Free C1000-180 practice test: 30 questions with answers

A condition that checks a stored variable controls whether a step runs, which matches the requirement. The fallback, a webhook on every message and a version snapshot do not decide whether one step should be reached based on earlier answers.

A session variable keeps a value available to later steps in the same conversation, so the customer is not asked twice. Training an intent, using a digression and changing environment settings do not pass a value between steps.

A clear re-prompt that states the expected format and a handoff after repeated failures are standard recovery patterns. Accepting any answer, publishing untested paths and deleting variables either hide the problem or break the flow.

Repeating only the failed question while keeping earlier variables preserves the customer's progress. Restarting the flow, storing invalid text as a postcode and removing the step all either lose data or produce bad data.

Intents with varied example phrases let the natural language understanding component classify free-text messages, which is what the team wants. A numbered menu removes the flexibility requested, routing everything to agents defeats the automation, and exact string matching fails on ordinary paraphrases.

Digressions allow a user to switch to another topic during a flow and then continue where they stopped. Entity extraction, webhook retry and snapshot versioning each do a different job and do not handle topic switches.

A fallback or anything-else response catches input that no intent matches confidently, which is where a clarifying question or a handoff belongs. The other options describe publishing, logging and translation, none of which is the fallback's role.

Goal-based intent names and merging or sharpening overlapping intents keep the model clear and easy to review. A single generic intent, near-duplicate examples and never retiring unused intents all make the assistant harder to maintain.

Retrieval-augmented generation grounds a foundation model's answer in the approved documents, which the requirement needs. Fine-tuning on public pages, answering from general knowledge and returning file names do not keep answers tied to the company's approved content.

Grounding and guardrails tie answers to retrieved content and keep the model within its approved scope. Longer answers, learning from each chat and skipping intent checks do not reduce unsupported claims and can increase them.

Explicit prompt instructions on length, tone and format, with the context clearly delimited, steer the model toward scannable answers. Limiting the customer's input, randomizing style and removing instructions all work against the goal.

Pre-publication testing and reviewing logged answers, then tightening prompts or sources, are the standard quality controls. Hiding citations, removing scope limits and dropping the knowledge source all reduce grounding and make errors harder to detect.

Reviewing unresolved messages shows the real phrasing customers use, and adding examples or a new intent addresses the gap. Deleting the fallback, creating versions and moving regions do not improve how the assistant understands delivery questions.

Logs show how customers actually phrase requests and where conversations fail, which drives concrete design fixes. Logs do not replace testing, training data should be reviewed by people, and billing totals do not show design gaps.

Comparing completion or resolution rates before and after a change measures whether customers reached their goal more often. Intent counts, greeting length and theme image counts do not show whether the booking experience improved.

Reviewing unresolved conversations and tracking handoffs with their reasons are recurring steps that feed real evidence into the next change. Testing only once, guessing user needs and deleting intents without review all skip the evidence the cycle depends on.

A read-only role limited to the review's needs gives the contractor what they need and nothing more. Sharing an owner's key, adding them to an administrator group and granting full access with a request not to change things all break least privilege.

Revoking an exposed key and rotating in a new one stops misuse, and the application must then be updated. Leaving the key active, renaming the assistant and waiting for a release leave the exposed credential usable.

Data privacy and logging settings on the instance are the administrative control for retention, which must match company policy. An intent cannot delete stored logs, example phrases do not affect retention, and publishing to another region does not limit how long data is kept.

IBM Cloud Activity Tracker records management events on IBM Cloud services, which supports auditing who did what and when. A foundation model, a web chat theme and the example phrase library do not provide an audit trail of configuration changes.

Webhooks and custom extensions call external REST APIs and let the flow map response fields into variables. An entity is for recognizing user input, a search skill does not query live order systems, and a snapshot copies configuration rather than live data.

Keeping credentials in the platform's secured settings limits who can read them and avoids exposing them in flows. Putting keys in training data, intent names or public web chat code all expose the credential to users or to anyone who views the source.

A failure path that explains the problem and offers a next step keeps the conversation moving. An endless retry loop, silence and switching off all integrations leave the customer stuck or worsen the outage.

A unique request identifier lets the back-end recognize a repeated create request and avoid saving the order twice. Removing retries, calling less often and storing order numbers as training phrases do not stop duplicate orders.

Voice callers hear responses and cannot click buttons or links, so prompts must be short and speakable. Long link lists, wide tables and HTML markup all assume a visual screen that a phone call does not provide.

Channel-specific response types let the same flow use rich elements where they work and fall back to simpler content elsewhere. Forcing one HTML layout, speech translation and disabling fallbacks do not solve the channel difference.

Reading the number back in groups and asking for confirmation catches transcription errors before they cause harm. Accepting the first transcription, storing numbers as training phrases and skipping the step entirely either trust a risky transcription or break the task.

Publishing a tested version from draft to live moves a checked change to customers in a controlled way. Editing live directly, copying data to a new account and sharing draft links expose customers to untested behaviour or the wrong data.

Separate environments let a team test and review changes safely while live conversations continue unaffected. The other reasons do not match how environments are used, and backup or model choice are not the purpose of the split.

Restoring the last known good version gives customers a working flow quickly, and the fix can then be tested in draft before publishing. Editing live without testing, recreating the environment and asking customers to skip the step all prolong or worsen the outage.