# Paid-Delivery Failure Lab: agent guide

Paid-Delivery Failure Lab 0.1.2 is available from Lumen House for **USD 7.00
(700 cents), one time**. There is no recurring charge. Read the exact offer,
[customer terms](TERMS.md) and [purchaser-use license](LICENSE.md) before approving
payment. Purchaser permission allows use in the buyer's own workflows, including
commercial work and agents acting on the buyer's behalf; software redistribution
and resale are not granted.

## Decide whether the Lab fits

The Lab is a portable Python reference model and regression corpus for local,
synthetic payment and delivery recovery. It contains nine cases, three SQLite
stores, schemas, source, tests, an offline wheel, examples and a broken negative
control. There are no runtime dependencies outside Python's standard library.

It exercises recoverable local interruptions, callback idempotency, recovery with
the same identity, safe no re-dispatch when payment state is uncertain, and exact
artifact bytes. Payment acceptance, entitlement, delivery effect and acknowledgement
are distinct model facts. Retries may occur; repeated durable effects are violations.

Passing this model does not test a buyer application or provider integration.
The corpus does not cover terminal asynchronous job/payment failure or exact private HTTP
status/header/body replay. It does not establish production access control,
power-loss/rollback durability or an exactly-once guarantee. Its callback HMAC uses
a public fixture key to illustrate payload binding; it is not authentication
certification. No real paid fulfillment is verified. Demand has not been validated.

The CLI rejects credential-like environment names and installs a supplemental
network/process audit guard. This guard is not an OS sandbox. Use synthetic inputs
only; do not add live-provider adapters, customer records or payment credentials.

## Inspect the standalone free sample

[Download the free sample](free-sample.zip), or inspect its [source](free-sample/sample.py),
[fixture](free-sample/fixture.json), [expected text](free-sample/expected-output.txt),
[expected JSON](free-sample/expected-output.json), [README](free-sample/README.md),
[tests](free-sample/test_sample.py), [checksums](free-sample/SHA256SUMS) and
[sample license](free-sample/sample-license.txt). It imports none of the paid kit.
One sequential callback replay leaves two entitlements in the deliberately broken
model and one in the corrected model, which recovers the same saved receipt.
The sample's 14-test offline suite passed during release preparation.

Sample runtime: Python 3.10+ with SQLite on POSIX; Linux verified, macOS untested;
Windows unsupported. No install step or runtime package dependencies. After
extracting the sample into a local folder, use a clean process and a new,
nonexistent output directory whose parent already exists:

```sh
env -i PATH="$PATH" python3 -I -B sample.py --output ./demo-output
python3 -I -B -m unittest discover -s . -p 'test_sample.py' -v
```

The sample README describes refusal of existing paths, symlinks, changed fixtures
and credential-like environment names. The in-process guard is supplemental.

## Purchase through a compatible Stripe MPP client

Exact price: USD 7.00 / 700 cents, charged once. The purchase route is:

`GET https://lumenhouse.xyz/api/wares/paid-delivery-failure-lab-v1/purchase/card`

A compatible payment-aware client, Stripe-issued Shared Payment Token, authorized
credential and explicit buyer approval are required. This is not an ordinary
browser card-entry form. Read-only MCP discovery can inspect the offer and its
public documents; it cannot purchase, hold recovery keys or deliver paid files.
No native MCP payment or download tools are enabled for this Lab.

Before requesting a challenge, a trusted client must create and durably save a
cryptographically random 64-lowercase-hex private recovery key. Send it in
`X-Lumen-Recovery-Key` before the challenge. Keep it outside model context,
MCP tool arguments, MCP results, URLs, chat, logs and source control. Never email it.
The key belongs in trusted private client storage.

After inspecting the exact price, terms and artifact descriptor and obtaining
explicit buyer approval, that compatible client submits the single approved
`Payment` credential. After a lost, pending or uncertain response, use only
`Authorization: Bearer <saved private key>` reads on the same purchase route
to locate the existing purchase; do not pay again. Private downloads use
`GET /api/lab-downloads/{order_id}`, with Bearer authorization. Public payment
references are not delivery credentials. Never put the private key in a URL.
Offline recovery checks do not establish successful live payments or delivery.

## Verify and extract the buyer package

- Filename: `paid-delivery-failure-lab-0.1.2.zip`
- Version: `0.1.2`
- Bytes: `166390`
- SHA-256: `1fa6f6391cf1be0b8c66b2b6bf9f335a31737340ae9c77ab46a5ef9e3db6d9f3`

The paid ZIP is delivered privately and is excluded from public assets. Verify
its bytes before extraction. Extract the ZIP into a parent directory, then run
`cd paid-delivery-failure-lab-0.1.2` before the commands below. The archive
contains that single top-level folder. Run where `README.md` and `dist/` are
directly present, in a POSIX shell.

## Run offline

Python 3.10+ is the declared compatibility target, with SQLite, venv and pip.
Recorded verification: Linux x86_64 / Python 3.12.14;
64 source tests,
64 clean installed-wheel tests and
9 cases passed. The negative control exited
1 as expected. Other Python versions, macOS and
Windows are unverified by that record. Read the shipped docs/VERIFICATION.md.

```sh
python3 -m venv .venv
.venv/bin/python -m pip install --no-index --no-deps ./dist/paid_delivery_lab-0.1.2-py3-none-any.whl
.venv/bin/paid-delivery-lab version
.venv/bin/paid-delivery-lab list
.venv/bin/paid-delivery-lab run all --output ./lab-run-001
.venv/bin/paid-delivery-lab verify ./lab-run-001
.venv/bin/paid-delivery-lab negative-control --output ./lab-negative-001
```

Use a new output directory for each command that creates a run. Successful run
and verify exit 0. The negative control should exit 1 and report invariant
violations; investigate if it passes. Keep the complete run directory to inspect
or verify it later. No account, network, wallet, credentials or package download
is required to run the Lab. Consult README.md, START_HERE.md and SKILL.md.

Delivery or refund requests use buscheo@gmail.com with non-secret order
information. Never send recovery keys or payment credentials. Requests receive
manual review; automatic refunds, future updates and indefinite hosting are not
promised. Read the exact customer terms.
