WORKFLOW EVALUATION KIT

Your code. One decision record.

Connect your original and replacement software paths. Run recovery against the local engine. Review the same evidence with engineering, operations and the budget owner.

Current profile: synthetic USD 10 movement between your own accounts. Two software hooks. No real funds or native provider connection.

Start with one workflow.

The free SDK quickstart demonstrates recovery using fixed HTTP fixtures. The Workflow Evaluation Kit adds your application hooks and a separately supplied local engine. The published SDK stays unchanged.

Inspect the customer adapters

Python and JavaScript templates, workflow input and hook guards.

Request the local evaluator

Confirm permitted access, dependencies and workflow fit. A paid assessment is optional.

Use Linux, Python 3.12+ with venv/pip and OpenSSL 3+ with Ed25519 support. JavaScript also needs Node.js 22+ with npm. The supplied evaluator installs the pinned SDK offline. Arrange your permitted application dependencies during setup.

Start with the setup helper.

In Workspace → Evaluation Kit, select your saved workflow, confirm readiness and accept the supplied evaluation license. After access review, your account receives the exact ZIP, version and SHA-256. Your browser checks the archive before saving it.

Extract outside Git. From the bundle directory:

python3 kit.py doctor --language python
python3 kit.py verify
python3 kit.py reference --language python --output /tmp/haltseal-reference-01
python3 kit.py init --language python --workspace /tmp/haltseal-workflow-01

Use javascript for that language, or both for the doctor and reference. The starter remains unconfigured. Open its NEXT.md, connect your actual code, then run the supplied evaluate command. python3 kit.py results --output /tmp/haltseal-reference-01 verifies the report checksums and explains the next step. Missing tools and mismatched files stop the helper with a concrete correction.

Keep the evidence boundary clear. An empty adapter returns INCOMPLETE. An internal reference adapter verifies the kit. Customer evaluation requires observed calls through the connected application hooks. Entry-point names alone do not establish coverage.

Connect five points in your code.

Hook	Your responsibility	What the run observes
Shared reference	Map both paths to the same agreed transaction.	The mapping matches the supplied workflow reference.
Original path	Connect the original execution point using the supplied client.	One SDK call with the retained operation ID and approval.
Replacement path	Connect the alternate execution point for the same obligation.	One SDK call under the same retained payment authority.
Recovery	Look up the exact original operation or recover the retained attempt.	Original identity is preserved. Recovery does not create another payment.
Stop branch	Stop the application on HOLD or REFUSE.	The hook returns STOP and performs no further SDK operation.

Call your actual permitted nonproduction application functions with the injected client. Python can import sibling application modules beside the adapter; JavaScript resolves relative imports from the adapter file. Use the exact public source commit named in your kit manifest.

The runner observes calls through this SDK client. Direct bank calls, manual portals, hidden senders and omitted paths remain unobserved. An adapter hash binds that file; it does not authenticate every imported dependency. ACCEPT is authorization, not settlement.

Run the supplied local evaluator.

Copy the workflow input and adapter outside Git. Use aliases for the transaction, entry points and owners. The workflow input schema is the exact contract pinned in the supplied kit. Keep confidential code and generated reports local until a separate channel is agreed.

python services/payment-api/customer-evaluation/run.py \
  --sdk-checkout ./payments-sdk \
  --workflow /tmp/my-workflow.json \
  --language python --adapter /tmp/my-code/adapter.py \
  --output /tmp/my-haltseal-evaluation

Run from the supplied evaluator bundle's root. For JavaScript, use --language javascript and your adapter.mjs. Choose a new private output directory outside Git. An existing output is never overwritten.

Scenario	Required observation
Original reply lost	Recover its identity without another original create.
Original UNKNOWN	HOLD the replacement and retain the reserved principal.
Process restart	Retain the original identity and principal.
Closure and old approval	HOLD. Closure requires fresh eligibility.
Approval signed before closure	HOLD until a fresh postclosure approval.
Fresh approval and eight contenders	One accepted replacement and one replacement create.
Observed PAID and another path	REFUSE another full-principal payment.

A missing configuration returns INCOMPLETE; a configured adapter with failed scenarios returns FAILED. Neither produces a passing conclusion. Retain the exact operation or attempt ID, inspect the private stage log and rerun corrected code in a fresh directory. The full integration guide covers input validation and optional Lab attachments.

Three outputs. One owner review.

Output	Reader	Decision it supports
`integration-map.html`	Integration engineer	Which declared application points were observed?
`evaluation-report.json`	Engineering and operations	Which versions, source hashes, calls and scenarios produced this run?
`pilot-decision.html`	Operations and budget owner	Retain current controls, evaluate an additional control, or complete the evidence?

The HTML pages are offline and printable. Their JSON record hash and SHA256SUMS.txt bind the outputs to the same run. A separate Payment Control Lab report keeps its own scope and requires a workflow match review.

KIT_VERIFIED_REFERENCE_ONLY means internal kit verification. LOCAL_CUSTOMER_HOOKS_EVALUATED means the operator-declared customer adapter completed the observed local calls. Neither status establishes an independent customer purchase, savings, native-provider qualification or production readiness.

Review existing controls, excluded paths, remaining conditions and a named owner for the next step. A passing local test does not automatically recommend buying.

Keep the next step scoped.

Save the non-confidential outline in your Workspace. Keep customer code and raw run outputs outside public forms and repositories.

Start a free Workspace Sandbox to define one workflow and retain its next steps. This local kit requires separate scoped access and setup; it is not a qualified customer payment path. Optional implementation assistance is separately agreed. Review pricing and qualification.

Supplier, payroll and subscription collections, native bank/provider connections, additional paths and production deployment require separate qualification. Free public diagnostics and scoped local evaluator requests remain independent of a paid assessment.

Next stepOutline your workflow