frontend-patterns/research/plan/AuthorityProjectionComparison.md

Next bounded comparison — authority placement and projection

Prepared: 2026-09-27 · WP-0003 T01 · Status: completed bounded documentary comparison. Originally prepared before execution. LEARN-0015 was preregistered at 2df1d94; the completed review records six documents, six seed dispositions and two synthetic traces. Zero new drafts. The original method and limits below are retained for comparison.

Why this question

The classification inventory finds that seven of the original sixteen seeds have developed drafts. None of Server Authority, Local Authority, Replicated Authority, Server Projection, Client Projection or Streaming Projection has a dedicated developed draft. Current authority-family membership instead describes acceptance/precondition mechanisms. The language needs to explain placement decisions as well as recovery after those decisions.

Do these six seed names describe reusable patterns, context choices, implementation strategies or delivery mechanisms? Which independent decisions are conflated when we pair server authority with server rendering, or local authority with client rendering?

Bounded method

1. Register one prospective cycle before new evidence collection, with a new comparison group for this documentary/conceptual protocol. Record effort and source access limits; do not pool it with navigation runtime cycles.
2. Reuse existing Carniato/local-first and navigation/rendering reviews as prior interpretations, not new independent evidence. Inspect at most six primary-source documents across at least two independent source families. Pin versions or record access dates and claim locators. Sources may corroborate only parts of a contract.
3. For each of the six seeds, describe the relevant authority scope, acceptance boundary, replica/cache semantics, projection inputs, execution location and delivery timing. Keep optimistic display distinct from accepted authority and streaming delivery distinct from where projection is computed.
4. Construct a decision matrix for two synthetic briefs already represented by VC-01 transactional administration and VC-02 offline field notes. Vary projection placement while holding authority assumptions explicit. State incompatible or unsupported combinations rather than invent implementations. These are conceptual probes, not new validated walkthroughs or deployed known uses.
5. Record a disposition for each seed: develop a candidate, retain as a seed, treat as context/strategy, merge, or reject—with evidence and unresolved questions. Produce at most two new drafts, and only where context, competing forces, consequences and alternatives justify distinct patterns. Zero is a valid result.

Acceptance, limits and handoff

Deliver a claim table, the two decision traces and six seed dispositions. Identify at least one conflation or counterexample, or explicitly report that none was located in the bounded evidence. Review responsibility/force mappings against resulting candidate changes, then regenerate coverage and update evolution profiles/ledger at the normal review boundary. No automatic canonical promotion or gate closure.

Stop after six primary documents or 45 active research minutes, whichever occurs first; allow up to 10 additional verification minutes. Register these limits with the cycle before starting. Preserve unknowns instead of widening silently. Live inference, if needed later, uses llm-connect; this brief requires no model calls or new runtime fixture. It does not establish measured performance, production guarantees or human usefulness.

WP-0003 T01 retains the bounded Zulip caller/transport/test follow-up as an unresolved navigation applicability question. Broadening does not require declaring that question resolved; preserve its limitations and keep the authored navigation fixture paused. WP-0002 T03 retains independent seed/usefulness review. M1/M2 and WP-0004 prerequisites remain unchanged. Classification is a research aid, not a demand to fill every cell.