Typed Alignment and Audit Signatures

This paper defines the realization-relative comparison layer of the Sectorized Observable Framework (SOF). Two versioned SOF Reports are input evidence, not yet a comparison object: record kinds, capabilities, carriers, sectors, observables, thresholds, depth conventions, parameter samples, and comparison policies must first be admitted and explicitly aligned. The paper defines the SOF Report Alignment datum from a reference report, a target report, and declared sector and observable maps. A comparison specification fixes normalization, threshold, depth, path, metric, weighting, and aggregation semantics. Together they form the comparison object. A versioned Audit Profile then selects a sparse typed Audit Signature whose coordinates remain aligned, mismatched, unavailable, incomparable, or unresolved rather than being coerced into one total numerical vector. The No Comparison Without Alignment proposition makes explicit alignment evidence and comparison semantics part of the domain of definition. Alignment-Relative Audit Faithfulness requires each emitted coordinate to retain its report-item bindings, carrier, policy, evidence, and provenance qualifications or to record an explicit negative state. A nonzero mismatch records an alignment-relative difference; it does not by itself establish defect, correctness, causal failure, or a warranted action. On a fixed retained comparison frame, the admitted structural sub-signature carries a weighted Hamming pseudometric, with the weights included in the comparison specification. The paper keeps operator/word and Lie/Hall channels independently typed: an exact support inclusion relates simple Lie bridges to ordered length-two word supports, while a finite witness shows that word accessibility need not imply Lie accessibility. No metric, pushforward, transport, connection, or curvature is claimed across comparison frames. The source-addressed v2 corpus contains 20 F1-F4 source payloads, five F5 compatibility payloads, and three transformation controls, giving 28 migrated legacy/control audits, plus one separate native GridWorld F4 factual audit. The migration preserves unavailable states and does not recertify legacy comparison values. The native Object Certificate independently recomputes selected support, word, and Lie coordinates from frozen source matrices and records factual states ALIGNED, ALIGNED, and MISMATCH with mismatch counts 0, 0, and 8. A separately promoted finite AB/BC/AC control establishes only candidate-relative gluing and nonuniqueness results, not a general descent or sheaf theorem. The results do not make report pairs comparable without alignment, treat a declared reference as truth, turn unavailable coordinates into zero, promote diagnostic analogues into strict carriers, infer defect from difference, define cross-frame transport, or prescribe an interpretation or action. Version 2.0 replaces the version-1 fixed eight-coordinate comparison presentation with capability-aware Audit Profiles and sparse typed coordinate maps. It separates the alignment datum, comparison specification, Audit Profile, Audit Signature, claim target, and evidence basis; introduces source-addressed profile and coordinate-semantics registries, digest-closed validation receipts, and a migration/native evidence split; narrows the pseudometric to one fixed retained frame; and keeps presheaf, descent, and cross-frame geometry as an explicitly non-evidentiary research program. Source repository: https://github.com/dooven-prime/rime-lite Paper XIII of the Representation-Induced Mechanics and Evolution (RIME) program. Independent researcher; no institutional affiliation.