Initial structural experiment attempts

These are source-structure results, not specialist or legal acceptance.

• baseline-initial-01: retained unchanged. The adapter treated missing per-unit legal-effect fields on already authored, complete-within-excerpt units as legal promotion. That diagnosis was wrong: the frozen catalogue explicitly says these boundaries are not legal or specialist acceptance. Do not use those provenance failures to judge the baseline.
• candidate-initial-01: retained unchanged. Four of four cases failed the overall gate. All four met passage and governing-heading checks, but navigation-role regions, some statute/memo references and the explicit source-instruction boundary were missing. No adapter exception occurred.
• baseline-initial-02: corrected adapter reads the frozen catalogue's explicit document-wide limitations, keeping excerpt completeness separate from legal resolution. Four of four cases still fail: governing headings, explicit scoped references and/or navigation-role regions are missing. The coherent source excerpts themselves are retained. No adapter exception occurred.

The correction changes no frozen source, fixture expectation or parser. Fourteen validator controls pass, including a new control separating declared excerpt completeness from legal acceptance. The first reports and their source/implementation hashes remain available for inspection. None of these first attempts qualifies for the eight-case expansion. Later attempts, including any passing initial gate and doubled run, have their own retained report directories.