[masarray]

Baseline (immutable stack)

Stacks on #644 at 80e4836affce0b1f5f142c2966199a2ef5af41b3 (itself stacked on combined #642). Engine remains ARIEC61850 #183 c85d4e20db722bd0c2729f2533b323b13d3ef735. No engine repin or main integration.

Root cause / reproduction

IndexPointReference previously used TryAdd for both normalized and canonical aliases. Two different selected points could share a canonical alias; whichever was registered first silently won. The rare Report fallback used Points.Values.FirstOrDefault for ancestor/descendant prefixes, permitting MMXU1.A to route as whichever phsA/phsB child happened to be enumerated first. It then overwrote the alias cache, making the misroute persistent.

Change

• Retain one existing session-scoped ReportReferenceIndex; record its ambiguous keys in a bounded same-session set. Collision removes the alias, forbids future re-insertion until reset.
• Replace first-match prefix selection by exact canonical leaf equality only; sibling phases and even a single selected child of a three-phase parent fail closed. Typed engine schema projection must emit the exact leaf to authorize routing.
• Never overwrite a pre-existing alias during fallback, and clear ambiguity quarantine with the session association.
• Preserve exact point routing, read-only report ingestion, and all existing RCB/command/data ledger authority. No MMS reads or writes, no polling, no global scans on successful indexed hot path.

Regression

Five positive/negative cases: parent/child ambiguity independent of ordering, single selected child rejected for parent aggregate, exact FC-encoded aliases accepted, segment-boundary prefix safety, canonical collision quarantine, and stable same-point cache.

Gate

New code/test committed; exact-head Windows CI and physical report member/FC/schema FAT still required. This prevents a known class of wrong point routing, but does not claim analogue Unknown is solved or that schema-proven per-member projection is complete. No release or merge without matched on-wire and UI evidence.

[masarray]

2026-10-20 exact-SHA Windows qualification checkpoint (code-only)

• Application PR P0-H: fail-closed canonical Report alias collisions and sibling-phase fallback #645 HEAD: 2b4e4ea0b917b6c10fbe4b64b6a3a7b127870598 (stacked P0-F: fail-closed Browser Read admission against active MMS model #644 -> P0-FAT: combined Command Dock + Browser leaf + IED-card progress trial #642; no main merge).
• Engine lock read from same HEAD: ARIEC61850 PR fix: recover offline static DataSet members in Signal Selection #183 c85d4e20db722bd0c2729f2533b323b13d3ef735 (no engine repin).
• GitHub Actions: all 10/10 commit-associated workflows completed SUCCESS; canonical Windows Build run 39410426905 SUCCESS.
• Windows runner logs: ARSAS 2,418 passed / 0 failed / 0 skipped; targeted pinned ARIEC 24/24; packaged managed 6/6; portable single-EXE smoke successful.
• Portable FAT code trial artifact #12259286963: inner EXE SHA-256 bd3e74ad9d54b4e3abb054cd3e2721c1801b06ccfd0975a73021cfa6eb130de7; artifact ZIP digest sha256:8f9933a8df3a0ebe92e02f819486dba6f3cdeb9d3468c0d0aab8b66ef2e57cd5. Test evidence #12258379256.
• Five new Report alias regressions prevent first-phase/parent inference, including when only one phase is selected. Exact canonical leaf authority required; this prevents wrong projection, not an assertion that all previous Unknown points now read.
• NOT ACCEPTED: no matching physical relay read of SV/OR/BL; no same-SHA DataSet member/RCB report proof, stable live latency, WPF rendered DPI/geometry, physical command sequencing, or exact Inno installer qualification. No automatic command/RCB changes, no main merge, no public release. Independent Smart Discovery Field Capture workflow should not substitute for the above exact engine-locked portable.