chore(evals): align model judge scripts

evals/model-judge/README.md

@@ -24,6 +24,13 @@ judge semantics.
 
 ## Manual Commands
 
+This repo follows the eval-kit two-config standard:
+
+- `evals/eval-kit.config.json` stays deterministic, default, and CI-safe.
+- `evals/eval-kit.model-judge.config.json` enables the manual pointwise judge lane.
+- Pairwise judging remains deferred unless a future calibrated comparison workflow adds a separate
+  config.
+
 Use the deterministic suite first:
 
 ```bash
@@ -33,15 +40,23 @@ pnpm eval:validate-fixtures
 pnpm eval:case -- --case case-contract-grounded-prd-v1 --candidate evals/cases/case-contract-grounded-prd-v1/candidate-good.md --run-id local-det-contract-good
 ```
 
+Validate the manual judge config before running provider-backed judging:
+
+```bash
+pnpm eval:judge:doctor
+pnpm eval:judge:list
+pnpm eval:judge:validate-fixtures
+```
+
 Run pointwise judging only when you explicitly want a local model-assisted review:
 
 ```bash
-pnpm eval:judge:coverage -- --case case-contract-grounded-prd-v1 --candidate evals/cases/case-contract-grounded-prd-v1/candidate-good.md --model <model> --provider openai --effort medium --run-id manual-pointwise-contract-good
+pnpm eval:judge:coverage -- --case case-contract-grounded-prd-v1 --candidate evals/cases/case-contract-grounded-prd-v1/candidate-good.md --model gpt-5.5 --provider openai --effort medium --run-id manual-pointwise-contract-good
 ```
 
-This command can call a provider through Promptfoo. It expects a deliberate local Promptfoo setup
-that provides `node_modules/.bin/promptfoo`; the default repo install and `pnpm check` do not require
-Promptfoo. Do not run the command in CI, broad checks, or unattended automation. Do not commit raw
+This command can call a provider through Promptfoo. `gpt-5.5` is an account-supported example for
+local calibration, not a permanent model policy; use any approved model supported by the current
+account. Do not run the command in CI, broad checks, or unattended automation. Do not commit raw
 result bundles unless they have been reviewed and intentionally curated.
 
 ## Calibration

package.json

@@ -28,11 +28,14 @@
     "eval:list": "eval-kit list-cases --config evals/eval-kit.config.json",
     "eval:validate-fixtures": "eval-kit validate-fixtures --config evals/eval-kit.config.json",
     "eval:case": "eval-kit run-case --config evals/eval-kit.config.json",
+    "eval:judge:doctor": "eval-kit doctor --config evals/eval-kit.model-judge.config.json",
+    "eval:judge:list": "eval-kit list-cases --config evals/eval-kit.model-judge.config.json",
+    "eval:judge:validate-fixtures": "eval-kit validate-fixtures --config evals/eval-kit.model-judge.config.json",
     "eval:judge:coverage": "eval-kit judge-coverage --config evals/eval-kit.model-judge.config.json",
     "eval:report": "eval-kit report --config evals/eval-kit.config.json"
   },
   "devDependencies": {
-    "@agentic-workflow-kit/eval-kit": "github:agentic-workflow-kit/eval-kit#v0.1.2",
+    "@agentic-workflow-kit/eval-kit": "github:agentic-workflow-kit/eval-kit#v0.1.3",
     "prettier": "^3.6.2",
     "promptfoo": "^0.121.17"
   }