frontend-patterns/research/reviews/M1-SeedReview.md

M1 seed review — 0.1 packaging boundary

Reviewed 2026-09-27 by Codex, author-model. Candidate baseline 8cc7016. M1 result: revise/defer; not passed. The connected subset is suitable for a qualified research reference preview, not canonical admission or proven usefulness. Publication 0.1 and PatternSchema 0.1 are different version axes.

Gate assessment

M1 criterion	Outcome	Basis
Reconciled metadata contract	pass	WP-0002 T01, vocabulary and worked contract fixtures; candidates are not full canonical-schema records
First source and local-first contrast	pass, scoped	Source readiness; source identity and uncertainty explicit
6–10 connected drafts	pass for editorial subset	CAND-0001–0008 cover local persistence/propagation/divergence and shared-record editing/recovery; selection is not a global catalog cap
Two completed validated walkthroughs	revise	VC-01/02 include decisions, baselines, changed constraints and failure traces, but remain synthetic with unverified implementation assumptions
Evidence/counterexamples	pass for traceability; defer for recurrence	Documentation, pinned code and bounded runtime mechanisms are separate evidence kinds; full-contract independence unassessed
Independent human validation and usefulness	defer	Frozen pilot 18/24 in each condition after model-review adjudication; no naming benefit, equivalence or decision-time result established
Quality and recovery guarantees	revise	Requirements for retention, keyboard recovery, receipt windows and crash boundaries are stated; complete validation absent

Core editorial inspection

All eight records were read in full. Each has context, competing forces, a proposed solution, consequences, alternatives, evidence and explicit gaps. “Pass” below means usable as qualified hypothesis prose; it does not mean empirical correctness or compliance with every canonical template field. Reference manifest selects identities; candidate prose remains the contract source.

Candidate	Context and solution / resulting decision	Trade-off and alternative checked	Evidence limit / preview disposition
CAND-0001 Conflict Copy	Preserve detected divergent content separately; resolution remains pending	Recoverability versus clutter/review; immediate review or domain-aware merge instead	Joplin and optional Obsidian mechanisms, not atomic preservation proof; pass as hypothesis
CAND-0002 Local Commit Boundary	Acknowledge a revision and pending intent after a recoverable local commit	Offline recovery versus transaction/feedback cost; memory-only or server-required saving cannot satisfy the fixed brief	Storage motivation, no verified crash/flush contract; pass as hypothesis
CAND-0003 Pending Synchronization	Retain queued work and distinguish sent from acknowledged	Offline usability versus freshness/retry complexity; online-only writes are a different requirement	Joplin progress metadata, not proposed retry/supersession proof; pass as hypothesis
CAND-0004 Explicit Reconciliation	Compare preserved alternatives and commit a new decision	Information completeness versus interruption; justified automatic merge is an alternative	Manual comparison evidence, not provenance/atomicity/AT validation; pass, retain separately provisionally
CAND-0005 Local Draft State	Keep base, editable proposal and refreshed observation distinct	Recovery versus state complexity/privacy; direct binding only when no unfinished intent	Partial Zulip/Element and authored reload mechanisms, not whole lifecycle; pass as hypothesis
CAND-0006 Version-Checked Mutation	Check the proposal's expected version at the authority	Consistency versus repeated review; unconditional/commutative operations need their own contract	Conditional API mechanism, not complete frontend implementation; pass as hypothesis
CAND-0007 Mutation Outcome Tracking	Track operation-specific evidence rather than infer rejection from transport failure	Recoverability versus retention/backend cost; lookup or reconciliation when replay cannot be justified	Bounded API/SDK/source mechanisms, not universal deduplication; pass as hypothesis
CAND-0008 Invalidate and Reproject	Refresh affected observations after acceptance without losing drafts	Freshness versus network/invalidation work; complete authoritative response or subscription may avoid refetch	Framework documentation, no full frontend recurrence; pass as hypothesis

The preview links provisional relations with conditions and rationales, not mandatory architecture dependencies. A request for globally exclusive offline approval cannot be satisfied by combining these names. Backup, authorization and crash guarantees remain explicit implementation obligations.

Retain / merge / defer

Preview ontology review retains Conflict Copy and Explicit Reconciliation as distinct retrieval decisions, conditionally connected. Their ultimate split/merge remains testable. Mutation Outcome Tracking is retained as a shared review model/hypothesis; no new claim of standalone recurrence. The nine extension candidates remain available with their existing limitations and are not needed for core closure. No candidate promotion, ID rename or evidence-profile upgrade occurs.

Residual owners

• WP-0002 T03: independent human review, usefulness and fresh-scenario replication; this task stays progress and the M1 gate stays open.
• WP-0003 T04: storage atomicity, receiver replay/ancestry and recovery validation transferred from the completed source-review deliverable.
• WP-0003 T01/T02/T03: independent application contrasts, quality/AT and wider cases, and later ontology/M2 judgment.
• WP-0004 T01/T02/T03: canonical graph, held-out retrieval and empirical applicability retain their prerequisites. Preview candidate export is not M3 completion.

These are explicit work records, not a claim that the remaining work was completed by packaging the reference. The frozen pilot and original investigation receipts are unchanged. Review this gate again when those tasks deliver new evidence.