Review a claim file for conduct

# Assemble the Decosa claims-file conduct pack on this machine

You are setting up a self-hosted claims-file conduct review on this Linux machine for an insurer's claims-quality team or a
TPA. It reads a claim file (first notice of loss, adjuster notes, letters, estimates) and the policy wording. It returns a
conduct checklist:
- acknowledgement, decision and payment deadlines computed from the file's own dates, under the NAIC model (Model #900 and
  #902), California (10 CCR 2695) or Texas (Insurance Code ch. 542);
- each denial reason checked for a cited provision, and grounded in the policy wording;
- the Department of Insurance review notice, where the state requires it;
- typed judgments on misrepresenting the policy, low offers and missing investigation;
- a record of any AI system that touched the claim.

Everything is sealed in a signed, hash-chained file review. A person's sign-off is a second signed record.

Work step by step, show me each command before running anything that needs sudo, and stop if a check fails.

**Before anything else, remind me:**
- Claim files hold personal, financial and often health data. Keep everything on this machine: the model route stays
  local (`direct`), and nothing goes to a hosted service.
- This is triage for a claims-quality reviewer. It never says a file is compliant, it does not decide coverage, and it is
  not legal advice. Every flag needs a person to read the file.

Repeat both points in your final summary.

## Step 0: set up with a coding agent, rehearse on mock data, then go private

This prompt is for a coding agent running on the machine that will host the service. We recommend Claude Code with
Claude Opus 5.5; any capable coding agent works. Work in this order:

1. Set up on mock data only. During the whole setup you (the agent) work with the synthetic sample bundle below and
   nothing else. Do not ask me for real data, and do not open, read, list or copy files that hold real data, even to
   "test with something realistic".
2. Rehearse. When the steps below are done and the service is healthy, fetch the mock-data bundle for this tool,
   https://decosa.ai/samples/claims-conduct-pack.zip (4 KB, 8 checks, synthetic or openly licensed: see `licence` in expected.json),
   show me what is in it, and run the rehearsal against the local API:
   `docker compose exec api python scripts/rehearse.py claims-conduct-pack` (the api image carries the same bundle under /app/rehearsal/claims-conduct-pack/;
   with no key set, the script asks the local API for a short demo token). From a decosa-api checkout instead:
   `python scripts/rehearse.py claims-conduct-pack --bundle claims-conduct-pack.zip --base-url http://127.0.0.1:<PORT>`.
   It sends the mock inputs to the local API and prints PASS or FAIL for each expected property (for example: "the late acknowledgement is flagged", "its deadline is 15 calendar days after the first notice", "the denial reason that misstates Exclusion 4 is flagged"). Show me
   the full output. Every check must pass. If one fails, fix the install and run it again; never edit `expected.json`
   to make a check pass.
3. Stop there. Once the rehearsal passes, tell me, and I will run my own data against the local API myself, on this
   machine.

For the person running this: a coding agent that runs in the cloud sees everything in its context, including files it
reads, command output and anything pasted into the chat. Keep real data out of the chat and out of anything the agent
can read. Switch to your own data only after the rehearsal has passed and the agent's work is done.

## What you are building

| service | image | model | port |
|---|---|---|---|
| `llm` | `${DECOSA_REGISTRY}/decosa-llm:0.1.0` (vLLM 0.29.0, `vllm/vllm-openai@sha256:c2914767605584b6d8f45686b82de173ecc99e781897aa3d0a66dacd72c51ae1`) | `nvidia/Qwen3.8-27B-NVFP4` @ `482ca0f3832238542f8f5295dde86b5f22711d80`, Apache-2.0 | internal 8000 |
| `api` | `${DECOSA_REGISTRY}/decosa-api:0.1.0` (no GPU) | none | `127.0.0.1:8445` |

## 1. Check the GPU, driver and Docker

1. Run `nvidia-smi`. I need one NVIDIA GPU with at least 48 GB and driver 580 or newer.
   - Blackwell (RTX PRO 6000, B200): use the defaults below (NVFP4). This is the measured setup.
   - Hopper (H100/H200) or 48 GB Ada/L40S: set `LLM_MODEL=Qwen/Qwen3.8-27B-FP8` and `LLM_REVISION=main`. On a 48 GB card, also
     set `LLM_MAX_LEN=32768`. Not measured.
   - Under 48 GB: stop and tell me it will not fit.
2. Check `docker --version`, `docker compose version` and `docker run --rm --gpus all ubuntu nvidia-smi`. If Docker or the
   NVIDIA Container Toolkit is missing, install them from the official Docker and NVIDIA repositories. Then run
   `sudo nvidia-ctk runtime configure --runtime=docker` and restart Docker.
3. Confirm about 60 GB of free disk.

## 2. Get the images

The images are **on request** while self-host is in early access: ask at https://decosa.ai/contact?topic=self-host, and Decosa sends the registry (set it as `DECOSA_REGISTRY`), pull access and the compose file. Try `docker pull ${DECOSA_REGISTRY}/decosa-{llm,api}:0.1.0`.

If a pull fails, build from source once the `decosa-api` source is published:
- clone it;
- in the clone, run `docker build -f docker/api/Dockerfile -t ${DECOSA_REGISTRY}/decosa-api:0.1.0 .`;
- run `docker compose build llm` from its compose file.

If neither works, stop and tell me.

## 3. Write the compose file

Create `~/decosa-claims/.env`:

```bash
DECOSA_TAG=0.1.0
DECOSA_GPU=0
LLM_MODEL=nvidia/Qwen3.8-27B-NVFP4
LLM_REVISION=482ca0f3832238542f8f5295dde86b5f22711d80
LLM_MAX_LEN=65536
LLM_GPU_UTIL=0.85
DECOSA_SIGNER_NAME="<who signs these file reviews, e.g. Example Mutual claims quality>"
```

Create `~/decosa-claims/docker-compose.yml` with exactly these services:

```yaml
name: decosa-claims
x-health: &health
  interval: 15s
  timeout: 5s
  retries: 5
services:
  llm:
    image: ${DECOSA_REGISTRY}/decosa-llm:${DECOSA_TAG}
    deploy: { resources: { reservations: { devices: [ { driver: nvidia, device_ids: ["${DECOSA_GPU:-0}"], capabilities: [gpu] } ] } } }
    ipc: host
    restart: unless-stopped
    volumes: [hf-cache:/root/.cache/huggingface]
    command: ["${LLM_MODEL}", "--revision", "${LLM_REVISION}", "--served-model-name", "qwen3.8-27b",
              "--language-model-only", "--max-model-len", "${LLM_MAX_LEN}", "--gpu-memory-utilization", "${LLM_GPU_UTIL}",
              "--max-num-seqs", "16", "--kv-cache-dtype", "fp8_e4m3", "--speculative-config", '{"method":"mtp","num_speculative_tokens":3}',
              "--seed", "0", "--enable-force-include-usage", "--disable-uvicorn-access-log", "--host", "0.0.0.0", "--port", "8000"]
    healthcheck: { <<: *health, test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8000/health', timeout=4)"], start_period: 900s }
  api:
    image: ${DECOSA_REGISTRY}/decosa-api:${DECOSA_TAG}
    restart: unless-stopped
    depends_on: { llm: { condition: service_healthy } }
    environment:
      DECOSA_LLM_ROUTE: direct                  # local model only; receipts are signed by this box's key ("attested")
      DECOSA_LLM_URL: http://llm:8000/v1
      DECOSA_LLM_MODEL: qwen3.8-27b
      DECOSA_LOCAL_SIGNING: "on"                # Ed25519 key created at /data/attest/ed25519.pem on first start
      DECOSA_SIGNER_NAME: ${DECOSA_SIGNER_NAME}
      DECOSA_SESSIONS_PER_IP_HOUR: "1000"
      DECOSA_BUDGET_LLM_TOKENS: "200000"        # per session; one file review needs up to about 4,500 generated tokens
      DECOSA_SESSION_TTL_S: "28800"
      DECOSA_CLAIMS_MAX_CONCURRENT: "3"
      DECOSA_CORS_ORIGIN_REGEX: '^https?://(localhost|127\.0\.0\.1)(:\d+)?$$'
    ports: ["127.0.0.1:8445:8445"]
    volumes: [decosa-data:/data]
    healthcheck: { <<: *health, test: ["CMD", "python", "-c", "import urllib.request; urllib.request.urlopen('http://127.0.0.1:8445/healthz', timeout=4)"], start_period: 20s }
volumes: { hf-cache: {}, decosa-data: {} }
```

Use the named volume `decosa-data` exactly as written. A host bind mount owned by root makes the API fail on
`/data/keys.sqlite`.

Run `docker compose up -d`, then poll `docker compose ps` until both services are healthy. The LLM takes 5-10 minutes
the first time. `curl -s localhost:8445/healthz` should show `"llm": true` (`asr` is false here, which is fine).

## 4. Smoke test

```bash
API=localhost:8445
TOKEN=$(curl -s $API/demo/session -H 'content-type: application/json' -d '{"vertical":"claims-conduct-pack"}' | jq -r .token)
curl -s $API/claims/check -H "authorization: Bearer $TOKEN" -H 'content-type: application/json' \
  -d '{"sample_id":"ca-auto-planted"}' > /tmp/claims.json
jq '{counts, receipts: (.receipts|length), statuses: [.receipts[].status] | unique}' /tmp/claims.json
jq -r '.checklist[] | "\(.status)\t\(.id)\t\(.reason)"' /tmp/claims.json
jq '{record}' /tmp/claims.json | curl -s $API/record/verify -H 'content-type: application/json' -d @- | jq '{ok, summary}'
```

This sample is a synthetic California auto denial (fictional insurer Harbor Oak Mutual). Pass if:
- `ack` is `flag`, and its reason says it was acknowledged 29 days after notice (deadline 2026-06-03);
- `reason:1` is `flag`, and `.grounding.spans[0].quote` contains "organized race or speed contest";
- `review_notice` is `flag`;
- every receipt has `"status": "attested"`;
- the record verifies (`ok: true`).

Then run the clean file:

```bash
curl -s $API/claims/check -H "authorization: Bearer $TOKEN" -H 'content-type: application/json' \
  -d '{"sample_id":"ca-home-clean"}' | jq '.counts'
```

Pass if `flag` is 0.

A person's sign-off makes a second signed record:

```bash
jq '{record, reviewer: {name: "QA reviewer", role: "claims quality"}, decisions: [{id: "ack", decision: "confirm"}]}' /tmp/claims.json \
  | curl -s $API/claims/signoff -H "authorization: Bearer $TOKEN" -H 'content-type: application/json' -d @- | jq '.record.statement.review_of'
```

Pass if it prints the first record's root hash.

## 5. Point the app at the local API

- Base URL: `http://localhost:8445`. For a web app, set `NEXT_PUBLIC_DECOSA_API=http://localhost:8445`. Add other origins to
  `DECOSA_CORS_ORIGINS`.
- `POST /claims/check` takes `{claim: {id?, line: auto|homeowners|other, state: NAIC|CA|TX}, documents: [{kind, title?,
  date?, text}], policy: {title?, text}, timeline?, ai_use?}`. It returns JSON, or streams Server-Sent Events when asked
  with `Accept: text/event-stream`.
- `GET /claims/info` lists every rule with its source and the date it was read, the deadlines, the conventions and what
  is not checked.
- The server stores nothing. Keep each signed file review (JSON) with the claim file. Anyone can re-check it with
  `POST /record/verify` against the key at `GET /attest/signing-key`.
- If the claims system exports a dated activity log, send it as `timeline`: those dates are used as given, and the model
  does not read them from the notes.
- Keep the API on 127.0.0.1. For other users on the LAN, put a TLS reverse proxy with authentication in front and set
  `DECOSA_TRUSTED_PROXIES`.

## 6. Keep the direct route

`DECOSA_LLM_ROUTE=gateway` would send prompts to the hosted Decosa
API, and those prompts contain the claim file and the policy.
Never use it for real claims. At most, use it for synthetic training material.

Finish with a summary: what is running, the health output, the smoke-test results, and the reminders above.