docs: clarify evidence ownership and verification guidance

docs/ghidra-dos.md

@@ -23,17 +23,19 @@ not name its entry function `entry`. MCP uses `open_binary`, `list_procedures`,
 `procedure_info`, `read_function_instructions`, and `analyze_function` with the
 same admitted target and provider semantics.
 
-The classifier validates MZ page counts, initialized module length, header and
+The DOS MZ parser validates page counts, initialized module length, header and
 relocation-table bounds, relocation destinations, and the initialized entry
-point. DOS structural facts are parsed independently from Windows-header
-evidence. A recognized PE signature remains PE; recognized NE, LE, and LX
-headers are explicitly unsupported. A nonzero Windows-header pointer without a
-recognized signature remains an unresolved candidate, even when the DOS
-structure validates: automatic target selection fails closed and reports the
-DOS facts alongside the candidate. REA does not infer DOS intent from padding
-or infer a damaged Windows header from the pointer alone. Inventory
-classification returns `unknown` for that candidate and when its bounded prefix
-cannot establish the format; this is distinct from complete DOS validation.
+point using the available structural bytes and actual file length. It reports
+those DOS facts separately from Windows-header evidence. A recognized PE
+signature remains PE; recognized NE, LE, and LX headers are explicitly
+unsupported. A nonzero Windows-header pointer without a recognized signature
+remains an unresolved candidate, even when the DOS structure validates:
+automatic target selection fails closed and reports both observations.
+REA does not infer DOS intent from padding or infer a damaged Windows header
+from the pointer alone. Inventory classification is a bounded-prefix format
+hint; it returns `unknown` for that candidate or when its prefix cannot
+establish the format. That result is distinct from complete DOS validation and
+does not decide target admission.
 
 Ghidra imports a private target snapshot with `MzLoader`, language
 `x86:LE:16:Real Mode`, and compiler specification `default`. These settings,

docs/testing.md

@@ -236,7 +236,13 @@ instrumentation remains equivalent.
 
 ## Choose verification for the change
 
-| Change or claim                           | First check                                                   | Additional coverage                                         |
+`test:local` is fast source-level feedback. For the strongest practical
+regression, prefer a real public CLI/MCP workflow, then production-boundary
+integration, then a producer golden; use the table to select quick feedback and
+claim-specific additional evidence. Provider-behavior claims require the
+matching real-provider lane.
+
+| Change or claim                           | Fast feedback                                                 | Claim-specific coverage                                     |
 | ----------------------------------------- | ------------------------------------------------------------- | ----------------------------------------------------------- |
 | Source semantics                          | `npm run test:local -- PATH...`                               | Relevant boundary cases when the behavior crosses a port    |
 | CLI/MCP workflow                          | `npm run test:focused -- PATH...`                             | Actual public calls; an injected provider has limited scope |

docs/tool-design.md

@@ -30,8 +30,13 @@ inside one tool.
 
 A primitive reports a reusable fact about an identified object or relationship.
 A format-specific decoder qualifies; one application's business interpretation
-belongs outside the general contract. Expose provider APIs when they add a
-useful capability that existing tools cannot express.
+belongs outside the general contract. The format owner reports source
+observations and structural validation; the application or workflow owner
+applies use-case-specific admission policy without reinterpreting raw fields.
+For example, DOS MZ structure and Windows-header evidence stay separate, while
+target resolution decides whether the unresolved header permits admission.
+Expose provider APIs when they add a useful capability that existing tools
+cannot express.
 
 Compose a workflow when repeated use demonstrates a multi-source outcome REA
 can join without hiding caller choices or uncertainty. Batch homogeneous reads