docs: retire stale model-judge guidance

CHANGELOG.md

@@ -7,6 +7,12 @@ documented with migration notes.
 
 ## [Unreleased]
 
+### Fixed
+
+- Marked extraction-era technical-design Promptfoo follow-up docs as historical so they do not
+  override current two-config pointwise model-judge guidance.
+- Replaced a stale release-process dependency example with a version placeholder.
+
 ### Planned
 
 - Additional docs for suite-specific adoption.

README.md

@@ -192,6 +192,7 @@ v0.1.0
 v0.1.1
 v0.1.2
 v0.1.3
+v0.1.4
 v0.1.5
 v0.2.0
 ```

docs/design/README.md

@@ -8,5 +8,6 @@ Read:
 - [`decisions.md`](decisions.md) - append-only decision log.
 - [`bootstrap-cli.md`](bootstrap-cli.md) - init/scaffold/doctor/list-cases design.
 - [`consumer-integration.md`](consumer-integration.md) - how consumer repos adopt the package.
-- [`technical-design-follow-up.md`](technical-design-follow-up.md) - required technical-design Promptfoo variable fix.
+- [`technical-design-follow-up.md`](technical-design-follow-up.md) - historical pre-v0.1.5
+  technical-design Promptfoo follow-up; do not use as current integration guidance.
 - [`skills.md`](skills.md) - agent-facing skill design.

docs/design/technical-design-follow-up.md

@@ -1,64 +1,22 @@
-# Technical-design follow-up
+# Historical technical-design follow-up
 
-## Problem
+This note records a pre-v0.1.5 follow-up from the original eval-kit extraction. It is not current
+integration guidance and should not be used to plan new consumer work.
 
-After extracting eval-kit, `technical-design` consumes the shared package successfully for deterministic evals. The deterministic smoke path passed, but the model-assisted Promptfoo path needs compatibility hardening.
+Current model-judge guidance lives in:
 
-The shared generic prompts use generic variables:
+- [`../guides/model-assisted-evals.md`](../guides/model-assisted-evals.md)
+- [`../guides/model-judge-calibration-reporting.md`](../guides/model-judge-calibration-reporting.md)
+- [`../guides/consumer-integration.md`](../guides/consumer-integration.md)
 
-- generation: `source_material`, `candidate_instructions`, `output_format`;
-- pointwise: `source_material`, `case_rubric`, `expected_items`, `candidate_path`, `candidate`;
-- pairwise: `source_material`, `case_rubric`, `expected_items`, `candidate_a`, `candidate_b`, randomization fields.
+## Current policy
 
-Technical-design adapters may still return older variables such as `product_md`, `source_map_md`, `source_facts`, and DDD-specific template keys. That can leave bundled prompts with missing variables.
+- Keep `evals/eval-kit.config.json` deterministic and CI-safe.
+- Put manual pointwise judge coverage in `evals/eval-kit.model-judge.config.json`.
+- Use consumer-owned prompts, expected items, and adapter semantics for the meaning being judged.
+- Keep provider-backed commands manual and outside `pnpm check` and CI.
+- Treat pairwise and generation lanes as separate, opt-in future work with explicit configs and human
+  calibration.
 
-## Required fix
-
-In `technical-design/evals/adapter.mjs`:
-
-1. Update `resolveGenerationVars` to return:
-
-```js
-{
-  source_material: [
-    "# Product Brief",
-    productText,
-    "# Source Map",
-    sourceMapText,
-  ].join("\n\n"),
-  candidate_instructions: [
-    authorSkill,
-    technicalDesignTemplate,
-    boundedContextTemplate,
-    enforcementMapTemplate,
-    dddEvalExpectations,
-  ].join("\n\n"),
-  output_format: "Markdown technical design",
-}
-```
-
-2. Update `resolvePointwiseVars` to return `source_material`. Keep `source_facts` temporarily only if local code still needs it.
-
-3. Update `resolvePairwiseVars` to return `source_material` and `expected_items` in the shape expected by the shared prompt.
-
-4. Add tests that call these adapter exports and assert required keys exist.
-
-5. Run the fast/offline gate and local on-demand deterministic smoke:
-
-```bash
-pnpm check
-pnpm eval:case -- --case case-tiny-laundry-pickup-v1 --candidate evals/cases/case-tiny-laundry-pickup-v1/reference-design.md --run-id verify-shared-eval-kit
-```
-
-Manual/advisory if local auth exists:
-
-```bash
-pnpm eval:generate -- --case case-tiny-laundry-pickup-v1 --model <model> --provider openai --effort medium --run-id verify-generation-vars
-pnpm eval:judge:coverage -- --case case-tiny-laundry-pickup-v1 --candidate evals/cases/case-tiny-laundry-pickup-v1/reference-design.md --model <model> --provider openai --effort medium --run-id verify-pointwise-vars
-```
-
-## Alternative
-
-If technical-design should keep fully specialized prompts, add local prompt templates under `technical-design/evals/prompts/` and configure `prompt_templates` in `evals/eval-kit.config.json`.
-
-Preferred path: align the adapter to generic variable names and keep specialized details inside `candidate_instructions` and source material.
+The old extraction-era advice to wire generation and pairwise variables for `technical-design` is no
+longer required by the standard pointwise model-judge lane.

docs/guides/technical-design-integration-fix.md

@@ -1,67 +1,20 @@
-# Fixing the technical-design integration
+# Historical technical-design integration fix
 
-This guide is for the follow-up PR in `agentic-workflow-kit/technical-design`.
+This guide is retained as extraction-era history only. It described a `technical-design` follow-up for
+early eval-kit tags and should not be used as current guidance.
 
-## Current issue
+For current consumer setup, use:
 
-`technical-design` consumes `@agentic-workflow-kit/eval-kit` through `github:agentic-workflow-kit/eval-kit#v0.1.0`. Deterministic evals pass, but model-assisted Promptfoo commands need variable alignment with the shared generic prompts.
+- [`consumer-integration.md`](consumer-integration.md)
+- [`model-assisted-evals.md`](model-assisted-evals.md)
+- [`model-judge-calibration-reporting.md`](model-judge-calibration-reporting.md)
 
-## Fix steps
+Current consumers should keep the two-config pattern:
 
-1. Read `technical-design/AGENTS.md` and `technical-design/evals/README.md`.
-2. Inspect `evals/eval-kit.config.json`.
-3. Inspect `evals/adapter.mjs`.
-4. Update generation, pointwise, and pairwise variable resolvers to include the generic keys expected by eval-kit prompts.
-5. Add unit tests for adapter variable contracts.
-6. Run the fast/offline `pnpm check` gate and the local on-demand deterministic smoke.
+- `evals/eval-kit.config.json` remains deterministic and CI-safe.
+- `evals/eval-kit.model-judge.config.json` owns manual pointwise judge coverage.
+- `eval:judge:*` scripts target the model-judge config.
+- Provider-backed commands remain manual and outside `pnpm check` and CI.
 
-## Required keys
-
-Generation:
-
-```text
-source_material
-candidate_instructions
-output_format
-```
-
-Pointwise:
-
-```text
-source_material
-case_rubric
-expected_items
-candidate_path
-candidate
-```
-
-Pairwise:
-
-```text
-source_material
-case_rubric
-expected_items
-candidate_a
-candidate_b
-randomization_method
-randomization_seed
-original_order
-candidate_order
-```
-
-## Verification
-
-Required:
-
-```bash
-pnpm install --frozen-lockfile
-pnpm check
-pnpm eval:case -- --case case-tiny-laundry-pickup-v1 --candidate evals/cases/case-tiny-laundry-pickup-v1/reference-design.md --run-id verify-shared-eval-kit
-```
-
-Manual/advisory if auth is available:
-
-```bash
-pnpm eval:generate -- --case case-tiny-laundry-pickup-v1 --model <model> --provider openai --effort medium --run-id verify-generation-vars
-pnpm eval:judge:coverage -- --case case-tiny-laundry-pickup-v1 --candidate evals/cases/case-tiny-laundry-pickup-v1/reference-design.md --model <model> --provider openai --effort medium --run-id verify-pointwise-vars
-```
+Do not revive generation or pairwise wiring from this historical note as required work. Generation
+and pairwise lanes require separate explicit configs, consumer-owned semantics, and human calibration.

docs/reference/release-process.md

@@ -6,7 +6,7 @@ Consumers depend on tags like:
 
 ```json
 {
-  "@agentic-workflow-kit/eval-kit": "github:agentic-workflow-kit/eval-kit#v0.1.0"
+  "@agentic-workflow-kit/eval-kit": "github:agentic-workflow-kit/eval-kit#vX.Y.Z"
 }
 ```