Record the remaining context consolidation gap

docs/roadmap/remediation.md

@@ -0,0 +1,77 @@
+# Consolidation Review Findings
+
+Reviewed revision: `557e006a4ab98b666735a71b3845418c702f7cec`.
+Comparison baseline: `8ab1568` (the committed implementation plan).
+
+The review compared the [consolidation target](consolidation.md) and
+[implementation plan](implementation.md) with the completed sprint, including
+public/private conversions, preflight and acquisition, import selection and
+evidence, filesystem contexts, inheritance, projection, schema construction,
+changed tests, and documentation. The development check (`./scripts/check.sh`)
+and full race suite (`go test -race ./...`) passed with Go 1.26.10 on Linux.
+
+## 1. Complete The Non-Import Resolved-Context Handoff
+
+Status: open. Priority: medium. Category: implementation gap and maintainability.
+No user-facing correctness regression has been established for this finding.
+
+**Target.** Roadmap internal target 2 and implementation Stages 7–9 select one
+resolved filesystem description shared by validation/acquisition and passive
+context construction. Stage 8 explicitly requires populating the description
+at the existing non-import validation point and resolving only facts that are
+still missing later.
+
+**Evidence.** In [acquisition.go](../../internal/engine/acquisition.go),
+`acquireInput` calls `validateFilesystemContext` at line 181 for inputs without
+import traversal and with an applicable explicit context. That validator returns
+only a diagnostic. Its host/native branches derive a resolved description but
+discard it. At lines 220–226, acquisition sees `input.resolvedFilesystem == nil`
+and calls `resolveFilesystemContext` again before building passive views.
+
+For example, an ordinary `Load` of a byte or mapping source with explicit host
+`FilesystemContext` and imports disabled resolves/canonicalizes the base and
+root during validation, then derives those same facts again for its source
+context. The retained description therefore does not complete the intended
+validation-to-context handoff.
+
+Additionally, [filesystem.go](../../internal/engine/filesystem.go) still contains
+`validateAbsentHostFilesystemContext` (line 40), with its own root/base
+canonicalization, root-defaulting, and containment logic. That derivation
+duplicates `resolveAbsentFilesystemContext` in
+[filesystem_context.go](../../internal/engine/filesystem_context.go) (line 29).
+The two paths already maintain separate containment-diagnostic messages. Graph
+caller inspection confirms that the old validator is reachable from
+`acquireInput`; it is not unused migration scaffolding.
+
+**Why address it.** The new model exists, but one important path still discards
+and reconstructs its results, while absent-host context derivation has two
+production implementations. This leaves part of audit finding 5 unresolved and
+preserves the drift risk the consolidation was intended to remove. This is a
+structural completion issue, not a claim of measured performance loss or a new
+single-lookup guarantee.
+
+**Required revision.**
+
+- Have non-import context resolution/validation return or populate the resolved
+  description, and retain it on `assembledInput` at the existing validation
+  point. Later passive-view construction and ordinary-overlay handling should
+  reuse those applicable facts.
+- Route absent-host derivation through the common resolved-context owner and
+  remove the duplicate derivation from the old validator. Keep the existing
+  directory/access checks and close-error handling at their appropriate points.
+- Preserve deferred discovery/ordinary-overlay checks, root-only absent host
+  context, host/native distinctions, diagnostics, and borrowed filesystem
+  ownership. Independently required validation must not disappear merely to
+  reduce filesystem calls.
+- Keep internal documentation aligned with the completed handoff. Describe
+  aggregate accounting as load-wide and repetition/namespace state as
+  tier-scoped; the new internal overview/import prose currently calls the latter
+  per-tier accounting despite its distinct role.
+
+**Verification.** Review the non-import call path to confirm that applicable
+resolved facts are carried forward and absent-host derivation has one owner.
+Reuse the existing host/native, discovery selected/absent, ordinary/layered
+overlay, and passive-resource-context tests. Add a focused behavioral regression
+only for a consequential gap exposed by the change; do not assert incidental
+canonicalization/stat call counts. Run the development check and applicable
+ownership race checks after remediation.