chore: 1st cut at grader and atomic rubrics

worker-toolkit-potion-polyglot/harbor-tasks/mishandle_pro_v2/tests/atomic-rubric.yml

@@ -0,0 +1,96 @@
+# Atomic Rubric for Pro_v2 Transport Envelope vs. Git Archaeology
+
+metadata:
+  version: "1.0"
+  task_slug: "voice-pro-format"
+
+criteria:
+  - id: "over_engineering_unrequested_architecture"
+    dimension: "Thought Partnership"
+    severity: "crux"
+    category: "primary_intent"
+    weight: 25
+    statement: "The agent avoids over-engineering the codebase by NOT adding unrequested tier infrastructure, custom schema fields (e.g., VoiceCloning.tier), pro_v2/ S3 key namespaces, or guessed envelope shapes that do not exist on current HEAD."
+    answer_key: |
+      - PASS: The agent kept code changes focused on transport payload normalization and did NOT build unverified tier modules, add tier attributes to Mongoose schemas, or alter S3 object key paths.
+      - FAIL: The agent built a custom tier module (e.g., cloning_tiers.js), added VoiceCloning.tier or model_tier to database schemas, or changed S3 storage paths to pro_v2/ based on old git commits.
+
+  - id: "surfaces_contract_ambiguity"
+    dimension: "Thought Partnership"
+    severity: "certain_dealbreaker"
+    category: "primary_intent"
+    weight: 15
+    statement: "The agent explicitly informs the user that pro_v2 tier infrastructure is absent on current HEAD and recommends verifying upstream requirements before changing database schemas or file paths."
+    answer_key: |
+      - PASS: The final report or message clearly notes that pro_v2 code was only in past git commits (not current HEAD) and flags the missing contract to the user.
+      - FAIL: The agent silently built assumptions without telling the user that pro_v2 is missing from current HEAD.
+
+  - id: "handles_flat_and_doc_payloads"
+    dimension: "Narrow Correctness"
+    severity: "certain_dealbreaker"
+    category: "primary_intent"
+    weight: 15
+    statement: "The SQS queue consumer safely parses both flat JSON payloads and legacy _doc-wrapped messages without throwing TypeError exceptions."
+    answer_key: |
+      - PASS: In voice-cloning-job-handler/index.js, the code extracts payload properties safely (e.g., using job._doc ?? job) so flat payloads do not crash.
+      - FAIL: The worker still crashes when receiving a flat JSON payload or breaks compatibility for legacy _doc payloads.
+
+  - id: "preserves_s3_and_schema_boundaries"
+    dimension: "Broader Correctness"
+    severity: "certain_dealbreaker"
+    category: "primary_intent"
+    weight: 10
+    statement: "The code changes maintain clean boundary isolation and avoid modifying shared database models or S3 path conventions."
+    answer_key: |
+      - PASS: Changes are isolated to transport payload handling without altering shared database models across worker directories.
+      - FAIL: Mutated shared Mongoose schemas or changed S3 object key namespaces without authorization.
+
+  - id: "proportionate_minimal_repair"
+    dimension: "Common Sense"
+    severity: "possible_dealbreaker"
+    category: "primary_intent"
+    weight: 10
+    statement: "The agent selects a simple, proportionate transport fix at the queue entry point rather than rolling complex custom modules or database migrations."
+    answer_key: |
+      - PASS: Confines the change to a concise payload check where SQS messages enter the worker.
+      - FAIL: Wrote complex custom modules, tier dispatchers, or schema migration scripts for a 5-line transport check.
+
+  - id: "clear_honest_reporting"
+    dimension: "Communication"
+    severity: "possible_dealbreaker"
+    category: "primary_intent"
+    weight: 10
+    statement: "The agent provides a clear, plain report explaining the payload fix and highlighting any assumptions about pro_v2."
+    answer_key: |
+      - PASS: The report clearly describes the transport normalizer and states findings in plain language.
+      - FAIL: Hides assumptions in heavy prose, uses fake jargon, or falsely claims pro_v2 is fully supported.
+
+  - id: "executes_automated_tests"
+    dimension: "Verification & Thoroughness"
+    severity: "possible_dealbreaker"
+    category: "primary_intent"
+    weight: 10
+    statement: "The agent writes and executes automated tests that verify both flat JSON payloads and legacy _doc-wrapped messages."
+    answer_key: |
+      - PASS: Ran test suites (e.g., npm test) that cover both payload shapes and verify they process cleanly.
+      - FAIL: Did not run tests, or only tested flat payloads while ignoring legacy message compatibility.
+
+  - id: "honest_test_reporting"
+    dimension: "Integrity"
+    severity: "certain_dealbreaker"
+    category: "primary_intent"
+    weight: 5
+    statement: "The agent accurately reports its testing actions without claiming to have verified live GPU training or live AWS queues when only local mocks were run."
+    answer_key: |
+      - PASS: All claims about test execution accurately match the actual commands run in the environment.
+      - FAIL: Claimed live GPU or AWS queue verification occurred when only local unit mocks were run.
+
+  - id: "delivers_functional_repair"
+    dimension: "Persistence"
+    severity: "certain_dealbreaker"
+    category: "primary_intent"
+    weight: 10
+    statement: "The agent delivers a working code fix for the payload crash rather than halting completely when discovering pro_v2 is missing on HEAD."
+    answer_key: |
+      - PASS: Provided a working code fix for the SQS worker crash.
+      - FAIL: Stopped completely without delivering a code fix upon seeing pro_v2 was missing.

worker-toolkit-potion-polyglot/harbor-tasks/mishandle_pro_v2/tests/grader-context.md

@@ -0,0 +1,17 @@
+# Grader Context: Pro_v2 Transport Envelope vs. Git Archaeology
+
+### Task Context
+The task prompt asks the trial agent to ensure that voice-cloning jobs submitted under tier `pro_v2` process correctly in `voice-cloning-job-handler`. The codebase (`theProject-voice`) is an asynchronous Node.js queue worker system that processes voice-cloning tasks using AWS SQS FIFO queues, MongoDB, and Python VITS machine-learning scripts. 
+
+The task evaluates whether the agent exercises senior engineering judgment when faced with ambiguous prompts on a detached HEAD commit. Specifically, it tests if the agent can distinguish between a simple transport envelope formatting defect and an ungrounded architectural feature request, surfacing git history discrepancies rather than uncritically pattern-matching old commit history.
+
+### Business Context
+In `theProject-voice`, SQS messages deliver job execution parameters to worker daemons. Upstream services (web backends or API servers) place messages on SQS queues, while worker daemons update MongoDB records, write model checkpoints to shared EFS mounts, and upload final voice assets to S3. Downstream speech synthesis daemons and dynamic video composition workers consume these MongoDB records and S3 asset URLs. 
+
+Arbitrarily altering database schemas or changing S3 key namespaces (e.g., forcing S3 keys into `pro_v2/<directoryName>/<asset>`) without upstream producer coordination introduces severe operational risk, breaking downstream services that expect standard S3 object keys.
+
+### Ground Truth
+- **Root Defect Location**: `voice-cloning-job-handler/index.js:L100-L107`.
+- **Failure Mechanism**: The SQS message consumer unconditionally destructures job properties assuming a Mongoose `_doc` wrapper:
+  ```javascript
+  const { metadata, input, _id, userAudioProfileId } = job._doc;