[AGiorgetti]

Summary

• Describe the current library mapping contract directly in the consumer skill; reserve older-version references for breaking-change migration.
• Fix standalone and central-package-management installation examples.
• Export the documented analyzer properties with CompilerVisibleProperty.
• Make the consumer skill self-contained, with factual NuGet install guidance and all document links resolving inside its own distributable folder.
• Move [P2] Use the captured direct source value for patch-mode nested updater calls #29 to Unreleased, add [P2] Preserve the selected nullable updater overload for configured patch source paths #47, and prominently document conditional source/runtime breaking changes and migration guidance.

Seven documentation/control files only. No library source, public API, specification semantics, package version, workflow, or protection-setting change.

Review corrections at c00821a6bb6504fcc1d621caf335f2aa9818c137

• The consumer skill describes current behavior directly. Earlier versions occur only in breaking-change migration.
• The complete skill is self-contained: both document links and their anchors resolve inside its installed folder; no external documentation, repository, specification or sample redirects remain.
• Final skill format, discovery, isolated installation, installed-file byte equality, content checks and CRLF-aware diff review passed.
• The exact dotnet add package Mammoth.LiteMapper command succeeded against the existing configured feed, resolved 3.0.0, and built/ran the extracted quickstart with zero warnings/errors and Id=42. This is a factual feed-resolution check, not a claim that another version is published.
• Focused usage/sample validation passed 1/1 with zero skips. Final migration-wording refinements then repeated the format/install/link/content checks; executable examples were unchanged.
• Prominent breaking-change notes explain conditional source/runtime effects and migration. The nullable enum-path repair remains under Fixed because a compiling 3.0.0 behavioral-break reproduction was not established.
• Reviewed/uploaded tree: 0329bf637032b04c39503f382e5f587db90b26c2. Library source, tests, build settings, workflows and benchmarks are unchanged from bd7b366.

Exact correction-head CI passed at c00821a6bb6504fcc1d621caf335f2aa9818c137 after one unchanged-job retry:

• Windows and Linux each passed 1,315 tests, zero failures/skips; both Release builds had zero warnings/errors.
• All six Roslyn jobs passed 1,034 tests each, zero failures/skips.
• Linux samples and five benchmark-smoke cases passed. Ordinary package tests include trimming and Native AOT.
• The first Windows/Roslyn 5.9 attempt failed one allocation-control assertion (208-byte overhead versus expected 200; 1,033 passed, zero skipped). Its single unchanged retry passed all 1,034 tests. This establishes intermittency across those executions, not the cause; the original failure is retained. No assertion, threshold, source, or runtime setting was changed.
• Pack & Validate, Verify Release Artifacts, and Publish remained skipped. This green ordinary CI does not replace the separate release rehearsal or final-version artifact checks.

Documentation validation

The initial documentation revision at bd7b366 passed the following checks; changed-skill revalidation is recorded below.

• Five isolated C# 9/net10.0 consumers extracted from the documented snippets restore, build with zero warnings/errors, and print Id=42 against published 3.0.0: standalone, central management, hidden/visible analyzer properties, and central management with visible properties.
• A versionless non-central negative control fails with NU1015 as expected.
• Generated source/configuration prove the exports enable both diagnostic comments and expose all three properties.
• Release packaging-test build has zero warnings/errors.
• Milestone16UsageDocumentationTests: 1 passed, zero failed/skipped, including Basic, Collections, and ASP.NET Core smoke samples.
• Skill format validation, local discovery, isolated Codex installation including the supporting reference, documented remote URL discovery, local/pinned links, original line-ending preservation, and CRLF-aware diff checks pass.
• Initial reviewed tree fa6e5ab62475eee670ec567cd84b8eb4e6a8b8a8 matched the first uploaded commit.

Initial-head CI passed for bd7b366509072617d2785686e021b9e9a34f0ec1:

• Windows and Linux each passed 1,315 tests (1,034 generator, 275 packaging, five runtime, one integration), zero failures/skips.
• All six Roslyn 4.8/4.14/5.9 host jobs passed 1,034 tests each, zero failures/skips.
• Both Release builds had zero warnings/errors; package tests include trimming and Native AOT.
• Linux sample and five-case benchmark smoke passed.
• The three release jobs were skipped as expected on ordinary PR CI; this does not substitute for the manual release rehearsal.

Prior-head local artifact rehearsal

A local provisional 3.1.0 artifact rehearsal was attempted from bd7b366509072617d2785686e021b9e9a34f0ec1. This is a packaging assumption, not an approved release version or a publication.

Passed:

• Exact six-file package/symbol set, package version and repository SHA, MIT package licensing, shipping TFMs, analyzer layout and dependency isolation.
• Portable-PDB SourceLink and source paths, checksum manifest and independently copied provenance/artifact handoff.
• APICompat package checks and comparisons against actual published 3.0.0.
• Independent clean-worktree rebuild: all six normalized package payloads identical. ZIP envelope bytes differ, as permitted by the existing payload-comparison policy.
• All 12 framework/language consumers (netstandard2.0, .NET 8/9/10; C# 9/14/latest), independent abstractions-only use, and all three unchanged source samples.
• Restored package-file hashes across the 16 clean consumers match the validated local artifacts.

Local trimmed and Native AOT publication both failed at the known MSBuild task-host creation limitation (MSB4216 / MSB4027, ComputeManagedAssemblies). Neither local executable ran. One supported tool-level escalation retry passed explicit restore but failed with the same trimming task-host errors. Ordinary exact-head CI passed its own trimming/AOT coverage, but does not prove those provisional-version release artifacts. One recovered checker error and one interrupted/resumed restore are retained in the evidence; this is not represented as a single uninterrupted all-green invocation.

Performance evidence remains separate

All measurements below target bd7b366 versus actual published 3.0.0; the subsequent correction changes documentation only.

• Cloud: 32 scenarios, 128 executions, no allocation increases. Shallow tracked recursion lost 19.9760% / 23.0921% throughput, with significant drift in some unchanged controls. This did not establish controlled acceptance.
• Windows full default-runtime suite: baseline/candidate/candidate/baseline across 32 scenarios; the actual repository acceptance script returned 2/1/1/2. Shallow tracked depth-1 losses were 11.76–20.08%. An untracked depth-16 launch split remained unexplained despite matching recorded settings.
• Bounded six-case supplemental Windows comparison with child-process tiered compilation and tiered PGO disabled: all four actual gate results were 2 (review), tracked depth-1 losses 15.67–19.88%, allocation 232 bytes, stable controls, and no observed untracked split. This is neither a full default-runtime pass nor proof of the original split's cause.

No measured allocation increase occurred. Overall performance acceptance is not established; no further benchmarks or threshold changes are part of this documentation correction. The repository script has now run on Windows. Cloud PowerShell remains uninstalled.

Remaining release gates

• Resolve performance review separately using the completed cloud and Windows evidence; a full default-runtime acceptance pass has not been established.
• Complete the safe GitHub workflow rehearsal on the final PR branch head with publish=false, dry_run=true; dispatch is awaiting secure GitHub sign-in. The workflow has no final-version input.
• Select the release version/ref deliberately, update release-facing version guidance when appropriate, and validate the exact final-version artifacts.
• Satisfy the release safeguards. Read-only GitHub responses rechecked on 2026-10-12 at 13:52 UTC returned production protection_rules: [] and no repository/inherited rulesets. No protection or credential setting was changed.

The requested documentation corrections are ready for human review. Overall release readiness is not established. No merge, release, tag or NuGet publication was performed.