docs: add public feedback goal fidelity evaluation

docs/architecture/evaluator-rag-prototype.md

@@ -123,6 +123,10 @@ Current corpus:
   expected-output comparisons, representative repository/commit histories, and
   regression commands before deep-analysis behavior can be promoted.
 
+- `public-feedback-goal-fidelity`: compares a shortened draft with the chosen
+  outcome and input/process/result chain, preserving verification limits and
+  separating optional alternatives. Decisions are illustrative.
+
 ## ECC Tools Mapping
 
 ECC Tools already flags missing RAG/evaluator evidence for retrieval,

examples/evaluator-rag-prototype/public-feedback-goal-fidelity/candidate-playbook.md

@@ -0,0 +1,47 @@
+# Public Feedback Goal Fidelity Playbook
+
+Candidate id: `preserved-outcome-and-chain`
+
+This is a synthetic, read-only comparison, with expected rubric decisions rather
+than a historical reproduction or measured model result. It responds to
+[reported feedback](https://github.com/affaan-m/ECC/issues/3695).
+
+## Compare Meaning Before Shortening
+
+1. Extract the user's chosen outcome and important causal steps.
+2. Shorten wording while retaining `INPUTS -> PROCESS -> RESULT` and why the
+   result matters. Here that purpose is sustained throughput.
+3. Compare the result with the original task. A human-assisted helper can be an
+   explicitly offered alternative, but must not silently replace automation.
+4. Retain uncertainty: a requested capability is not an implemented architecture;
+   technical limits and required verification still apply.
+
+The accepted and rejected drafts in `scenario.json` deliberately use the same
+topic. The rejected draft changes the outcome even though it sounds helpful.
+Evaluate meaning and connections, not just shared keywords or brevity.
+
+## Existing Artifact Control
+
+For an HTML-editing variant, provide the original artifact as evaluation input
+before asserting preservation. The contract can require supported technical
+corrections and removal of outdated emphasis while preserving useful facts,
+article purpose, substantial narrative, section structure and image references.
+Return copyable HTML when that is requested; do not promote the narrow correction
+to a new article theme. Supported additions and removals remain allowed.
+
+If the original is unavailable, preservation remains unverified. A revised code
+block establishes format only. Version comparison, rendering, factual checking
+and user acceptance are distinct evidence states. Do not invent the missing
+original or claim those checks passed.
+
+The source reports a corrected public comment, but its earlier revision was not
+independently recovered. For the HTML follow-up, original HTML and the earlier
+response remain unavailable; revised output was reported provided, with quality
+unverified. Exact ECC activation, version and causal contribution are unknown.
+
+## Repository Validation
+
+Run `node tests/docs/evaluator-rag-prototype.test.js`, Markdown lint on touched
+docs, and `git diff --check`. These repository checks do not validate a missing
+historical HTML artifact or establish user acceptance. No publication occurs in
+this evaluator example.

examples/evaluator-rag-prototype/public-feedback-goal-fidelity/report.json

@@ -0,0 +1,23 @@
+{
+  "schema_version": "ecc.evaluator-rag.report.v1",
+  "scenario_id": "public-feedback-goal-fidelity",
+  "run_id": "illustrative-goal-fidelity",
+  "read_only": true,
+  "result": "illustrative_expected_result",
+  "findings": [
+    {
+      "id": "meaning-before-length",
+      "severity": "warning",
+      "summary": "A shorter draft can retain vocabulary while losing the requested outcome or causal connection."
+    },
+    {
+      "id": "format-not-quality",
+      "severity": "info",
+      "summary": "A revised HTML code block establishes output format, not verified preservation, rendering or factual correctness."
+    }
+  ],
+  "recommended_next_action": {
+    "candidate_id": "preserved-outcome-and-chain",
+    "action": "Compare meaning and limits with the task before publishing; offer alternatives explicitly."
+  }
+}

examples/evaluator-rag-prototype/public-feedback-goal-fidelity/scenario.json

@@ -0,0 +1,48 @@
+{
+  "schema_version": "ecc.evaluator-rag.scenario.v1",
+  "scenario_id": "public-feedback-goal-fidelity",
+  "title": "Preserve the chosen outcome and causal chain when shortening feedback",
+  "mode": "read_only_prototype",
+  "objective": "Compare a shortened draft with the requested outcome, causal steps and verification limits before publication.",
+  "sources": [
+    {
+      "kind": "public_feedback",
+      "url": "https://github.com/affaan-m/ECC/issues/3695",
+      "purpose": "Reported corrected goal drift and proposed evaluation"
+    },
+    {
+      "kind": "public_comment",
+      "url": "https://github.com/evilbruce666/1c-odata-mcp/issues/56#issuecomment-6385767160",
+      "purpose": "Corrected public text cited by the reporter; not recovery of its earlier revision"
+    }
+  ],
+  "evidence_limits": "The earlier published revision was not independently recovered. Original HTML and earlier HTML response are unavailable; revised output was reported supplied but not compared, rendered, fact-checked or accepted by the user. ECC activation/version and causal contribution remain unknown.",
+  "inputs": {
+    "chosen_outcome": "OUTCOME_A: an end-to-end automated workflow producing the prepared result",
+    "causal_chain": [
+      "INPUTS: incoming data sources",
+      "PROCESS: intermediary tool layer prepares the result",
+      "RESULT: final prepared result, with sustained throughput as the purpose"
+    ],
+    "constraint": "Describe the requested capability, not a claim that it is implemented. Preserve genuine feasibility and verification limits.",
+    "accepted_draft": "Request an end-to-end flow from incoming sources through the intermediary tool layer to a prepared result, supporting sustained throughput. Implementation and feasibility remain to be verified.",
+    "rejected_draft": "Request a convenient helper that prepares suggestions and leaves routine exceptions for a person."
+  },
+  "retrieval_questions": [
+    "Does the draft preserve OUTCOME_A?",
+    "Are the inputs, process, result and throughput rationale still connected?",
+    "Are optional alternatives labeled as alternatives?",
+    "Are implementation and verification limits retained?"
+  ],
+  "forbidden_actions": [
+    "publishing feedback from this read-only evaluator",
+    "silently replacing OUTCOME_A with human-assisted OUTCOME_B",
+    "claiming implementation or verified HTML correction from a revised code block alone"
+  ],
+  "acceptance_gates": [
+    "chosen outcome remains unchanged",
+    "important causal steps and throughput rationale remain explicit",
+    "alternatives are separate proposals, not substituted requirements",
+    "unknown feasibility and verification remain explicit"
+  ]
+}

examples/evaluator-rag-prototype/public-feedback-goal-fidelity/trace.json

@@ -0,0 +1,33 @@
+{
+  "schema_version": "ecc.evaluator-rag.trace.v1",
+  "scenario_id": "public-feedback-goal-fidelity",
+  "run_id": "illustrative-goal-fidelity",
+  "read_only": true,
+  "illustrative": true,
+  "events": [
+    {
+      "phase": "observation",
+      "summary": "The synthetic task requests shorter public feedback about an automated end-to-end workflow."
+    },
+    {
+      "phase": "retrieval",
+      "summary": "Read the supplied outcome, causal chain and limits before comparing drafts."
+    },
+    {
+      "phase": "proposal",
+      "candidate_ids": [
+        "preserved-outcome-and-chain",
+        "silent-human-assisted-substitution"
+      ]
+    },
+    {
+      "phase": "verification",
+      "summary": "Expected rubric: retain automation, causal chain and throughput purpose; reject a convenient helper substituted for the requested outcome."
+    },
+    {
+      "phase": "promotion",
+      "promoted_candidate_id": "preserved-outcome-and-chain",
+      "summary": "Illustrative content recommendation only; no public comment is created or edited."
+    }
+  ]
+}

examples/evaluator-rag-prototype/public-feedback-goal-fidelity/verifier-result.json

@@ -0,0 +1,29 @@
+{
+  "schema_version": "ecc.evaluator-rag.verifier.v1",
+  "scenario_id": "public-feedback-goal-fidelity",
+  "run_id": "illustrative-goal-fidelity",
+  "read_only": true,
+  "evaluation_status": "illustrative_expected_decisions_not_historical_reproduction",
+  "candidates": [
+    {
+      "candidate_id": "preserved-outcome-and-chain",
+      "decision": "accepted",
+      "reasons": [
+        "Preserves the end-to-end automated OUTCOME_A",
+        "Connects incoming sources through the intermediary tool layer to the prepared result",
+        "Preserves sustained throughput as the rationale",
+        "Does not claim the requested architecture is implemented or verified"
+      ]
+    },
+    {
+      "candidate_id": "silent-human-assisted-substitution",
+      "decision": "rejected",
+      "reasons": [
+        "Substitutes human-assisted OUTCOME_B for the requested outcome",
+        "Drops incoming sources and the sustained throughput rationale",
+        "Presents an alternative as if it were the original requirement"
+      ]
+    }
+  ],
+  "promoted_candidate_id": "preserved-outcome-and-chain"
+}