[masarray]

P0-B exact DataSet scoped bounded RCB reconciliation

Engine HEAD: adffc518305926e1aa5b4829ab4bb37ff7a80e1e, stacked on P0-A engine PR #177. Windows .NET CI exact SHA SUCCESS: https://github.com/masarray/ARIEC61850/actions/runs/39238764969

Behavioral contract

• Removes global pre-read abort for 66/300+ discovered RCBs. Selects exact live DatSet-bound candidates in one scoped index first, then exact configured references as read priority only, then bounded unknown candidates. Only real live DatSet equality enters operational inventory.
• Read-only paged interrogation with max reads/page, bounded pages/invocation, explicit Partial, examined/deferred counts and resumable continuation token.
• Continuation is SHA-256 bound to target/catalog plus accepted MMS association nonce; reconnect, changed inventory or wrong target invalidate the cursor.
• Wrong DatSet, wrong FC, unrelated LD excluded; no discovery writes and no reservation state changes. Operational availability / foreign reservation gating retains existing authoritative policy.

Synthetic regression evidence

• 66 and 300 catalogs with only two proven target RCBs: 2 candidates selected; remaining candidates skipped when existing live binding is established.
• 300 unknown RCB: five bounded pages at 64 candidates/page, no truncation or duplicate.
• Busy/foreign RCB rejected by eligibility policy and independent exact-free RCB eligible.
• Stale catalog or association epoch cursor rejected; exact configured name without DatSet is not binding evidence.

Limits: Synthetic selection-count evidence is not an on-wire MMS request-count measurement. Full ARSAS consumer Windows CI in linked PR #626; physical field acceptance is separate and remains unverified. Keep draft.