Investigate a Reg E dispute

# Assemble the Decosa Reg E dispute investigation file on this machine

You are setting up a self-hosted Regulation E dispute file on this Linux machine for a bank, credit union or fintech
disputes team. It reads a consumer's notice of an electronic fund transfer error (a chat, call transcript, form or
letter), the account's transactions as CSV, and the investigation notes, events and results letter, and returns:
- every Reg E clock computed in code with the rule cited (12 CFR 1005.11, 1005.18(e) for prepaid, 1005.33 for
  remittances): notice timeliness, 10 or 20 business days, provisional credit and its notice, 45 or 90 days, the report
  within 3 business days and the correction within 1, plus a timeline;
- the notice's required elements, each quoted word for word;
- a checklist of the file's gaps, including a check that a no-error letter explains the findings for this case and
  notes the consumer's right to request the documents relied on;
- a signed, hash-chained investigation 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:**
- Consumer notices and account data are nonpublic personal information. Keep everything on this machine: the model
  route stays local (`direct`), and nothing goes to a hosted service.
- It never decides a claim. A named investigator enters the determination; the tool records it as entered. It is not
  legal advice, and it does not cover card-network rules, Regulation Z or state law.

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/reg-e-dispute-file.zip (6 KB, 14 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 reg-e-dispute-file` (the api image carries the same bundle under /app/rehearsal/reg-e-dispute-file/;
   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 reg-e-dispute-file --bundle reg-e-dispute-file.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 provisional credit notice is due two business days after the credit (4 Sep 2026) and came late (no model)", "the May charge was reported more than 60 days after its statement", "a debit card online purchase gets the 90-day period"). 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-rege/.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 files, e.g. Example Bank disputes operations>"
```

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

```yaml
name: decosa-rege
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 needs up to about 5,500 generated tokens
      DECOSA_SESSION_TTL_S: "28800"
      DECOSA_REGE_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

The clocks need no model:

```bash
API=localhost:8445
curl -s $API/rege/clocks -H 'content-type: application/json' -d '{"sample_id":"cnp-generic-denial"}' \
  | jq -c '[.rows[] | {id, due, status}]'
```

Pass if `provisional_credit_notice` is `missed` (due `2026-09-04`), `notice:T-0512` is `late_notice` (due `2026-08-01`)
and `determination` is due `2026-11-18` (90 days: a debit card online purchase).

Then the files (synthetic disputes):

```bash
TOKEN=$(curl -s $API/demo/session -H 'content-type: application/json' -d '{"vertical":"reg-e-dispute-file"}' | jq -r .token)
for s in cnp-generic-denial p2p-scam-consumer-sent p2p-takeover-open; do
  curl -s $API/rege/file -H "authorization: Bearer $TOKEN" -H 'content-type: application/json' \
    -d "{\"sample_id\":\"$s\"}" > /tmp/rege-$s.json
  jq -c '{status: .file_status.status, cannot_close: (.file_status.cannot_close != null), outcome: .determination.outcome,
          gaps: .file_status.gaps, receipts: (.receipts|length), statuses: ([.receipts[].status]|unique)}' /tmp/rege-$s.json
done
jq '{record}' /tmp/rege-cnp-generic-denial.json | curl -s $API/record/verify -H 'content-type: application/json' -d @- | jq '{ok, summary}'
```

Pass if:
- `cnp-generic-denial` is `gaps` with `cannot_close` true, the outcome `no_error` exactly as entered, and gaps including
  `explanation`, `right_to_documents` and `documents_relied_on`;
- `p2p-scam-consumer-sent` is `gaps` (the determination came late with no provisional credit, and the claim was held
  for an affidavit) and `.coverage[0].id` is `consumer_sent`;
- `p2p-takeover-open` is `open` with outcome `null`: no determination was entered, so none exists;
- every receipt has `"status": "attested"`, and the record verifies (`ok: true`).

You can also run the rehearsal bundle from the decosa-api source: `python scripts/rehearse.py reg-e-dispute-file
--base-url http://localhost:8445` (14 checks).

## 5. Point your case system 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 /rege/file` takes `{regime?: account|prepaid|remittance, notice: {text, received, channel?, written_confirmation?},
  account?: {first_deposit?, reg_t?}, transactions_csv, disputed_ids?, investigation: {investigator?, notes?, events?,
  documents_relied_on?, letter?, determination?: {outcome, by, date}}, closed_days?, today?}`. It returns JSON, or streams
  Server-Sent Events when asked with `Accept: text/event-stream`. `POST /rege/clocks` takes the same dates without text
  and calls no model.
- The CSV needs `id,date,amount,type,description`; add `statement_sent` (or `pending`) so the 60 days can be counted.
  Negative amounts are debits.
- Business days default to weekdays minus the Federal Reserve Banks' holidays. Reg E counts the institution's own
  business days (12 CFR 1005.2(d)): list any other day you are closed in `closed_days`.
- `GET /rege/info` lists each rule with its citation, link and the exact words relied on.
- The server stores nothing. Keep each signed record (JSON) with the dispute for at least two years
  (12 CFR 1005.13(b)(1)). 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 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 consumer's notice and
account data. Never use it for real disputes. 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.