fix: centralize advisory analysis and execution context

docs/engineering/ENGINEERING_REPORTING.md

@@ -213,11 +213,11 @@ reports never claim successful implementation or delivery.
 
 ## Advisory Codex analysis
 
-After a terminal report is written, the runner may request one separate Codex
-CLI analysis of that exact local report. The analysis is read-only, bounded and
-stored locally per run under `.engineering/report-analysis/<run-id>.md`. It
-distils findings, issues, risks, next steps and advice for the Product
-Architect. Its output is advisory and redacted before persistence. Every
+After a terminal report is written, the lifecycle dispatcher may request one
+separate Codex CLI analysis of that exact report. The analysis is read-only,
+bounded, redacted before persistence, and registered as an integrity-verified
+CENTRAL Markdown artifact for the same `(project_id, run_id)`. It distils
+findings, issues, risks, next steps and advice for the Product Architect. Every
 analysis also records a bounded **Analyseverwerking** status. If the provider
 is unavailable, fails, or returns an invalid structured response, the matching
 safe reason is shown there; raw provider output and diagnostics are never
@@ -230,15 +230,18 @@ and authoritative.
 
 The private Engineering Status dashboard exposes an **AI analysis** column in
 Prompt History next to the engineering report. View and download actions are
-available only when the analysis file belongs to that exact Run ID; analyses
-from another execution are never selected as a fallback.
+available only for the matching verified CENTRAL artifact; an older local file
+or another run is never selected as a fallback. A response that is unavailable
+or is not Markdown is rendered as a localized unavailable message, never as
+technical JSON.
 
 For a controlled temporary processing failure (`provider_failed`,
 `provider_unavailable` or `invalid_structured_response`), the analysis dialog
-also offers **Generate analysis again**. It is bound to that same indexed
-terminal report and regenerates only the advisory analysis; it never resumes,
-retries or changes the Engineering execution, checkpoint, branch or pull
-request. A successfully processed analysis is deliberately not retryable.
+also offers **Generate analysis again**. It is bound to that same centrally
+indexed terminal report, emits a redacted Console audit event, and regenerates
+only the advisory analysis; it never resumes, retries or changes the
+Engineering execution, checkpoint, branch or pull request. A successfully
+processed analysis is deliberately not retryable.
 
 If a terminal report temporarily cannot be read in its dialog, the dashboard
 offers **Reload report**. That action repeats only the read-only retrieval of

docs/engineering/P_CENTRAL_CONSOLE_AUTHORITY_MAP.md

@@ -19,7 +19,7 @@ physical execution binding only; it is not Console authority.
 | Provider usage | SERVER_PLATFORM_NATIVE | Current single-runtime Codex capacity is a Server/CENTRAL projection. |
 | Evidence report downloads | CENTRAL_NATIVE | Report index and artifact path are authorized by `(project_id, run_id)` in CENTRAL. |
 | Prompt chat history | CENTRAL_NATIVE (read projection) | Immutable transcript lookup is scoped by CENTRAL project/run lineage. |
-| Provider-backed chat mutation and report analysis | RETIRED/UNREACHABLE from migrated routes | No CENTRAL Server authority is invented for historical root-backed mutation. |
+| Advisory report analysis and retry | CENTRAL_NATIVE (bounded project mutation) | An exact-run, integrity-verified Markdown artifact is read from CENTRAL. Retry is allowed only for a controlled transient analysis status and uses the matching CENTRAL report; it is audited and cannot change the execution. |
 | Configuration | CENTRAL_NATIVE (read projection) | Only CENTRAL maintenance/capacity policy is exposed; old root-local settings fail closed. |
 | Component logs | CENTRAL_NATIVE | Server-owned CENTRAL log index, explicitly `PLATFORM` scoped. |
 | Worktree, provider-login and update actions | RETIRED/UNREACHABLE | They are not supported installed Console actions in this phase. |

docs/engineering/P_TRANSPORT_CONSOLE_ROUTE_OWNERSHIP_MATRIX.md

@@ -18,7 +18,7 @@ change authority, select a checkout, or delegate that route.
 | PLATFORM | `GET/POST /api/configuration`; `GET /api/central-data/export`; `POST /api/central-data/{relocate,relocate/browse,relocate/discard,import}`; `GET/POST /api/central-database/configuration` | Server settings and Central data operations; paths and imported bytes are never placed in the audit log |
 | HOST_ADMIN | `GET /api/host-admin/diagnostics` | Bounded installation-only disk and managed-runtime observation; no project, queue, execution or mutation authority |
 | PLATFORM | `GET /v1/operations/projects` | Operations Console platform listing |
-| PROJECT | `GET /api/prompt-history`, `/api/prompt-history/{run}/{report,chat,details}`, `/api/telemetry/{date}`; `POST /api/telemetry/clear`, `/api/execution-{dismiss,retry}`, `/api/dashboard-translate` | Project history, telemetry and project actions; no valid selected project returns `409 CONSOLE_PROJECT_UNAVAILABLE` |
+| PROJECT | `GET /api/prompt-history`, `/api/prompt-history/{run}/{report,analysis,chat,details}`, `/api/telemetry/{date}`; `POST /api/prompt-history/{run}/analysis-retry`, `/api/telemetry/clear`, `/api/execution-{dismiss,retry}`, `/api/dashboard-translate` | Project history, centrally indexed advisory analysis, telemetry and project actions; no valid selected project returns `409 CONSOLE_PROJECT_UNAVAILABLE` |
 | TRANSPORT_INTERNAL | `/diagnostics/topology`, `/healthz`, `/readyz`, `/v1/projects/{project}/submissions`, `/v1/agent/{pair,register,heartbeat,attachment}` | Transport probes and authenticated transport endpoints, not Console delegation |
 | HISTORICAL_UNREACHABLE | `POST /api/runtime-directory/open` | Explicitly retired checkout-bound runtime action (`410 RUNTIME_DIRECTORY_RETIRED`) |
 
@@ -45,6 +45,11 @@ Mission, Engineering Action, correlation and target repository. For Forge, the
 accepted versioned provenance also supplies host, mission/intent revisions,
 runtime-prompt identifier and digest, and retry correlation when present.
 
+The active card names two deliberately separate EP states: **EP execution
+phase** is the state recorded for the Execution Host run; **EP dispatcher
+state** is the FIFO dispatch record's orchestration state. They can both be
+`RUNNING` during normal operation, but neither is inferred from the other.
+
 The Console never derives a branch, checkout, tracked-file count, Mission
 summary or prompt content. Those remain absent until the Execution Host records
 run-specific evidence or Forge supplies a separately versioned context.
@@ -57,6 +62,15 @@ audited, but the log contains only action, actor, outcome and canonical target
 identifiers: no prompt text, AI-chat content, queue reason, local path, import
 payload or downloaded bytes.
 
+An advisory AI analysis is a separately indexed, integrity-verified CENTRAL
+Markdown artifact for the exact terminal `(project_id, run_id)`. The Console
+never falls back to a repository checkout or a legacy local analysis file.
+Only the bounded retryable processing states may be regenerated from that
+run's CENTRAL report; regeneration records a `dashboard_action_completed`
+audit event and cannot change the execution outcome. Document responses that
+are absent, malformed, or not Markdown are shown as a localized unavailable
+message, never as a JSON payload in the Console.
+
 ## Test and qualification coverage
 
 The matrix is not documentation-only. The following checks are the normal

package.json

@@ -1,7 +1,7 @@
 {
   "name": "engineering-platform-browser-validation",
   "private": true,
-  "version": "2.3.19",
+  "version": "2.3.20",
   "scripts": {
     "test:engineering-dashboard": "PYTHONPATH=src python3 -m engineering_platform.dashboard_browser_validation",
     "test:engineering-dashboard-logic": "node --test tests/engineering/dashboard_status_store.test.mjs tests/engineering/ui_localization_contract.test.mjs",

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "engineering-platform"
-version = "2.3.19"
+version = "2.3.20"
 description = "Local-first Engineering Platform execution operations runtime"
 readme = "README.md"
 requires-python = ">=3.11"

src/engineering_platform/ENGINEERING_PLATFORM_CONFIG.json

@@ -3,7 +3,7 @@
   "platform": {
     "id": "engineering-platform",
     "name": "Engineering Platform",
-    "version": "2.3.19",
+    "version": "2.3.20",
     "generation": 2,
     "documentation_namespace": "engineering-platform",
     "capability_registry_version": 1

src/engineering_platform/ENGINEERING_PLATFORM_VERSION.json

@@ -1,15 +1,15 @@
 {
   "bootstrap_contract": "2026.12",
   "checkpoint_format": 1,
-  "dashboard_version": "2.3.19",
+  "dashboard_version": "2.3.20",
   "handoff_protocol": 1,
   "memory_format": 2,
   "minimum_codex_cli": "0.146.0",
   "inbox_protocol": 1,
-  "platform_version": "2.3.19",
+  "platform_version": "2.3.20",
   "report_format": 2,
-  "runner_version": "2.3.19",
+  "runner_version": "2.3.20",
   "status_model": 1,
   "storage_schema": 44,
-  "watcher_version": "2.3.19"
+  "watcher_version": "2.3.20"
 }

src/engineering_platform/assets/dashboard.js

@@ -1618,18 +1618,25 @@ function replaceWithLocalFilesystemLink(element, replacementValue = element?.tex
   element.replaceWith(link);
   return link;
 }
-function executionContextField(label, value, badge = false, folder = false) {
+function executionContextField(label, value, folder = false) {
   const field = document.createElement("p"), caption = document.createElement("span"), content = document.createElement("span");
   field.className = "field";
   caption.className = "label";
   caption.textContent = label;
   const supplied = executionContextValue(value);
   const output = folder && supplied.startsWith("/") ? localFilesystemLink(supplied) : content;
   if (output === content) content.textContent = supplied || t("execution_context.not_supplied");
-  if (badge) output.classList.add("execution-context__phase");
   field.append(caption, output);
   return field;
 }
+function executionContextRuntimeStatus(value) {
+  const raw = executionContextValue(value);
+  return raw ? translate(raw) : "";
+}
+function executionContextTimestamp(value) {
+  const raw = executionContextValue(value);
+  return raw ? formatTimestamp(raw) : "";
+}
 function inheritModalAccent(modal, trigger) {
   const source = trigger?.closest(".current-run,[data-modal-accent-source],.dashboard-modal-shell");
   const sample = source && document.createElement("span");
@@ -1683,7 +1690,7 @@ function renderExecutionContext(context, execution = {}) {
   if (!context || typeof context !== "object") {
     card.replaceChildren(
       Object.assign(document.createElement("strong"), { textContent: t("ui.execution_context") }),
-      ...hostFields.map(([label, value, isExecutionMode]) => isExecutionMode ? executionModeField(value) : executionContextField(label, value, false, label === t("detail.target_checkout"))),
+      ...hostFields.map(([label, value, isExecutionMode]) => isExecutionMode ? executionModeField(value) : executionContextField(label, value, label === t("detail.target_checkout"))),
       Object.assign(document.createElement("p"), { textContent: t("execution_context.not_supplied") }),
     );
     return;
@@ -1696,15 +1703,15 @@ function renderExecutionContext(context, execution = {}) {
     [t("execution_context.engineering_summary"), context.engineering_summary],
     [t("execution_context.current_intent"), context.current_intent],
     [t("execution_context.current_engineering_action"), context.current_engineering_action],
-    [t("execution_context.execution_phase"), context.execution_phase, true],
+    [t("execution_context.execution_phase"), executionContextRuntimeStatus(context.execution_phase)],
     [t("execution_context.planning_confidence"), context.planning_confidence],
     [t("execution_context.current_iteration"), context.current_iteration],
     [t("execution_context.mission_progress"), context.mission_progress],
-    [t("execution_context.last_runtime_update"), context.last_runtime_update || context.last_updated_timestamp],
+    [t("execution_context.last_runtime_update"), executionContextTimestamp(context.last_runtime_update || context.last_updated_timestamp)],
     [t("execution_context.decision_evidence_reference"), context.decision_evidence_reference || context.decision_evidence],
     [t("execution_context.decision_type"), context.decision_type],
     [t("execution_context.execution_receipt_reference"), context.execution_receipt_reference || context.last_execution_receipt],
-    [t("execution_context.dispatcher_state"), context.dispatcher_state],
+    [t("execution_context.dispatcher_state"), executionContextRuntimeStatus(context.dispatcher_state)],
     [t("execution_context.approved_mission_queue_state"), context.approved_mission_queue_state],
     [t("execution_context.producer_host"), context.producer_host_id],
     [t("execution_context.mission_revision"), context.mission_revision],
@@ -1731,10 +1738,10 @@ function renderExecutionContext(context, execution = {}) {
   ];
   card.replaceChildren(
     Object.assign(document.createElement("strong"), { textContent: t("ui.execution_context") }),
-    ...hostFields.map(([label, value, isExecutionMode]) => isExecutionMode ? executionModeField(value) : executionContextField(label, value, false, label === t("detail.target_checkout"))),
+    ...hostFields.map(([label, value, isExecutionMode]) => isExecutionMode ? executionModeField(value) : executionContextField(label, value, label === t("detail.target_checkout"))),
     ...suppliedFields.map(([label, value]) => executionContextField(label, value)),
     ...(planningFields.length
-      ? planningFields.map(([label, value, badge]) => executionContextField(label, value, badge))
+      ? planningFields.map(([label, value]) => executionContextField(label, value))
       : [Object.assign(document.createElement("p"), {
         className: "execution-context__planning-empty",
         textContent: t("execution_context.planning_not_supplied"),
@@ -7313,19 +7320,14 @@ function loadPromptHistoryDocument() {
       (promptHistoryDocumentKind === "analysis" ? "/analysis" : "/report"),
     { cache: "no-store" },
   )
-    .then((response) =>
-      response.ok
+    .then((response) => {
+      const markdown = response.headers.get("content-type")?.toLowerCase().includes("text/markdown");
+      return response.ok && markdown
         ? response.text()
-        : Promise.reject(
-            Error(
-              t(
-                promptHistoryDocumentKind === "analysis"
-                  ? "history.analysis_unavailable"
-                  : "history.report_unavailable",
-              ),
-            ),
-          ),
-    )
+        : Promise.reject(Error(
+          t(promptHistoryDocumentKind === "analysis" ? "history.analysis_unavailable" : "history.report_unavailable"),
+        ));
+    })
     .then((text) => {
       if (!text)
         throw Error(
@@ -7382,19 +7384,15 @@ async function retryPromptHistoryAnalysis() {
       "/api/prompt-history/" + encodeURIComponent(promptHistoryReportRun) + "/analysis-retry",
       { method: "POST", headers: { "Content-Type": "application/json" }, body: "{}" },
     );
+    const markdown = response.headers.get("content-type")?.toLowerCase().includes("text/markdown");
     const text = await response.text();
-    if (!response.ok) {
-      let detail = "";
-      try { detail = JSON.parse(text).error || ""; } catch { /* controlled fallback below */ }
-      throw Error(detail || t("history.analysis_retry_failed"));
-    }
+    if (!response.ok || !markdown) throw Error(t("history.analysis_retry_failed"));
     promptHistoryReportText = text;
     renderMarkdownDocument(content, text);
     button.hidden = !reportAnalysisCanRetry(text);
-    void recordUserAction("prompt_history_analysis_regenerated", promptHistoryReportRun);
   } catch (error) {
     renderMarkdownDocument(content, promptHistoryReportText);
-    showDashboardError(error instanceof Error ? error.message : "", t("history.analysis_retry_failed"));
+    showDashboardError(t("history.analysis_retry_failed"), t("history.analysis_retry_failed"));
   } finally {
     button.disabled = false;
   }
@@ -7532,7 +7530,7 @@ function promptDetailExecutionSections(history) {
     detailField(t("detail.mission_id"), executionContextValue(context.mission_id) || t("execution_context.not_supplied")),
     detailField(t("execution_context.business_summary"), executionContextValue(context.business_summary) || t("execution_context.not_supplied")),
     detailField(t("execution_context.engineering_summary"), executionContextValue(context.engineering_summary) || t("execution_context.not_supplied")),
-    detailField(t("execution_context.execution_phase"), executionContextValue(context.execution_phase) || t("execution_context.not_supplied")),
+    detailField(t("execution_context.execution_phase"), executionContextRuntimeStatus(context.execution_phase) || t("execution_context.not_supplied")),
     detailField(t("execution_context.mission_lifecycle"), executionContextValue(context.mission_lifecycle) || t("execution_context.not_supplied")),
     detailField(t("execution_context.decision_evidence_reference"), executionContextValue(context.decision_evidence_reference || context.decision_evidence) || t("execution_context.not_supplied")),
     detailField(t("execution_context.execution_receipt_reference"), executionContextValue(context.execution_receipt_reference || context.last_execution_receipt) || t("execution_context.not_supplied")),