection to fair round-robin by exact observed RCB family within unchanged 64 candidate budget. It never infers DatSet bindings or permits writes from family names. ARSAS log/UX source now differentiates observed RCB instances, nonempty verified mapping evidence and actual Report ingress, with concise indexed group messages.

Exact app HEAD: 2ab60b726600957311c23a2223d92541032ccf79 (Draft #768 stacked on #767); engine candidate: 24077e48; approved production pin f75e323e unchanged.

ARIEC #203: .NET CI #44720030989 succeeded on attempt 2 (first attempt unrelated timing-sensitive enhanced-SBO negative error test; separate flake audit remains).

ARSAS P9 trial Windows CI #44721168600: SUCCESS; 965/965 focused reporting regressions PASS, native bridge and portable EXE smoke PASS.

  • Download exact P9 trial EXE
  • SHA-256 78a17dbf6df0abc459dde15bf1361ee828ab7bd930770febe261ab45e84fd22d (uploaded artifact digest = smoke-tested EXE).
  • Canonical Windows pinned-engine build #44721168573 SUCCESS at the same exact ARSAS source: 3,041/3,041 full application tests, 24/24 pinned engine tests and 6/6 packaged smoke tests PASS; native bridge and portable EXE smoke PASS. The production pin remained unchanged.

Never automate control or PurgeBuf to mask BufOvfl. Next field gate: confirm actual endpoint hardware/simulator isolation, open original IID for identity/parity, record live mapped DataSet count versus previous four and the 33 configured design bindings, prove each selected RCB has independent Report ingress, inspect BufOvfl/EntryID, and recheck Toshiba regression. No production promotion or protected merge until qualified.

…streams

BCUNR 33-RCB trial received first Reports after the early 3-second monitor-click deadline. Anchor bounded diagnostic observation to exact completed activation timestamps; summarize 64 scoped RCB groups rather than first 16; keep event log concise; distinguish pending display token from missing wire value. No additional MMS Read, GI, RCB write or process polling. Add 33-RCB, stale/future timestamp regressions.

Copy link
Copy Markdown
Owner Author

BCUNR follow-up field validation — P9 33-RCB milestone (2026-11-20)

Exact tested EXE: ARSAS 2ab60b726600957311c23a2223d92541032ccf79, engine 24077e48e393ce3dc2cd1409db6e96f27aa1e368 (ARIEC #203). Do not commit the raw field diagnostic or interface inventory.

Observed progression (not a synthetic test claim)

  • Previous trial: 4/4 selected configured RCBs had independent ingress; 125/125 selected values visibly populated.
  • Current trial: 33/33 exact planned static RCBs eventually routed independently, 63 Report frames across 33 streams, 1,642 process updates, 410/411 selected runtime values displayed, 1 value pending (BCUNRLD0/LTMS1.TmSrc.stVal). 39 DataSets / 445 static members remain model inventory and must not be equated with 411 selected scalar runtime points.
  • Exact control CBAUTOCSWI1.Pos passed operator-directed Open and Close enhanced-SBO with positive CommandTermination and matching feedback (click-to-result ~569 / ~529 ms); do not alter the ~400ms SBO async application-error grace.
  • Buffer/SOE not qualified: 25 IED-advertised BufOvfl flags across 63 frames, 1 observed SqNum reset (15→0 with GI-reason on the boundary); these do not prove 25 lost events or lossless replay. Preserve all forensic evidence, never auto-PurgeBuf.
  • Open SCL parity remains not captured in this session. Local NIC alias included endpoint IP, so physical hardware identity is not independently proved.
  • Current discovery found 560 RCB instances, 62 with nonempty observed DatSet metadata, six of 39 DataSets without verified binding, consistent with the earlier IID's six unbound design DataSets. This is stronger coverage, not proof of every RCB's ownership or historical SOE.

Root causes for misleading operator diagnostics

  1. The old three-second report evidence timer started before serial activation of 33 RCBs. A WARN “no routed frames” fired shortly before the first Report; later 33/33 ingress proved it was a premature diagnostic. Follow-up now anchors post-activation observation to the latest exact completed RCB activation timestamp seen in the bounded window. No additional MMS I/O.
  2. Per-RCB Copy Diagnostic and UI triage previously took first 16 grouped RCBs only, resulting in shown=16, routed=12, noRoutedFrames=4 while the independent final per-stream tracker correctly counted 33/33. Copy Diagnostic now summarizes up to 64, including all 33, and labels shown/total.
  3. The one pending value carried Live / static report (semantic-structured-leaf) status and general-interrogation reason, yet was not visible. The new diagnostic includes its UI display token, not purported wire raw value. Root cause remains unknown until that evidence is captured; do not substitute a sibling/assumed enum or MMS polling.

Candidate change (no production promotion)

ARSAS PR #768 head 048e902ac81278313f40e86b7f59fa23271247d0: post-activation proof-window policy and tests; all-33 diagnostic summary; bounded Event Log details; pending display-token evidence. The engine #203 candidate remains unchanged. Do not merge until exact Windows CI and operator field verification gates are green.

… new activation deadline contract

Routed traffic triage correctly requires LastFrameReceivedAtUtc; fix synthetic proof fixture, and update pre-existing static FAT structural assertion to the new bounded post-activation observation API.

Copy link
Copy Markdown
Owner Author

P9 diagnostic field follow-up implementation and exact-SHA Windows trial — PASS

Updated PR HEAD: ab91f66b8e58c37b85709d3afab8b64fe6993049; ARIEC engine unchanged at 24077e48e393ce3dc2cd1409db6e96f27aa1e368; production pin f75e323e9f8443aa20e0a3818a1d5d77255b2c0f unchanged.

  • Exact-SHA nonpromotion Windows run #44728113201 SUCCESS: 968/968 focused regression tests PASS, native bridge and portable EXE smoke PASS, same SHA verified before and after artifact upload.
  • Download P9 updated EXE.
  • EXE SHA-256: 5b5ed4ab2d43af64893196f58f1b5e5abd29773b869bc216864863fba2d224ec.
  • Earlier unrelated fixture assertions were updated to match the exact received-frame timestamp and the new post-activation proof-window contract. No MMS/control protocol code was modified.
  • Still not field-qualified: LTMS1.TmSrc.stVal remains 1/411 pending pending a new displayToken capture; initial BRCB BufOvfl indications, one GI-adjacent SqNum reset, SCL-versus-live parity and endpoint identity require separate verification. Do not merge or promote before qualified operator evidence.

Canonical pinned-engine Windows build and IO test workflows for this exact HEAD are separate gates; verify their final conclusions independently.

Copy link
Copy Markdown
Owner Author

Final validation of exact P9 diagnostic candidate (2026-11-20)

Field acceptance still pending: new EXE must prove post-activation diagnostic now reflects all 33 RCBs; collect unresolved BCUNRLD0/LTMS1.TmSrc.stVal display token; do not dismiss 25 BufOvfl-indicated frames, 15→0 SqNum with GI, the absence of Open SCL parity, or local-host endpoint ambiguity. A diagnostic-only CI PASS is not proof of history-complete buffered SOE or production hardware interoperability.

… ingress

Only exact schema-resolved scalar Report HasValue with semantic-structured-leaf provenance can preserve literal Unknown; default placeholders, MMS Read, compatibility and stale association remain rejected. No additional MMS IO or RCB/control mutation.
…evidence

Evaluate value visibility once per point, require matching live Report reference+FC+association before showing semantic Unknown as populated; thread canonical read-only verdict through FAT/Copy Diagnostic/RCB triage. Add negative/cross-FC/offline/session regression and audit note; zero new MMS I/O.

Copy link
Copy Markdown
Owner Author

P9 source-level repair: canonical literal Unknown + association-safe completeness (2026-11-20)

Candidate (Draft PR #768, no production merge): exact ARSAS e0cf0aae409d4e52a8ce2035e08619fa31808fb6; experimental ARIEC engine remains 24077e48e393ce3dc2cd1409db6e96f27aa1e368 (PR #203); pinned production engine f75e323e9f8443aa20e0a3818a1d5d77255b2c0f unchanged.

Source-supported root cause and repair

  • Canonical IedBrowserLiveValues.TryPublish always rejected Unknown, even when exact schema-resolved semantic-structured-leaf delivered HasValue=true in an actual InformationReport. Initial-image diagnostic separately classified all display tokens Unknown as unreceived. BCUNR's one pending LTMS1.TmSrc.stVal had semantic-structured-leaf/GI process provenance; this is a viable source-level explanation for the mismatch, NOT independent physical proof of the enum value or source transport.
  • Report ingress now enables only exact semantic value Unknown, on a verified model-resolved FC and current association; generic string sentinel, MMS Read, compatibility, stale/cross-association, -, and Not read remain rejected.
  • Display initial-image and up-to-64-RCB triage now resolve literal Unknown via the same exact current-association canonical Report SSOT used by Explorer/Browser. One visibility decision is computed per point and reused for aggregate/group/pending totals. No new MMS IO, GI, RCB Write, control command or polling.
  • Project-owned regressions for positive and missing proof, wrong FC, read/compat rejection, stale reset, and unknown q (no fabricated Good).

CI and executable

Closure discipline — unresolved field gate

The original SG7_BCU recurrence (2026-10-28: 56/77 and 1/4 RCB) has NOT yet been replayed on this exact P9 engine/ARSAS pair. The newer BCUNR 33/33 stream trial is a different IED and cannot substitute physical SG7 acceptance. BRCB BufOvfl-marked frames, a GI-adjacent SqNum reset, raw TmSrc value identity, physical endpoint identification, and Open SCL parity still need field evidence. This source-level repair does not prove 77/77 SG7 recovery or complete SOE history.

Request: keep SG7 P0 issue #707 OPEN until original physical acceptance is demonstrated; do not close it as fixed or promote trial to production based solely on green CI. The PR is ready for exact-head trial, not unconditional production merger.

Copy link
Copy Markdown
Owner Author

P9 canonical verification completed — exact candidate e0cf0aa, 2026-11-20

  • Nonpromotion P9 consumer exact-SHA run #44734562650 SUCCESS: 971/971 reporting-focused tests; native ArdIrec bridge and portable Windows EXE smoke passed; artifact 14628002645; EXE SHA-256 a1655ddb3627cf1288df2df336631b9115742b137f5f58cb23621ecc8b2d1439.
  • Full canonical production-pinned Windows build #44734562671 SUCCESS: 3,047/3,047 ARSAS application tests, 24/24 pinned-engine tests, 6/6 packaged regression tests; native bridge and portable single EXE packaging completed.
  • IO List Testing #44734562645 SUCCESS; interoperability guard SUCCESS.
  • Exact candidate PR ARSAS P8/P9 consolidated mainline RC — field-tested SG7 77/77, full integration CI pending #768 e0cf0aae409d4e52a8ce2035e08619fa31808fb6 remains Draft and unmerged. Stable ARSAS main and production ARIEC lock unchanged.

Close decision: the source defect 'literal Unknown canonical Report omission / false pending' is code- and CI-remediated, with physical BCUNR validation pending. The older #707 SG7_BCU physical recurrence cannot be marked completed because no 77/77 + independent 4/4 RCB recovery on SG7 with this exact experimental candidate has been supplied. The 33/33 result came from BCUNR, a different IED; buffered SOE continuity with observed BufOvfl remains unproven. Keep #707 OPEN for this explicit field gate instead of falsely closing a P0 substation-reporting incident.

…riage

Display rows may overlap multiple configured RCB DataSets. Preserve independent activation-only RCBs with zero displayed attribution; never borrow member values or ingress proof. Add four-RCB regression.
…tic RCB reporting remains honest

Add one-shot exact-reference MMS initial-image reads through the existing per-IED IO ownership gate, with 15s/77-point bounds, association fencing, Stop cancellation and no invented quality, Report attribution or SOE. Preserve RCB traffic status; label Snapshot Read distinctly in Device Explorer. Add SG7 regression and engineering field notes.

Copy link
Copy Markdown
Owner Author

SG7_BCU field reproduction and tested P9 repair — 2026-11-20

Physical/operator-supplied trial diagnostic: exact nonproduction engine 24077e48e393ce3dc2cd1409db6e96f27aa1e368 and ARSAS reportingtrial. Static mode no cyclic MMS polling: 56/77 displayed, 21 pending, 2/4 RCBs ingress (13 Report frames, 2 BufOvfl advertised). Four individually accepted RptEna and GI writes are activation, not proof of initial Report arrival.

  • A01 / TT6DataSet3: 7 frames, 56/56 selected displayed, GI tagged.
  • B01 / TT3DataSet02: 6 frames (some subsequent control-related changes), but 3 unique selected value rows remain pending; no GI-tagged frame observed.
  • D01 / TT3DataSet01: 0 Report frames, overlaps metering members attributed by UI to I01. The former Copy Diagnostic omitted D01 because it grouped solely by the 77 point rows.
  • I01 / TT6DataSet4: 0 Report frames; 18 MX metering points pending.
  • Reporting endpoints' IPv4 also appeared on local tester NIC; physical remote IED identity needs separate verification. BufOvfl means buffered SOE continuity cannot be certified.

Exact new ARSAS head: a3dca3e8483a7a8b1cb8047e7034db75380f8baa on draft PR #768 (nonproduction only):

  1. StaticDataSetRcbImageTriage unions independent activation-only RCBs with the UI-attributed groups: now four SG7 RCBs can be diagnosed while D01 reports NOT ATTRIBUTED instead of fabricated 0/18 process values. No borrowed frame or quality truth. Regression constructs all four independent targets with required receive timestamps.
  2. Explicit operator-confirmed Read pending action in Device Explorer. One bounded pass (up to 77 exact ST/MX selected pending leaves, max 15s aggregate / 2s per attempt) through existing canonical MMS client, read-only; exact returned wire leaf and association fence required; Stop cancels and control cannot queue behind it; no GI/RCB writes, hidden polling, synthetic SOE or RCB-health promotion. Canonical MMS Read is displayed distinctly as Snapshot Read, and real Report remains authoritative. Existing ConsumeStaticReadJournal projects the value without duplicating a value source.
  3. Source and negative regressions for old RCB omission, wrong FC/non-pending candidates, alternative read leaf, provenance, and Report-over-Read priority.

CI exact-head: P9 Windows trial workflow #44741754523 SUCCESS, 975/975 tests, portable EXE smoke PASS. Artifact #14630533360 SHA-256 d1b7990aedef3a840d7cf608f058a1cc73ab689a99e4e35cf307add06c2c7558; experimental engine 24077e48, production pin unchanged. IO List #44741754435 SUCCESS. Canonical production-pinned Windows build #44741754423 still running when this comment was authored.

Closure gate still blocked: no proof that D01 and I01 began delivering InformationReports, or that B01 delivered three missing process values, or no BufOvfl / replay gap. A one-shot Snapshot Read can populate the visible initial image but must not count as 4/4 Report ingress or close P0 #707. Re-run exact candidate and compare 77 process sources plus per-RCB receipts; gather raw/header evidence for missing GI without blind re-arming. Keep PR draft and issue OPEN.

Copy link
Copy Markdown
Owner Author

Exact-head canonical Windows build has now completed (2026-11-20)

Build ARSAS run #44741754423 SUCCESS for same experimental PR head a3dca3e8483a7a8b1cb8047e7034db75380f8baa. Canonical Windows test gate: 3051/3051 ARSAS app tests, 24/24 engine tests on existing production pin, 6/6 package regressions, Windows portable single EXE smoke PASS. Canonical package artifact, EXE SHA-256 5151f437e26a7ca3e875e84ac54a5c59a1dc26ca74d52805d6aac8358b19d6e2. The P9 experimental executable for SG7 testing is a different, separately qualified artifact: P9 #14630533360 SHA d1b7990aedef3a840d7cf608f058a1cc73ab689a99e4e35cf307add06c2c7558; 975/975 reporting tests and portable smoke PASS. These artifacts differ in engine composition; do not substitute the pinned canonical package for P9 physical trial. No main/staging merge or production promotion; P0 #707 remains open until independent 4/4 Report ingress and 77 point provenance are verified on SG7.

@masarray
masarray changed the base branch from integration/p9-multircb-exact-windows-trial-20261105 to release/arsas-integrated-rc0-20261024 November 20, 2026 19:50
@masarray masarray changed the title P9 BCUNR: source-aware RCB absence + bounded targeted discovery recovery P8→P9 consolidated reporting RC: SG7 77/77, SIEBCU reliability, multi-RCB diagnostics Nov 20, 2026
Preserve original P9 operator-tested source and all 38 RC commits. Integrate 55 non-overlapping public-site, media and release-governance changes from main plus 7 exact action-version updates. Combine operator screenshot provenance and source-licensed Material vector provenance without overwriting either. 3-way path audit: 63 main deltas, 8 explicitly resolved overlapping paths, 0 candidate-only changes lost; merge remains draft awaiting exact combined Windows CI and IEC 61850 qualification. No stable release or engine repin.
@masarray
masarray changed the base branch from release/arsas-integrated-rc0-20261024 to main November 20, 2026 19:58
@masarray masarray changed the title P8→P9 consolidated reporting RC: SG7 77/77, SIEBCU reliability, multi-RCB diagnostics ARSAS P8/P9 consolidated mainline RC — field-tested SG7 77/77, full integration CI pending Nov 20, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant