Job search
CoderPad AI Fluency Interview: Work with Reference Files in the Shared Document Workspace
CoderPad AI Fluency uses a shared document workspace for non-coding work. Learn to verify reference files, review AI-assisted revisions, and distinguish activity playback from video and transcription.
In CoderPad’s AI Fluency interview, prepare to produce and defend a document-based work sample. Your task is to use the supplied materials, make an intelligible decision, and leave a final document whose claims you can defend. Know which references support the answer, what remains uncertain, and whether you are asking AI a question, planning work, or allowing it to change the document.
LandOffer publishes this guide and offers job-search tools. I read current public documentation on October 21, 2026, without entering an interview, uploading files, invoking AI, or recording anyone. The examples below are original fiction with a small local consistency check, not a candidate session or a transcript. For opportunities to prepare toward, browse recent roles on LandOffer then confirm the requested deliverable, reference materials, and assistance rules for that role.

Confirm the document task before using AI
The AI Fluency guide, updated July 6, 2026, describes non-coding work in a shared document environment. It names CSV and Markdown references and AI-generated HTML; Excel is unsupported. Ask reads or answers without directly changing the document; Edit changes it; Plan structures the work. The guide describes reviewable activity and says video calls are not recorded.
That description does not tell you the employer’s particular task, time allowance, or permitted sources. Read those instructions first. A product-operations memo, a customer-response draft, and a campaign analysis can share a workspace while requiring different evidence. Do not substitute a polished generic answer for the requested deliverable.
Before work begins, identify the final document, audience, reference versions, deadline, and boundaries on external tools. If a supplied workbook is unreadable, report that rather than inventing its missing contents. A local CSV conversion is not automatically authorized just because the platform supports CSV; ask how the interviewer wants the source provided.
The candidate guidance, updated October 16, recommends a full-size desktop browser and says the recruiter or interviewer must enable AI for the relevant session. General platform availability does not settle what you personally may use. If the expected AI panel is absent, raise the mismatch with the interviewer rather than quietly moving the task into another service.
Keep questions, plans, and edits separate
The broader AI Tools documentation distinguishes shared AI history from a private, interviewer-only Ask channel. It also says an interviewer can disable AI for an individual pad. A candidate should not expect access to private interviewer assistance or notes.
Use Ask when you need an explanation or critique that you will evaluate before changing the deliverable. Use Plan to expose the sequence you intend to follow. Use Edit for a bounded revision after deciding what may change. These are preparation suggestions based on the documented modes, not a promise that a particular model will follow every constraint.
Reading a reference and naming it in a prompt are also different from verifying that the assistant has used the right contents. State the relevant version and facts, then compare the response with the source. I did not inspect a live file-attachment control or test automatic context selection, so this guide does not prescribe an unverified attachment-button sequence.
Make the revision status explicit: “This is the proposal; I have not accepted it into the final memo.” Once you do accept a revision, reread the affected claims and the surrounding text. A correct changed sentence can leave a contradictory total or outdated recommendation elsewhere.
Worked reference pack: make an internal rollout recommendation
Consider a completely fictional product-operations task. The interviewer requests an internal recommendation about which cohorts fit the next support window. The references are cohorts.csv and policy-v1.md. They contain no real customers, private business data, or proprietary assessment material.
The policy allows eight forecast support hours. It prioritizes eligible, ready East cohorts before West, then uses ascending cohort ID. A legal hold is not readiness. An unknown state stays unknown. Requested seats are not active users or revenue. The memo does not authorize a launch or customer contact.
| Cohort | Region | Eligible | Requested seats | Forecast hours | Readiness |
|---|---|---|---|---|---|
| A101 | East | Yes | 12 | 3 | Ready |
| A102 | East | Yes | 8 | 2 | Ready |
| A103 | West | Yes | 15 | 4 | Ready |
| A104 | West | Yes | 10 | 3 | Legal hold |
| A105 | East | No | 6 | 2 | Ready |
| A106 | West | Yes | 5 | 1 | Unknown |
The completed source review finds three explicitly ready, eligible cohorts: A101, A102, and A103. Together they request 35 seats and nine forecast hours, exceeding the eight-hour capacity. The policy does not say “maximize seats”; replacing its priority rule with an optimization objective would change the assignment.
Following the supplied priority, A101 and A102 use five hours and request 20 seats. The remaining three hours cannot fit A103’s four-hour forecast. Its deferral is a capacity decision, not a claim that it is ineligible or unready. Keep A104’s hold, A105’s ineligibility, and A106’s unknown readiness separate; a single “excluded” total would conceal the next action for each.
Here is a complete example of an Ask request, prepared but not sent:
Review the six fictional rows in cohorts.csv against policy-v1.md. Capacity is eight forecast hours; prioritize East before West and then cohort ID. Identify any contradiction in a recommendation of A101 and A102. Do not edit the memo, assume a legal hold is cleared, or call requested seats active users. Separate supported conclusions from questions for the source owner.
The locally authored final memo reads:
Recommend A101 and A102 for the next planning window: 20 requested seats and five forecast support hours against capacity eight. Follow the stated East-first priority. A103 is ready and eligible but needs four hours; three remain, so defer it. Hold A104 for legal clearance, exclude A105 as ineligible, and ask the data owner to resolve A106’s unknown readiness. Confirm the forecasts and unresolved holds before implementation. This recommendation does not authorize contact or launch.
This is a finished deliverable, not a claim about what CoderPad’s AI produced. It preserves the difference between arithmetic, policy, and an owner’s future approval.
Worked revision: a new capacity does not clear every hold
Now stipulate that the fictional interviewer supplies policy-v2.md: capacity increases to ten forecast hours, and all other rules remain unchanged. This clarification is invented for the example. It was not received from a real employer.
The completed Plan request is likewise unsent:
Outline the changes needed when policy-v2.md replaces capacity eight with ten. Preserve eligibility, readiness, East-first ordering, and the internal-only recommendation. Identify the affected cohort selection, hours, requested seats, remaining capacity, and rationale before drafting changes.
The review plan has five concrete actions: confirm the new source supersedes the old capacity; retain the same eligible-ready rows; rerun the allocation; update the selected cohorts and totals together; check that exclusions and operational limits remain. Name the source version so the reviewer can trace which instruction changed the recommendation.
The proposed bounded Edit request says:
Revise only the selection, capacity, totals, and capacity-deferral explanation in the memo for policy-v2.md. Include A103 if it fits after A101 and A102. Keep A104’s legal hold, A105’s ineligibility, and A106’s unknown readiness. Preserve the statement that this is an internal recommendation, not launch approval. Do not add revenue or observed-user claims.
An intentionally defective editorial proposal would include A104 because “more capacity means the hold is resolved.” Reject that reasoning. Capacity and legal clearance are different fields with different owners. This is a made-up bad proposal for review practice, not an observed model response.
The corrected complete memo is:
Apply the clarified capacity of ten forecast hours. Recommend A101, A102, and A103: 35 requested seats and nine forecast support hours, leaving one hour. Preserve East-first ordering. A104 remains on legal hold; A105 remains ineligible; A106 remains unknown. Confirm forecast assumptions and unresolved holds with their responsible owners before operational action. Capacity has changed the recommendation, not authorization to launch. Requested seats are not observed active users or revenue.
The finished change record identifies one new selected cohort, A103; four added forecast hours; 15 added requested seats; capacity changing from eight to ten; and remaining capacity changing from three to one. All other holds and boundaries survive. This change record gives a reviewer a traceable explanation of the revision; it does not predict how the work will be scored.

What the local checks establish
I saved the original six-row CSV, both fictional policies, and both completed memos. A Python standard-library script checked these six synthetic rows against the declared ordering, eligibility, readiness, and capacity rules, including their integer totals. With capacity eight, it selected A101 and A102: five hours, 20 requested seats, three hours remaining. With capacity ten, it selected all three eligible-ready cohorts: nine hours, 35 requested seats, one hour remaining.
The check also preserved explicit rejections of three wrong interpretations: the capacity change clears the legal hold; all six rows are ready and eligible; requested seats prove active users. The last rejection is a semantic boundary stipulated by the source definition, not something a numerical sum could discover by itself.
These checks verify the declared miniature policy and arithmetic. They do not verify CoderPad’s CSV rendering, file-context retrieval, AI accuracy, document editing, playback, or a candidate’s performance. The script is a local evidence aid, not a hidden scoring rubric or an instruction to run Python during a non-coding interview.
For discussion, prepare a concise explanation: “The recommendation follows the supplied priority rather than maximizing seats. The capacity clarification admits A103 but changes no legal or readiness status. I checked every affected total and preserved the unknowns.” If the interviewer asks for a different objective, obtain that clarification before treating it as part of the original task.
Understand what can be reviewed afterward
Do not equate a session’s reviewable work history with a stored video of your face. The AI Fluency guide’s playback description covers prompts, responses, document edits, and available notes. That does not establish that a candidate receives an export or access to the reviewer’s private materials.
The separate AI Notetaker documentation, updated April 24, describes transcription on Enterprise and Custom plans. It requires organizational enablement and each participant’s agreement. People who opt out can still join the call. Its FAQ says the result is transcript-only, with no stored audio or video, and notes remain while the pad is retained.
A transcript is still a record. Ask what is enabled, who can review it, and what retention policy the hiring organization applies. The feature documentation says organization members can review notes in playback; it does not guarantee that only the person conducting the interview sees them. Nor does it establish whether a separate meeting service or employer system records anything outside CoderPad.
Before ending, identify the current final document and explain any unresolved question. Do not assume an AI chat answer has become the deliverable, a plan has applied its own changes, or a revised capacity has granted operational approval. For another role’s preparation, review recent LandOffer openings and rebuild the task brief from that employer’s actual instructions.


