Release v1.1.0: fix payer eligibility and published tool paths

CHANGELOG.md

@@ -0,0 +1,56 @@
+# Changelog
+
+## 1.1.0 — 2026-10-06
+
+### Fixes
+
+- Restore the published tool-reference path in the Docker image. The directory
+  alias `/opt/healthverse-task-assets` points to `/opt/chi-bench-task-assets`, so
+  existing task instructions see the selected per-task tool reference.
+- Allow a completed payer intake with failed member eligibility to end in the
+  existing `denied` state with reason `ELIG_TERMED`. The agent generates and
+  delivers a denial letter. Approval and clinical routing are blocked after a
+  failed eligibility check.
+- Add coverage status and known coverage dates to the existing request-form
+  template. The data repair restores pa_t035's termination date from the original
+  synthesis facts and replaces its unreachable `requirement_checked` expectation
+  with denial and a denial letter. No separate evidence document is added.
+- Provide scripts to repair downloaded pa_t035 task copies and regenerate the
+  25 saved payer request forms and their marathon copies before rebuilding.
+  Remove rendering timestamps from regenerated PDFs so repeat runs preserve
+  their hashes when content is unchanged.
+- Align the local competition Rules and Overview references with the published
+  Kaggle pages.
+
+### Upgrade
+
+After obtaining handbook access and downloading the task data:
+
+```bash
+uv sync --extra dev
+uv run playwright install chromium
+uv run python scripts/repair_pa_t035.py --data-dir data
+uv run python scripts/regenerate_payer_request_forms.py --data-dir data
+uv run cb data verify
+uv run cb docker build
+```
+
+On Linux, use `uv run playwright install --with-deps chromium` if browser system
+libraries are missing.
+
+This is runtime version **1.1.0**. The downloaded dataset remains pinned to
+**chi-bench-v1.0.0**; the scripts apply local corrections. This change does not
+publish a new Hugging Face dataset revision. Preserve earlier trial results and
+record the runtime commit and image digest for new runs.
+
+### P2P clarification
+
+The P2P workflow is unchanged. The original `MdCaseDecision` records the referral
+to P2P; the completed P2P request supplies the outcome, and
+`determination.finalize` records the final determination. No second physician
+submission is required.
+
+Among the six public payer tasks requiring P2P, t026 and t032 expect the referral
+to have `signed_off=false`. Tasks t016, t019, t031, and t036 exclude MD review
+from their scored stages. A direct-decision task may require signed physician
+review even if an agent chooses P2P instead.

CLAUDE.md

@@ -151,6 +151,20 @@ Configs:
 
 ## Things to remember
 
+- **Write Kaggle FAQs as direct participant answers in ASD-STE100 style, using the known
+  CHI-Bench dates. Omit organizer-reply citations, review notes, repeated requirements, and
+  verbatim restatements of user directions. Keep contact links in FAQs and only necessary
+  source links in announcements.
+  Assess existing rules against the intended workflow before proposing them.**
+- **Use the CHI-Bench Cup sequence: competition deadline, private evaluation, result release,
+  then the IEEE-format camera-ready report. Do not add a paper-acceptance stage.** **Why:**
+  CHI-Bench evaluates competition systems; the report follows competition results. Use its
+  own schedule and participation policy; do not import another Cup's instructions. Use
+  organizer-specified submission routes; do not infer a requirement from a portal listing.
+- **Preserve published task paths with image compatibility aliases when fixing runtime renames.**
+  **Why:** Rewriting downloaded `instruction.md` files changes task checksums. For tool references,
+  alias `/opt/healthverse-task-assets` to `/opt/chi-bench-task-assets` as a directory so per-task
+  selection and tool-name rewriting remain visible through the published path.
 - **For simple, explicitly scoped maintenance PRs, work directly on a `codex/` branch in the
   current clean checkout unless the user asks for a worktree or design spec.** **Why:** The user
   prefers direct execution over extra process for low-risk changes such as a model-default update.

README.md

@@ -89,10 +89,23 @@ uv run huggingface-cli download actava/managed-care-operations-handbook \
 
 **5. Build the Docker image** (~5 min, one-time).
 
+Apply the `pa_t035` data correction before building. It restores the coverage
+termination date in the existing request form and requires a payer denial letter.
+The script changes only local data; it does not publish a dataset revision.
+
 ```bash
+uv run playwright install chromium
+uv run python scripts/repair_pa_t035.py
+uv run python scripts/regenerate_payer_request_forms.py
 uv run cb docker build
 ```
 
+The repair scripts render PDFs on the host. On Linux, use
+`uv run playwright install --with-deps chromium` if browser system libraries are missing.
+Runtime version `1.1.0` and dataset revision `chi-bench-v1.0.0` are separate;
+keep the dataset pin at the revision you downloaded and record the runtime commit
+and image digest with your results. See [the v1.1.0 changes](CHANGELOG.md).
+
 > [!NOTE]
 > `cb` is the short alias for `chi-bench`; both commands resolve to the same CLI. Pick whichever you prefer (the rest of this README uses `cb`). If your shell already aliases `cb` to something else (e.g. a clipboard tool), use `chi-bench`. For the full command surface and flag reference, read [`docs/cli.md`](docs/cli.md).
 

configs/submission_example.yaml

@@ -5,9 +5,11 @@
 #   cb submission validate -f <your-sub>.yaml
 #   cb submission run      -f <your-sub>.yaml
 #   cb submission status   -f <your-sub>.yaml
-#   cb submission package  -f <your-sub>.yaml
+#   cb submission prepare  -f <your-sub>.yaml
 #
-# The packet (<output_root>/<id>.zip) is what you upload to the leaderboard.
+# The prepared packet is under <output_root>/packet/YYYY-MM-DD-<id>/.
+# Follow the Kaggle organizer instructions for private-final package delivery;
+# the separate evidence leaderboard accepts the same packet through a GitHub PR.
 
 schema: chi-bench/submission/v1
 

docker/Dockerfile

@@ -99,6 +99,9 @@ COPY data/marathon /opt/chi-bench/tasks/marathon
 COPY data/skills/managed-care-operations-handbook /workspace/skills/managed-care-operations-handbook
 RUN mkdir -p /opt/chi-bench-task-assets
 COPY data/care_management/shared/tool_reference.md /opt/chi-bench-task-assets/tool_reference.md
+# Published task instructions still use the healthverse asset path. Alias the
+# directory so per-task selection and tool-name rewriting remain visible there.
+RUN ln -s /opt/chi-bench-task-assets /opt/healthverse-task-assets
 
 COPY docker/entrypoint.sh /usr/local/bin/entrypoint.sh
 RUN chmod +x /usr/local/bin/entrypoint.sh

docs/architecture.md

@@ -48,6 +48,11 @@
 5. After the agent stops (success / timeout / abstain), Harbor invokes the verifier in the same container.
 6. Verifier writes `verifier/scorecard.json` and `verifier/verdicts.json`; Harbor produces `result.json`.
 
+When the payer confirms failed member eligibility, processing stops at intake.
+`determination.finalize` can use the completed intake record to set the existing
+case status `denied`, with reason `ELIG_TERMED`. The agent then generates and
+delivers the denial letter. Clinical review is not required for this result.
+
 ## Why the LLM judge needs Anthropic credits
 
 The verifier always uses `claude-opus-4-7` (configurable but paper-faithful default).

docs/cli.md

@@ -250,6 +250,29 @@ echo "$REV" > data/.chi-bench-version
 
 `cb data download` exists so the pin file is written in one step; the README documents the raw HF CLI invocation as the canonical path.
 
+For the `pa_t035` correction, run the following after downloading data and before
+building the Docker image:
+
+```bash
+uv run playwright install chromium
+uv run python scripts/repair_pa_t035.py --data-dir data
+uv run python scripts/regenerate_payer_request_forms.py --data-dir data
+```
+
+These scripts render PDFs on the host, so Playwright needs its Chromium browser
+there. On Linux, use `uv run playwright install --with-deps chromium` when system
+libraries are missing.
+
+This restores the coverage termination date in the existing request form and
+changes the expected payer result to `denied`, with a denial letter. It updates
+the task's expectation copies and its marathon copy. It does not publish data
+or change the dataset version pin.
+
+The second script updates all saved payer request forms with the same template,
+including their marathon copies. Coverage status appears on each form. Coverage
+dates appear only when the shared world provides them. Case facts and scoring
+expectations are unchanged by this regeneration.
+
 ---
 
 ## `cb submission`