paybondpaybond
Sign in

T22 · Tutorial

Ship with receipts (production)

Close the live loop in Console: signed evidence, ARS receipts, Agent runs timelines, and Disputes — not local :9477.

~30 minIntermediateProductionproductoperatorReceipt cards

Outcome: Walk Machine access, Settlement, Intents, Agent runs, and Disputes; explain resize-to-evidence capture.

Learning view: objectives and extras expanded. Switch for commands only.

Mental model

Live release is still proof-gated. Operators inspect Intents / Agent runs / Disputes — not local :9477.

You will be able to

  • Separate funding webhooks from tool-completion evidence
  • Explain capture comes from validated evidence, not the authorize ceiling
  • Walk Console checklists: API keys, Settlement, Agent middleware keys, Intents, Agent runs, Disputes
  • Use Agent runs instead of paybond dev trace for production
  • Wire optional traceSink on production bind

Prerequisites

  • · Protect first agent tool (production) green — or any live attach execute with auto-evidence
  • · Evidence preset on the paid tool (cost_and_completion or stripe_charge)
  • · Console access for Intents, Agent runs, Disputes, and Machine access / Settlement when rotating

Production proof is operator-ready

Live attach still ends in authorize → evidence → release. Local paybond dev trace is for sandbox/dev only — production operators use Console agent runs, disputes, and optional traceSink on bind. Funding webhooks are not tool-completion evidence — see How intent funding works.

Authorization (live)

Operation
registered policy tool
Requested ceiling
policy / bind request cents
Decision
spend_authorized or spend_denied

Bound to funded intent and capability — not sandbox bootstrap.

Completion evidence

Preset
cost_and_completion | stripe_charge | …
Signed by
payee / attach production evidence
Capture driver
validated cost_cents

Gateway rejects empty recognition proofs on production mutations.

Portable receipt

Shape
Agent Receipt Standard (ARS)
Status
released / held / disputed
Operator view
Console Intents & Agent runs

Counterparties replay receipts — not log screenshots.

Console-only ops (first-class checklist)

Walk these named screens after a live execute. Do not rely on local :9477 for production investigations.

  1. Machine access → API keys — Confirm the live service-account key still maps to the expected tenant; rotate from here if leaked (never mint production via device login).
  2. Settlement — Verify the rail used on the intent still matches tenant config after deploy changes.
  3. Machine access → Agent middleware keys — Inventory attach credentials; revoke compromised bundles and mint replacements for funded intents.
  4. Intents — Open the live intent dossier: state, ceiling, evidence status, settlement outcome.
  5. Agent runs — Open the middleware timeline for authorize → execute → evidence stages (production substitute for paybond dev trace).
  6. Disputes — When a counterparty contests release, open the disputes workbench and gather the evidence pack — not chat logs alone.

Where to inspect (production)

  • Intents — funded state, evidence, capture/resize story
  • Agent runs — middleware timeline (not :9477)
  • Optional: traceSink / createAgentRunTraceReporter on production bind
  • Disputes — counterparty contests release
  • Compliance exports — evidence packs under operator policy

Standard & docs

Verify before you continue

Check these off against your terminal or timeline output — progress stays on this device.

0/6

If something goes wrong

  • If you see

    Treating Stripe or Shopify webhooks as tool completion evidence

    Do this

    Keep provider funding / order webhooks separate from payee completion evidence after the guarded tool returns — see How intent funding works.

  • If you see

    Expecting :9477 for production timeline

    Do this

    Local dashboard is dev-only. Production: Console Agent runs (timeline) and/or your traceSink integration.

  • If you see

    Overriding capture after evidence to force a larger amount

    Do this

    Settlement uses validated signed evidence only — caller's requested ceiling bounds spend; it does not force larger captures.

  • If you see

    Searching docs paths instead of named Console screens

    Do this

    Operator work is Machine access, Settlement, Intents, Agent runs, and Disputes — human labels first; routes are hrefs only.

Self-check

Answer without scrolling up — then reveal the model answer to compare.

If authorize approved 50_000 cents and evidence reports cost_cents 12_400, what should capture follow — and which two Console surfaces do you open first for a live exception?

Next steps

Pick a branch — not every path needs every tutorial.

Related docs & recipes

Recipes are copy-paste production smokes — not repeated inside this tutorial.