FrameCode-VibeWork reconciliation

Type: agentic-system-analyses/types/agentic-system-reconciliation-report.md

Reconciliation

The reports share the frozen boundary: FCVW supplies governance instructions, retained project knowledge and optional local tools, not the enclosing host executor. Runtime supplies the invocation baseline; memory traces retained artifacts and later consumption; epistemic separates evidence production, acceptance, integration and operational permission; workflow traces one authorized package-to-closeout protocol. This reconciliation uses the pinned members only. No source ambiguity required a new source read. No correction-answer files were supplied; the current findings, rather than the previous reconciliation's descriptions of them, determine the connections below.

Duplicate identities

Amendment: EPI-OBJ-wiki-content is superseded by MEM-OBJ-wiki. Both identify the natural-language bodies of versioned wiki notes, including reusable claims, raw imports and session content interpreted by hosts. Their SRC-1 FCVW/wiki/README.md passages distinguish knowledge pages from raw preservation. Epistemic inventory and wiki-route findings refer to this same content family, not another store. Identity preserves their different analytical fields and uncertainty about particular claim instances; it does not establish acceptance or semantic preservation.

Amendment: EPI-OBJ-wiki-metadata is superseded by MEM-OBJ-wiki-metadata. Both denote symbolic wiki YAML lifecycle, provenance and relation fields consumed by checks, graphs and indexing. SRC-1 FCVW/wiki/README.md, tools/validate_fcvw.py and the source/graph findings anchor that identity. Evidence-method checks and source-digest/access assessments concern the same metadata surface, not independent warrants for prose truth.

Amendment: EPI-OBJ-product-contract is superseded by MEM-OBJ-product-contract. Both identify product surface pages containing sourced expectations, stable IDs and case tables with a symbolic selected-contract hash consumer. SRC-1 FCVW/skills/QA/PRODUCT_WIKI.md and tools/qa_wiki_fcvw.py supply the identity evidence. Live-QA, contract-checking and reuse findings concern the same expected-behavior input. Approval provenance and behavior outside the encoded contract remain unresolved by hashing.

Amendment: EPI-OBJ-qa-run is superseded by MEM-OBJ-qa-run. Both identify retained application Markdown audits containing observations, evidence references, results, contract hashes and divergence decisions. Their SRC-1 FCVW/skills/QA/PRODUCT_WIKI.md and tools/qa_wiki_fcvw.py findings agree on the artifact and declared-only provenance limit. Epistemic acquisition/check/decision rows and memory exact-run reuse findings concern that one artifact family. Neither declaration authenticates execution or consent. Both IDs remain declared, as do all superseded IDs above.

Containers, parts and distinct routes

RT-OBJ-skill and its declared part RT-OBJ-skill-frontmatter carry procedural guidance and skill-routing metadata. They are not wiki knowledge bodies or wiki lifecycle metadata merely because they share Markdown storage. MEM-OBJ-session specializes the wiki content family by its continuity consumer, synthesis inputs and verification on resume. The wiki container already distinguishes sessions and product-specific content; its findings do not become wrong when those operative parts are assessed separately. No container supersession is warranted.

MEM-RTE-index-build produces MEM-OBJ-context-index, which RT-RTE-context-delivery can load on later queries. Production and query-to-return are not identical routes. MEM-OBJ-knowledge-graph supplies optional symbolic relationships, not the index's copied excerpt text. MEM-OBJ-decision-trace supplies content-free tool evidence, distinct from canonical knowledge and the reports/logs of RT-RTE-local-checks. No supersession joins these objects to runtime routes or each other.

MEM-RTE-source-review is explicitly part of RT-RTE-asset-admission. Its progression combines deterministic candidate production with subsequent host curation. EPI-RTE-source-review isolates digest, direct-dependent and review-date calculations. It is narrower than the combined memory route, not identical to its complete progression. EPI-RTE-wiki-review and EPI-RTE-wiki-integrate separately declare admission/revision and retention/integration parts of the runtime asset gate. Their overlap with source-triggered curation does not establish whole-route identity: they also cover other promotion triggers and distinguish consequential phases. The containers remain valid. Wired warning production and claimed host revision retain their own statuses.

MEM-RTE-session-handoff has a distinct synthesis, rotation and reload progression despite the asset gate's optional session-synthesis output. It is not identical to general asset admission or wiki promotion. This disposes of memory's continuity-route overlap question without amendment.

MEM-RTE-product-review spans discovery, authoring, replay, run reuse, structural checking and divergence admission. EPI-RTE-live-qa, EPI-RTE-qa-check and EPI-RTE-divergence-decision distinguish application observation, selected-representation checking and user direction within that broader account. None alone denotes the entire memory route. The valid container remains; these phase distinctions require no split supersession. Application QA also differs from the framework-local verification driven by RT-RTE-local-checks.

EPI-RTE-evidence-check is a validator function shared by multiple dispatch consumers, not identical to the MCP, hook or local-check wrapper. EPI-RTE-retrieval-benchmark is a declared part of the local-check driver and also has a separate CLI; its labeled retrieval evaluator does not replace the driver. EPI-CLM-sourced-knowledge, EPI-CLM-product-evidence and RT-CLM-context-controls name different claims, not duplicate identities.

Read-back issue dispositions

Memory's first integration issue still describes an earlier runtime finding: that no-write MCP handlers make later read-back inapplicable. The supplied RT-RTE-mcp-checks now explicitly identifies section-body read-back, outline-only requests, active-plan references, queue identities/metadata/recommendation and digest-only output. These findings agree with memory's MCP annotation, including its SRC-1 tools/mcp_server_fcvw.py return-content evidence and queue derivation in tools/plan_queue_fcvw.py. Thus the integration issue's description of what runtime says is stale; its requested distinction is present in the current runtime record. The earlier cross-report disagreement is no longer retained. This does not turn references, outlines or digests into claim-body delivery, or establish actual host reliance.

The same disposition applies to RT-RTE-hook-lifecycle and memory's hook annotation. The current runtime identifies narrow retained plan-identity read-back: all sorted in-progress names and up to five sorted pending names supplied as additionalContext. Memory retains the corresponding SRC-1 tools/hook_fcvw.py selection passage. Both distinguish this push from handoff-body reading, memory authorship and observed host use. Memory's integration issue still attributes the former inapplicability finding to runtime, but the supplied record no longer says that. The previous reconciliation's conflict therefore does not describe these inputs.

EPI-RTE-evidence-check and EPI-RTE-source-review likewise explicitly describe later consumption of persisted provenance/status, digests, dependency relations and review dates. Their lack of writes does not erase these consumers. They do not claim body delivery, semantic acceptance or automatic report retention. This is consistent with memory's separation of content and access metadata; it is not a new route identity or a stronger operation status.

Convergence and remaining limits

Runtime's RT-CLM-context-controls, memory's context-delivery annotation and epistemic's EPI-RTE-evidence-check distinguish independent consumers: builder metadata can override defaults, while the validator can flag incompatible authority. No mandatory validation-before-index chain is established. The check qualifies any inference that authority spoofing is never checked, not the narrower finding that ranking itself is not fail-closed. Stale copies, lifecycle ranking penalties, sliced excerpts and conditionally reachable rationale remain delivery limits, not evidence of universal withdrawal or a host-window bound.

The specialized every-divergence user gate in MEM-RTE-product-review and EPI-RTE-divergence-decision coexists with general curation's lack of universal per-content human approval. Their scopes differ. The checker can reject a declared run or withhold command success without authenticating consent; clearing a decision gate does not make a failed test pass. No conflict requires choosing a stronger gate for all content.

MEM-BAP-knowledge-consumption and RT-BAP-policy-consumption have distinct referents and force: evidence/continuity guidance versus interpreted policy/skill instructions. A note feeding a later skill update does not directly acquire the skill's downstream authority. No supersession joins these paths.

Workflow declares no new records. Its package annotation and change annotation extend the runtime routes with rejection, bounded bypass, recovery and delivery criteria. Main-agent review rather than worker success or plan status controls prescribed closeout. Specific R5 approval and approved irreversibility qualify the general flow, not contradict it. These annotations establish neither an executable scheduler nor an exercised repair loop.

Memory's remaining initialization/provenance, semantic-review runtime, host-context and rebuild-scheduling questions remain local limits in its integration issues. The epistemic bounded conclusion, runtime's RT-RTE-asset-admission and workflow's recovery account independently distinguish evidence-responsive revision procedures from exercised improvement. Faithful uncertainty about actual acceptance, compliance, learning benefit and causal self-improvement is not disagreement and does not erase the supported procedures or wired checks. No remaining substantive cross-report conflict is identified in these pinned findings beyond the stale integration-issue descriptions disposed of above.