Matching the best-artifact task

This overview explains how C21 compares seven methods under a common best-artifact contract and C22 freezes a guarded source-packed path. A valid factorization is not enough when the task requires the exhaustive-best artifact and complete charged work. Suppose two artifacts reconstruct the same target matrix exactly. Both satisfy a contract asking for any valid reconstruction. A stronger contract may require the unique winner under a frozen canonical ranking. For a simple illustration, rank by component count. A one-component artifact wins over a three-component artifact, even if both are exact. This toy field explains the distinction; the retained experiment uses its own complete frozen canonical ranking, which must be applied unchanged. Returning the larger artifact passes reconstruction while failing this stronger task. Correctness therefore includes more than a truth-vector match when the requested output specifies a particular best artifact. Compare methods under the same complete requirement. C21's seven-method table requires the exhaustive-best GF(2) artifact under a common bounded task, rather than any valid factorization. Construction, proposal, completion, checks, cleanup and wrappers belong to the charged request. Imagine two candidate artifacts that both reconstruct a truth vector. The frozen ranking selects the smaller canonical score. Returning the other valid artifact satisfies reconstruction but misses the best-artifact contract. In C21, source-packed construction is also distinct from proposal success: the representation path can run even when its cut proposal abstains and exact screened completion chooses the answer. The retained narrow source-packed advantage leaves little routing headroom. C22 freezes an opt-in source-packed screened path with exhaustive restoration, refusal fallback and shadow identity checks. That is implementation readiness, not fresh timing evidence or production promotion. A source-packed representation can be constructed even when a later cut proposal abstains. Construction and proposal success are different stages. A refusal to propose does not imply that the representation was never built. The route can continue with exact screened completion, choose the required artifact, verify it, and deliver the answer. All these stages, including the unsuccessful proposal work, belong to the charged request when the caller pays for them. An opt-in shadow path can record candidate behavior while preserving the control answer and checking restoration and identity. That establishes operational checks within its scope. It is not a substitute for fresh timing evidence or permission to change a production default. Separate construction, proposal, completion, and checking costs. Implementation readiness does not replace fresh timing evidence or production promotion.