Example sessions
See how a solution takes shape. Follow the decisions, inspect the changes, and learn how to work with AI.
37 of 37 walkthroughs
Repairing an expiring cache
An authored walkthrough of a cache repair: reproduce the bug, question an AI suggestion, and verify each change.
Growing a cache pool without reshuffling it
Investigate a cache-pool expansion that moves too many objects, and follow the repair through placement and inventory checks.
Retrying a request without repeating its side effects
Follow a retry-client incident from failing request traces to a reviewed repair and routed-delivery verification.
Keeping a cursor inside the right tenant and timestamp
Investigate missing activity rows, inconsistent page order, and misleading continuation links in a tenant feed.
Letting a new worker claim a job without trusting the old one
Work through a scheduler incident involving expired claims, delayed workers, and the safety of completion requests.
Keeping fractional quota across rejection and clock skew
Investigate quota decisions that disagree with elapsed time and compare the repair against real gateway traffic.
Recognizing a replay without trusting a reused event ID
Follow a payment-consumer repair involving duplicate deliveries, corrected retries, and conflicting event contents.
Repairing document permissions across tenants
Investigate a document leak, review the AI’s permission assumptions, and verify behavior across reads and updates.
Redacting secrets without damaging log data
Follow a logging leak from its first reproduction through an AI proposal, a targeted repair, and regression checks.
Planning deployment waves inside the requested dependency closure
Review a deployment planner whose ordering and diagnostics change with the requested release and service catalog.
Remembering a deletion when older events arrive later
Investigate a customer projection that revives old records when deliveries arrive out of order.
Batching order lookups without changing the report
Investigate an expensive order report and review an AI-assisted repair against both its performance and output requirements.
Repairing structural equality and difference paths
Investigate inconsistent assertion results and review how the AI explains and locates differences between JSON values.
Repairing quoted record imports
Follow a broken import through a small reproduction, an AI parsing proposal, and checks for preserved rows and field contents.
Rejecting stale edits before they erase newer work
Investigate a document workspace that loses edits when several writers act on the same starting snapshot.
Separating a blocked call from a failed dependency call
Trace a dependency breaker that opens unexpectedly and review how its behavior changes across failures and recovery.
Computing a percentile from the right window
Investigate misleading latency numbers using small timestamped examples and the dashboard that displays them.
Detecting edited audit content without rewriting the evidence
Find why an edited audit entry passes verification, then review a repair against the investigation console.
Repairing a classifier evaluation report
Follow a misleading evaluation report from the first failing example through an AI review and verified repair.
Repairing hybrid search rank fusion
Investigate unexpected search rankings and missing results, then review proposed AI changes against the documented evidence.
Building training features without future information
Trace a historical feature mismatch, challenge an AI shortcut, and check what each prediction could have known.
Making notification preference updates stick
Investigate settings that refuse to save, challenge an AI shorthand, and verify the saved state and change history.
Repairing cart totals without changing discounts
Trace an incorrect receipt, review the AI’s arithmetic assumptions, and verify the repair through saved-cart workflows.
Keeping private fields out of public profiles
Investigate a profile data leak and review how an AI-assisted change affects privacy, consent, and existing directory behavior.
Repairing inventory availability pages
Follow a stock-filter bug through reproduction, an AI review, and checks for correct inventory pages and totals.
Tracing a deployment preview with the wrong settings
Investigate a deployment preview that ignores a later configuration layer, then challenge an AI merge proposal and verify the application flow.
Investigating missing regional translations
Follow a regional translation failure through catalog lookup, message selection, and the saved preview while checking an AI fallback proposal.
Repairing a toolbar that moves focus unexpectedly
Reproduce confusing keyboard movement in a toolbar, distinguish navigation from rendering, and verify the resulting tab-stop projection.
Investigating missing and misleading form errors
Trace absent diagnostics and inconsistent validation totals while reviewing an AI proposal and exercising the saved-form preview.
Investigating values lost during snapshot capture
Follow a snapshot that changes saved values, test the difference between presence and output, and verify capture through the application.
Finding invoices that disappear from the ledger
Trace a missing invoice through customer lookup and payment aggregation, challenge an AI join proposal, and reconcile the final ledger.
Preserving saved configuration through an upgrade
Investigate configuration lost during a version upgrade, review an AI migration proposal, and verify the saved workspace inspection.
Investigating inconsistent telemetry counts
Trace telemetry outputs that disagree across bounds, question an AI total-count proposal, and verify the selected metric export.
Investigating inconsistent virtual asset paths
Follow a virtual asset manifest from surprising canonical keys through rejected paths and exact saved-workspace execution.
Finding a missing byte in a range response
Count a small byte request by hand, challenge a cosmetic metadata fix, and test resource-boundary behavior through the application.
Investigating an upload that publishes too early
Trace conflicting upload chunks and incomplete publication, review an AI write proposal, and verify the complete local upload lifecycle.
Finding credits lost during allocation
Trace an allocation with unused credits, challenge an AI tie-breaking proposal, and reconcile caps, row identity, and the final receipt.
Learn from the reasoning
These educational examples pair an authored conversation with code and verified checks. Full sessions require Pro and a submitted attempt of the same problem version.