# Decide what survives a provider failure Published and revised: October 13, 2026 · Version 1.0.0 Move beyond retrying: identify shared dependencies, limit unsupported service, and define evidence for controlled recovery. ## Goal and proposed time Write an honest degraded-operation decision that preserves useful supported work without inventing successful failover. Allow 90–120 minutes, including selected reading, making the outputs, and review. This is a proposed study budget, not measured completion time. Split the path across sessions if needed; stop at a checkpoint rather than rushing an evidence check. ## Prerequisites and inputs - Understand request limits and the distinction between retry and a new action - Have read a basic tool contract; software implementation is optional and outside this path Use only the linked guides and their public or synthetic fixtures. A text editor, calculator, and timer are enough. No provider connection, account change, publication, or outside communication is required. ## Work through the path ### 1. Design recovery before retrying an uncertain action Read [the guide](/agentics/design-retry-safe-recovery/). Walk the uncertain-outcome state machine first. Explain why a timeout does not prove an action failed and why a new operation key can duplicate an effect. Carry forward: A recovery note separating unknown writes from the capstone’s read-only provider outage. ### 2. Write a tool contract that can reject a bad request Read [the guide](/agentics/specify-tool-contracts/). Use the response-contract section to list the fields a replacement output must preserve. A syntactically valid response still needs applicable evidence. Carry forward: A compatibility checklist with missing-result behavior. ### 3. Map an agent's permissions before it acts Read [the guide](/agentics/map-agent-permissions/). Add a hypothetical provider switch as a separate proposed operation. Name the input data and receiving service; leave it blocked where authorization is absent. Carry forward: A transmission boundary that cannot be cleared by technical success. ### 4. Make a cost, latency, and quality decision from a run log Read [the guide](/agentics/measure-agent-cost-latency/). Define what you would measure during normal, degraded, and recovered operation. Keep failed attempts and required stops in the ledger. Carry forward: A blank run ledger and quality gates; no fabricated measurements. ## Worked practice Complete [Keep a useful service when the provider disappears](/agentics/practice/provider-loss-degraded-service/). Map shared dependencies, reject an unapproved or incompatible fallback, and preserve only the service that remains supportable. When gateway G fails, both A and B lose their shared route. A still-valid static hours card can support a labelled limited response; it cannot establish a live seat count. Provider C’s different route does not solve its missing authorization or incompatible evidence fields. ## Output A dependency map, output-contract check, degraded-response card, and recovery decision with a bounded measurement plan. ## Exit checks - Different provider names are not assumed to imply independent dependencies - Unknown action outcomes are reconciled before any proposed repeat - Authorization and fallback compatibility are separate gates - Partial service is labelled accurately; live facts are withheld when unavailable - Cost and latency stay unmeasured unless actual records exist ## Decide what happens next Keep the result when the checks pass. Revise a specific failed check in a new draft, preserving the first attempt. Hold any part that needs missing evidence or authority; an honest hold can complete this learning path even though the wider real-world task remains unfinished. Do not call an authored answer key an observed result. IIEN means infer intent and extrapolate needs here: use it to notice a useful next question and label assumptions. It does not authorize new actions, expand access, or tune model weights. ## Sources and scope This is an editorial sequence of the linked existing guides. Their source sections support the individual methods; the sequence, checkpoints, and study budget are original teaching choices. This path does not claim certification, deployed capability, or measured performance improvement.