feat: preserve full-text approval through writes and recovery

AGENTS.md

@@ -20,6 +20,8 @@ ConceptWeave owns automatic, evidence-bound **Semantic Model Engineering**. Do n
 - Full-text review views are read-only evidence; never strip their outer binding into a metadata-only patch and claim full-text decision or approval provenance.
 - Full-text decisions use the separate blank-start worksheet and atomic exact-view application; reverify the capture/report relation through finalization and whole-envelope governance. No reviewed-set downcast grants Zotero write authority.
 - Offline full-text commands reuse the private-file boundary and pass completed-view bytes unchanged into atomic validation. Finalized files await external approval verification; no CLI command issues approval or writes Zotero.
+- Full-text writes require a complete typed review, explicit destinations and mode; finish both local validation paths before real authority verification. Keep scope bindings through opaque execution/recovery receipts. Unknown original writes cannot become empty successful rollbacks, and serialized audit files are not executable authority.
+- Delayed original-write observations retain the exact submitted request and complete earlier receipt; matching metadata does not prove causal completion or authorize retry/rollback.
 - Published semantic truth is immutable; correction uses supersession/new release.
 - Public Rust APIs require beginner-readable documentation.
 - Owned production coverage target is 100% line/function/region/branch where tooling exposes it.

ARCHITECTURE.md

@@ -38,6 +38,10 @@ The separate Full-Text Review Worksheet starts blank and retains one capture ide
 
 ## Aggregate boundaries
 
+Full-Text Write Scope belongs to Research Intake and combines a complete capture-bound semantic review with independently authorized explicit metadata replacements and mode. Existing local validation is composed before either authority verifier. The admitted plan is opaque; bound execution and recovery wrappers reuse the existing conditional Local API cores, with no legacy-plan or free inverse-operation projection. Versioned scope commitments preserve the complete input identity without copying source text into receipts. The scope is not published semantic truth or an approval issuer. Audit serialization cannot restore executable authority; durable recovery and independently unknown original-write reconciliation remain explicit gaps.
+
+An Original Write Observation is a read-only audit projection inside this same context. It retains the opaque original receipt and exact submitted preconditions beside one later unverified response. It cannot clear uncertainty, prove request completion or enter execution/recovery as authority. The existing executor remains the sole mutation path.
+
 ### SemanticCandidate
 
 Smallest consistency boundary for a single proposed semantic artifact and its evidence-bound publication state. It cannot jump directly from Draft to Published.

CHANGELOG.md

@@ -10,6 +10,14 @@ All notable changes to ConceptWeave are documented here.
 
 ### Added
 
+- Research inventory now separates message ancestry, data changes and work dependencies from reviewed semantic relationships, with release limits and owner follow-up evidence.
+
+- Unresolved changes can be inspected again without resending them or losing the original outcome. Later observations do not authorize retry or recovery.
+
+- Research inventory now distinguishes three additional statistical-library candidates from adopted integrations and reviewed papers, with source-bound limitations and follow-up requirements.
+
+- Full-text-reviewed changes can now carry separately verified destination and execution approval through the local write and recovery workflow. Unknown writes remain unresolved until their state is proven; no command issues approval or changes the live library.
+
 - Private review commands now show saved text, accept completed decisions without replacing earlier work, and prepare a complete review for independent approval verification. They do not supply decisions or approve a review.
 
 - Private inspection of pending papers alongside their saved text, preserving missing material and leaving previous reports and decisions unchanged.

CLAUDE.md

@@ -11,3 +11,7 @@ Zotero source capture must not alter the metadata report or renew its approval.
 The separate full-text review view does not authorize decisions or writes. Keep its evidence binding intact; existing metadata-only apply/finalization cannot establish full-text-reviewed approval.
 
 Use the capture-bound blank-start worksheet for full-text decision work. Only the dedicated atomic view path may apply completed slots; finalization/evaluation reverify capture/report bindings and require whole-envelope external approval. Offline CLI view/application/finalization preserve that envelope and earlier files. They do not supply a reviewer, authenticate approval, or admit Zotero writes.
+
+The separate typed full-text write boundary additionally requires explicit destinations and mode verified against independent write authority, after all local checks. Keep the complete approved scope on the opaque plan and the same versioned commitment on every execution and recovery outcome. Do not extract legacy plans or mix inverse operations, restore authority from audit JSON, or call an unknown original write restored through empty rollback. No live-write CLI or approval issuer is introduced.
+
+Delayed original-write inspection preserves the actual submitted request and complete earlier receipt. Its later response is unverified evidence, even when metadata matches; it never clears uncertainty or grants retry/rollback authority.

crates/conceptweave-zotero/src/full_text_capture_tests.rs

@@ -1,6 +1,9 @@
 use super::*;
 use crate::{StewardReviewWorksheet, ZoteroItem, classify_snapshot};
 
+#[path = "full_text_write_tests.rs"]
+mod full_text_write_tests;
+
 fn completed_full_text_view(
     report: &ClassificationReport,
     worksheet: &FullTextReviewWorksheet,

crates/conceptweave-zotero/src/full_text_review.rs

@@ -1,6 +1,10 @@
 use super::*;
 use crate::StewardReviewWorksheet;
 
+#[path = "full_text_write.rs"]
+mod full_text_write;
+pub use full_text_write::*;
+
 /// Independently issued approval input for one full-text review context.
 /// The issuer must bind the complete reviewed labels as well as this capture.
 #[derive(Deserialize, Serialize)]
@@ -168,19 +172,29 @@ pub fn evaluate_full_text_review<F>(
 where
     F: FnOnce(&FullTextReviewedGoldenSet) -> bool,
 {
-    validate_review_capture(report, capture, &reviewed.capture_digest)?;
-    let review_evaluation = crate::evaluate_complete_reviewed_classification(
-        report,
-        &reviewed.reviewed_golden_set,
-        |_| verify_approval(reviewed),
-    )
-    .map_err(|_| FullTextError("full-text review is invalid or unverified"))?;
+    let review_evaluation = prepare_full_text_review(report, capture, reviewed)?;
+    if !verify_approval(reviewed) {
+        return Err(FullTextError("full-text review is invalid or unverified"));
+    }
     Ok(FullTextReviewEvaluation {
         capture_digest: reviewed.capture_digest.clone(),
         review_evaluation,
     })
 }
 
+fn prepare_full_text_review(
+    report: &ClassificationReport,
+    capture: &FullTextCapture,
+    reviewed: &FullTextReviewedGoldenSet,
+) -> Result<crate::GoldenSetEvaluation, FullTextError> {
+    validate_review_capture(report, capture, &reviewed.capture_digest)?;
+    if reviewed.reviewed_golden_set.labels.len() != report.classified_items.len() {
+        return Err(FullTextError("full-text review is invalid or unverified"));
+    }
+    crate::prepare_reviewed_golden_set(report, &reviewed.reviewed_golden_set)
+        .map_err(|_| FullTextError("full-text review is invalid or unverified"))
+}
+
 fn validate_review_capture(
     report: &ClassificationReport,
     capture: &FullTextCapture,