Answer a payer audit

# Assemble the Decosa payer audit response on this machine

You are setting up a self-hosted "Answer a payer audit" tool on this Linux machine for a practice (or the billing service
working for it) that has received a payer's medical-records request or post-payment audit. It reads the auditor's letter,
the charts and the payer's documentation policy, and returns:
- the respond-by date, worked out in code from the letter's own words, with the arithmetic and the letter line;
- each claim matched to its note, and each requirement of the payer's policy marked found (with chart, page and line and
  the words) or missing, the weak claims listed first;
- a draft cover letter whose sentences are each checked against the letter and the index;
- a bookmarked PDF response packet (letter, index, the records as kept), a worksheet, a CSV and a 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:**
- Charts are protected health information. Keep everything on this machine: the model route stays local (`direct`), and
  nothing goes to a hosted service.
- This is a review and drafting aid, not legal, coding or compliance advice. It never suggests adding to, changing or
  back-dating a record, and it never touches a payer or auditor portal. A person reads the weak claims with their
  compliance lead or counsel, decides what to send and signs.

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/payer-audit.zip (7 KB, 11 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 payer-audit` (the api image carries the same bundle under /app/rehearsal/payer-audit/;
   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 payer-audit --bundle payer-audit.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 respond-by date is the letter date + 10 calendar days (no model)", "the run gives the same date", "12 claims are read from the letter"). 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-payer-audit/.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 records, e.g. Example Counseling compliance>"
```

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

```yaml
name: decosa-payer-audit
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: "400000"        # per session; a 140-claim audit used about 71,000 generated tokens
      DECOSA_SESSION_TTL_S: "28800"
      DECOSA_PAYERAUDIT_MAX_CONCURRENT: "2"
      DECOSA_PAYERAUDIT_WORKERS: "4"            # claims checked in parallel
      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

The deadline needs no model (the sample letter is synthetic):

```bash
API=localhost:8445
curl -s "$API/payer-audit/samples?id=therapist-12" | jq '{letter, today: "2026-09-24"}' \
  | curl -s $API/payer-audit/deadline -H 'content-type: application/json' -d @- | jq '{respond_by, days_left, rows: [.rows[].arithmetic]}'
```

Pass if `respond_by` is `2026-10-01` (the letter is dated 21 Sep 2026 and gives ten calendar days).

Then the 12-claim sample (a made-up practice; the policy is BCBSM's published individual-therapy documentation list):

```bash
TOKEN=$(curl -s $API/demo/session -H 'content-type: application/json' -d '{"vertical":"payer-audit"}' | jq -r .token)
curl -s $API/payer-audit/run -H "authorization: Bearer $TOKEN" -H 'content-type: application/json' \
  -d '{"sample_id":"therapist-12"}' > /tmp/pa.json
jq -c '{verdict: .summary.verdict, weak: [.claims[] | select(.status=="weak") | .n], first: .summary.weak_first[0:4], respond_by: .deadline.respond_by, receipts}' /tmp/pa.json
jq -r '.claims[] | select(.status=="weak") | "\(.n)\t\(.reasons | join("; "))"' /tmp/pa.json
jq '{record}' /tmp/pa.json | curl -s $API/record/verify -H 'content-type: application/json' -d @- | jq '{ok, summary}'
curl -s "$API$(jq -r .export.packet /tmp/pa.json)" -H "authorization: Bearer $TOKEN" -o /tmp/pa-packet.pdf && head -c 8 /tmp/pa-packet.pdf
```

Pass if:
- claims 4, 5, 6 and 10 are weak and listed first (no start/stop times; an empty interventions field; times only in an
  addendum dated after the request; no start/stop times), and the other 8 are supported;
- every found element has a cite like `C3 p2:L14`; the record verifies (`ok: true`); the packet starts with `%PDF`;
- the receipts in `/tmp/pa.json` are signed by this box (`GET /receipts/<id>` shows `"status": "attested"`).

## 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`.
- `POST /payer-audit/run` takes `{letter: {text, date?, received_date?}, charts: [{id?, title?, text}], pack | policy,
  claims?, practice?, mode?: respond|self_audit, sample_size?, seed?}`. Charts are text; split pages with form feeds or
  `--- Page 2 ---` lines. Leave `claims` out to read the claim list from the letter.
- Use a pack for your payer's policy (`GET /payer-audit/packs` shows the shape: each requirement quoted from the payer's
  published text, with the source). Pasting the policy text instead makes the model read the requirements; review them.
- Exports: `GET /payer-audit/runs/<run_id>/export?format=packet|worksheet|csv|letter|record`, kept in memory one hour.
- The server stores nothing else. Keep the signed record with the audit file; anyone can re-check it with
  `POST /record/verify` against the key at `GET /attest/signing-key`.
- Keep the API on 127.0.0.1. For other users on the LAN, put a TLS reverse proxy with authentication in front.

## 6. Keep the direct route

`DECOSA_LLM_ROUTE=gateway` would send prompts to the hosted Decosa
API, and those prompts contain the charts. Never use it for
real patients.

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