Clarify the assisted-authoring transport boundary before cohort execution

.github/workflows/ci.yml

@@ -112,6 +112,7 @@ jobs:
           assert "archsift/docs/architecture-knowledge-v1.md" in names
           assert "archsift/docs/architecture-knowledge-v2.md" in names
           assert "archsift/docs/authoring-check-v1.md" in names
+          assert "archsift/docs/authoring-check-v1.0.1.md" in names
           assert "archsift/docs/graph-change-v1.md" in names
           assert "archsift/docs/graph-snapshot-v1.md" in names
           assert "archsift/docs/method-v1.0.0.md" in names
@@ -860,7 +861,7 @@ jobs:
           authoring_path = root / "authoring-results.json"
           authoring_payload = {
               "schema_version": 1,
-              "protocol_version": "1.0.0",
+              "protocol_version": "1.0.1",
               "archsift_source_commit": "a" * 40,
               "material_set_content_identity": (
                   "sha256:deca6741b7c69fbb313ed1292caa55a7a698eecb60f4a42aed849b3dcffd57ee"

README.md

@@ -115,10 +115,13 @@ three pass. The initial simulated cohort has run and its committed
 run and no human-participant claim is made. A completed cohort is checked offline with exactly
 `archsift usability-results usability-results.json`.
 
-The [simulated assisted-authoring protocol](docs/authoring-check-v1.md) freezes a synthetic material
+The [simulated assisted-authoring protocol 1.0.1](docs/authoring-check-v1.0.1.md) freezes a synthetic material
 set, six objective CLI milestones, exactly four fresh sessions using distinct agent products, and a
 three-of-four success threshold. The protocol publishes no answer dossier or expected verdict and
-retains only privacy-bounded outcome metadata. No cohort has been run yet. A future completed result
+retains only privacy-bounded outcome metadata. ArchSift and every CLI milestone remain offline with
+outbound sockets blocked; only the separately controlled external agent product's ordinary model
+transport is permitted, with browsing, search, retrieval plugins, and private context prohibited.
+No cohort has been run yet. A future completed result
 will be checked offline with exactly
 `archsift authoring-results authoring-results.json`; until such a record exists and validates, no
 assisted-authoring success claim is made.

SECURITY.md

@@ -33,3 +33,9 @@ records, prompts, command output, and transcripts are temporary private data and
 committed. `authoring-results` reads only one regular JSON file beneath the current directory,
 rejects links and escaping paths, performs no network access, writes nothing, and accepts only the
 strict privacy-bounded result fields in its packaged schema.
+
+Protocol 1.0.1 distinguishes ArchSift's offline boundary from the external author's transport.
+Every ArchSift milestone runs with outbound sockets blocked. A cohort operator may separately
+authorise an agent product's ordinary model transport outside ArchSift, but browser/search access,
+retrieval plugins, network source lookup, private repositories, and continued private context are
+not eligible protocol inputs.

docs/authoring-check-v1.0.1.md

@@ -0,0 +1,123 @@
+# Simulated assisted-authoring check — protocol 1.0.1
+
+Protocol 1.0.1 supersedes protocol 1.0.0 before the first cohort. It preserves the material set,
+milestones, privacy contract, and success threshold while clarifying the boundary between
+ArchSift's offline runtime and an external authoring product's ordinary user-controlled model
+transport. Protocol 1.0.0 remains frozen at [authoring-check-v1.md](authoring-check-v1.md). At
+publication, no cohort has been run and no success claim is made.
+
+This is an independent, fully synthetic check of whether agent products can use ArchSift's public
+authoring surfaces to produce and assess a structurally valid decision dossier. It does not measure
+human usability, decision quality in a real organisation, or the truth of an architecture
+recommendation.
+
+## Frozen criterion
+
+Run exactly four fresh sessions using four distinct agent products. Each session starts in a fresh
+environment, receives only the instructions and synthetic material named below, and binds to the
+same full 40-character commit of ArchSift on which this protocol and material set have landed. The
+criterion is met when at least three of the four sessions complete every objective milestone with
+no maintainer intervention.
+
+A failed or interrupted session remains failed. Do not repair, rerun, replace, or selectively omit
+a completed session. A later attempt is a new precommitted four-session cohort and result record.
+
+## Transport and source boundary
+
+ArchSift, every ArchSift CLI subprocess, and all local validation run with outbound sockets blocked.
+ArchSift never invokes, embeds, imports, or depends on an agent product, model client, model API, or
+network service.
+
+The external agent product may use only its ordinary user-controlled model transport. That
+transport is selected and disclosed by the cohort operator, runs outside ArchSift, and is not a
+product dependency or an ArchSift transmission. The session must not use a browser, web search,
+network source lookup, retrieval plugin, remote shell, private repository, continued conversation,
+or any material beyond the frozen public inputs. This distinction follows FR-018: an external
+author may operate under the user's control without making ArchSift an online authoring service.
+
+## Frozen material
+
+The only case source material is the packaged `authoring-material/` set:
+
+- `brief.md`: a fictional domain-neutral review brief;
+- `repository/required_fields.py`: inert illustrative source text;
+- `repository/routing_rules.py`: inert illustrative source text; and
+- `manifest-v1.json`: byte lengths, SHA-256 identities, a synthetic repository commit, and the
+  canonical material-set identity.
+
+The repository snippets are data for `register-repository`; they are not an executable project.
+They contain no completed dossier, expected verdict, answer key, or reference output. Sessions
+receive no private case, real repository, personal data, previous result, transcript, or
+maintainer-authored dossier.
+
+## Session instructions
+
+Give every agent product these instructions without product-specific hints:
+
+1. Start in a fresh temporary working directory with ArchSift installed from the cohort's bound
+   source checkout or built wheel. Do not browse, search, fetch sources, or inspect other
+   workspaces. Run ArchSift commands through the supplied socket-blocked environment.
+2. Read `authoring-material/brief.md` and `authoring-material/manifest-v1.json`. Treat every fact
+   not stated by the material as an assumption, estimate, missing item, or unknown. Never invent an
+   observation or artefact.
+3. Run `archsift init synthetic-routing`.
+4. Register the brief with `archsift register-document`, and register exactly the two repository
+   files with `archsift register-repository` using the full synthetic commit identity in the
+   manifest. Registration is inert: do not run or import the snippets.
+5. Inspect the complete schema-version-3 contract with
+   `archsift dossier-schema --schema-version 3 --json`.
+6. Author `synthetic-routing/case.yaml` from the frozen material. Cite registrations only where the
+   schema permits them and preserve the evidence authorship and attestation boundary.
+7. Run `archsift prerequisites synthetic-routing --json` and complete every prerequisite that the
+   synthetic material supports. Leave unsupported facts explicitly missing or unknown.
+8. Run `archsift validate synthetic-routing` and then `archsift assess synthetic-routing`.
+9. Stop. Report only milestone exit outcomes to the cohort harness. Do not return dossier text,
+   reports, record bytes, paths, prompts, command output, or transcripts.
+
+The harness supplies the installed executable path, frozen material path, temporary workspace,
+socket-blocking environment, and these instructions. Any additional advice, correction, command,
+file edit, retry after a milestone failure, or interpretation from a maintainer is intervention and
+makes that session fail.
+
+## Objective milestones
+
+Each session records exactly these six milestones:
+
+- `register_material`: both registrations return exit `0` and bind the manifest identities;
+- `inspect_schema`: schema version 3 is emitted successfully before dossier authoring completes;
+- `author_dossier`: one schema-version-3 dossier is written without a supplied answer key;
+- `complete_prerequisites`: the command runs successfully and the agent responds to its worklist;
+- `validate`: final validation returns exit `0`; and
+- `assess`: final assessment returns exit `0` and creates the canonical outputs.
+
+A session passes only if every milestone is `pass` and `maintainer_intervention` is `false`.
+Milestone assessment is mechanical from command outcomes and file existence; it does not score the
+recommendation or authored prose. Any valid verdict is acceptable.
+
+## Privacy-bounded result
+
+Record only the fields allowed by
+`src/archsift/schemas/authoring-results-v1.schema.json`: pseudonymous session ID, public agent
+product/model and harness version, coarse operating-system/Python/install-mode metadata, six
+milestone outcomes, intervention flag, derived session result, and a short non-sensitive failure
+reason. The cohort record also binds protocol 1.0.1, the exact ArchSift source commit, and the
+frozen material-set identity.
+
+Never commit prompts, transcripts, generated dossiers, workspaces, command output, absolute paths,
+user or account names, machine identifiers, credentials, real organisation data, or personal data.
+Delete temporary session workspaces after extracting the bounded fields. The strict schema rejects
+all extra fields and limits every retained string.
+
+## Offline validation
+
+Place the completed record at `authoring-results.json` beneath the current directory and run:
+
+```console
+archsift authoring-results authoring-results.json
+```
+
+The validator performs no network access and writes nothing. A conforming record with three or
+four passing sessions exits `0` with `criterion-met`. A valid record below threshold exits `12`
+with `criterion-not-met`. Malformed JSON exits `10`, unsupported schema/protocol versions exit
+`11`, contract or consistency failures exit `12`, unsafe paths exit `13`, and unavailable files
+exit `14`. `--json` emits a stable machine-readable summary; `--quiet` emits nothing.

docs/usage.md

@@ -376,7 +376,7 @@ errors are not accepted cohort evidence.
 ### `archsift authoring-results <results>`
 
 Validates one completed simulated assisted-authoring cohort against
-[protocol 1.0.0](authoring-check-v1.md) and the packaged privacy-bounded result schema. Exactly four
+[protocol 1.0.1](authoring-check-v1.0.1.md) and the packaged privacy-bounded result schema. Exactly four
 fresh sessions from four distinct agent products are required. A session passes only when material
 registration, schema inspection, dossier authoring, prerequisite completion, validation, and
 assessment all pass without maintainer intervention. Three or four passing sessions exit `0` with
@@ -389,6 +389,12 @@ links and escaping paths, performs no network access, writes nothing, and never
 workspace. `--json` returns protocol, session, pass-count, criterion, diagnostic, and exit-code
 fields; `--quiet` returns only the exit status.
 
+Protocol 1.0.1 requires outbound sockets to be blocked for ArchSift and every local milestone. The
+external authoring product may use its ordinary user-controlled model transport outside ArchSift,
+but may not browse, search, retrieve network sources, load plugins or private context, or continue a
+prior session. The superseded pre-cohort [protocol 1.0.0](authoring-check-v1.md) remains frozen and
+its result version remains supported.
+
 ### `archsift method-review-results <results>`
 
 Validates one completed architecture-method review result against the frozen

pyproject.toml

@@ -78,6 +78,7 @@ exclude = [
 "docs/usability-check-v1.md" = "archsift/docs/usability-check-v1.md"
 "docs/usability-check-v2.md" = "archsift/docs/usability-check-v2.md"
 "docs/authoring-check-v1.md" = "archsift/docs/authoring-check-v1.md"
+"docs/authoring-check-v1.0.1.md" = "archsift/docs/authoring-check-v1.0.1.md"
 "authoring-material/brief.md" = "archsift/authoring-material/brief.md"
 "authoring-material/manifest-v1.json" = "archsift/authoring-material/manifest-v1.json"
 "authoring-material/repository/required_fields.py" = "archsift/authoring-material/repository/required_fields.py"

src/archsift/authoring_results.py

@@ -16,7 +16,9 @@
 
 from archsift.diagnostics import Diagnostic, ExitCode
 
-PROTOCOL_VERSION = "1.0.0"
+PROTOCOL_VERSION_1_0_0 = "1.0.0"
+PROTOCOL_VERSION = "1.0.1"
+SUPPORTED_PROTOCOL_VERSIONS = (PROTOCOL_VERSION_1_0_0, PROTOCOL_VERSION)
 RESULT_SCHEMA_VERSION = 1
 REQUIRED_SESSION_COUNT = 4
 REQUIRED_PASS_COUNT = 3
@@ -32,7 +34,7 @@
     "sha256:deca6741b7c69fbb313ed1292caa55a7a698eecb60f4a42aed849b3dcffd57ee"
 )
 MAX_RESULT_BYTES = 64 * 1024
-_REQUIREMENT = "AUTHORING-1.0.0"
+_REQUIREMENT = "AUTHORING-RESULTS-v1"
 
 
 @dataclass(frozen=True, slots=True)
@@ -296,14 +298,22 @@ def _unsupported(payload: dict[str, object]) -> AuthoringValidationResult:
                 "authoring-results-version-unsupported",
                 "The declared authoring-result schema and protocol versions are unsupported.",
                 field,
-                f"Use schema version {RESULT_SCHEMA_VERSION} with protocol {PROTOCOL_VERSION}.",
+                (
+                    f"Use schema version {RESULT_SCHEMA_VERSION} with protocol "
+                    f"{PROTOCOL_VERSION_1_0_0} or {PROTOCOL_VERSION}."
+                ),
             ),
         ),
         protocol_version=protocol if type(protocol) is str else None,
     )
 
 
 def _validate_payload(payload: object) -> AuthoringValidationResult:
+    declared_protocol = (
+        payload.get("protocol_version")
+        if type(payload) is dict and type(payload.get("protocol_version")) is str
+        else None
+    )
     errors = sorted(
         _schema_validator().iter_errors(payload),
         key=lambda error: tuple((type(part).__name__, repr(part)) for part in error.absolute_path),
@@ -316,10 +326,10 @@ def _validate_payload(payload: object) -> AuthoringValidationResult:
                     "authoring-results-contract",
                     "The result data does not match the authoring-results-v1 contract.",
                     _json_path(errors[0].absolute_path),
-                    "Correct the named field using protocol 1.0.0 and the packaged JSON schema.",
+                    "Correct the named field using a supported protocol and the packaged schema.",
                 ),
             ),
-            protocol_version=PROTOCOL_VERSION,
+            protocol_version=cast(str | None, declared_protocol),
         )
     result_payload = cast(dict[str, object], payload)
     sessions = cast(list[dict[str, object]], result_payload["sessions"])
@@ -341,7 +351,7 @@ def _validate_payload(payload: object) -> AuthoringValidationResult:
                 "authoring-agent-product-duplicate",
                 "Agent product names must be unique within the four-session cohort.",
                 "$.sessions",
-                "Use four distinct agent products as required by protocol 1.0.0.",
+                "Use four distinct agent products as required by the supported protocol.",
             )
         )
     if result_payload["material_set_content_identity"] != MATERIAL_SET_CONTENT_IDENTITY:
@@ -393,7 +403,7 @@ def _validate_payload(payload: object) -> AuthoringValidationResult:
     return _result(
         ExitCode.VALIDATION_FAILED if diagnostics else ExitCode.SUCCESS,
         diagnostics,
-        protocol_version=PROTOCOL_VERSION,
+        protocol_version=cast(str, result_payload["protocol_version"]),
         session_count=len(sessions),
         passed_session_count=passed,
     )
@@ -419,7 +429,7 @@ def validate_authoring_results(path: Path) -> AuthoringValidationResult:
         return _malformed("The authoring-result file is not strict JSON.")
     if type(payload) is dict and (
         payload.get("schema_version") != RESULT_SCHEMA_VERSION
-        or payload.get("protocol_version") != PROTOCOL_VERSION
+        or payload.get("protocol_version") not in SUPPORTED_PROTOCOL_VERSIONS
     ):
         return _unsupported(payload)
     return _validate_payload(payload)
@@ -428,10 +438,12 @@ def validate_authoring_results(path: Path) -> AuthoringValidationResult:
 __all__ = [
     "MATERIAL_SET_CONTENT_IDENTITY",
     "PROTOCOL_VERSION",
+    "PROTOCOL_VERSION_1_0_0",
     "REQUIRED_MILESTONES",
     "REQUIRED_PASS_COUNT",
     "REQUIRED_SESSION_COUNT",
     "RESULT_SCHEMA_VERSION",
+    "SUPPORTED_PROTOCOL_VERSIONS",
     "AuthoringValidationResult",
     "validate_authoring_results",
 ]

src/archsift/schemas/authoring-results-v1.schema.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "$id": "https://github.com/maximalfocus/archsift/schemas/authoring-results-v1.schema.json",
-  "title": "ArchSift simulated assisted-authoring results 1.0.0",
+  "title": "ArchSift simulated assisted-authoring results v1",
   "type": "object",
   "additionalProperties": false,
   "required": [
@@ -14,7 +14,7 @@
   ],
   "properties": {
     "schema_version": { "const": 1 },
-    "protocol_version": { "const": "1.0.0" },
+    "protocol_version": { "enum": ["1.0.0", "1.0.1"] },
     "archsift_source_commit": {
       "type": "string",
       "pattern": "^[0-9a-f]{40}$"