Preserve the generated draft
Select one resume version and copy the relevant section into a review document without editing it. Record the document language, role context, and generation date if those details are available in your own records, but do not infer missing metadata. The goal is to create a stable reference for what the AI actually wrote. If the Radaryno Resume Kit produced the document, you can review the generated HTML or PDF text alongside your career notes; the kit’s documents remain on your computer. Keep the original separate from any corrected version so later changes can be traced to a specific claim.
Create one ledger row per claim
Build a simple ledger with fields for the original claim, its smaller factual parts, the source note, the evidence classification, and the proposed correction. Enter the claim before deciding whether it is accurate. A sentence may contain a tool, action, audience, date, ownership level, scale, and outcome, so one row may be too broad. Use separate rows when a phrase could be supported while another phrase cannot. This structure is useful whether the document was written manually or generated from buyer-provided career notes with the Radaryno Resume Kit.
Split compound statements before judging them
Read each sentence as a collection of factual assertions rather than as one polished unit. Separate who did the work, what they did, which technology or process was involved, for whom, when, and what changed. Pay close attention to verbs such as led, owned, managed, delivered, and designed because they can raise the level of responsibility beyond the notes. Also separate numerical outcomes from the underlying activity. This step is the main safeguard against accepting a sentence merely because some of its words match the career record.
Trace each part to a career note
For every small assertion, locate the closest supporting passage in the career notes and copy enough context to show what it actually says. Record the note’s wording rather than relying on memory. A source can support a technology but not an outcome, or an activity but not a leadership claim. If the notes do not contain a direct match, record that absence plainly. Do not fill the gap from common industry practice, the job description, or what would sound persuasive. The ledger should show the boundary between documented experience and an attractive inference.
Classify evidence with precise boundaries
Use three practical classifications: supported, unsupported, and uncertain. Mark a statement supported only when the notes clearly establish the same activity, scope, and level of responsibility. Mark it unsupported when the notes provide no basis for the detail, including an invented metric or tool. Use uncertain when the notes suggest the detail but do not establish it clearly enough to state as fact. Apply the classification to each claim part, not only to the sentence as a whole. This makes the review useful when a generated sentence mixes accurate and inaccurate details.
Rewrite and recheck the corrected claim
Replace unsupported wording with the narrowest statement the source can defend. Remove invented metrics, reduce ownership verbs when necessary, and keep a supported technology or activity if it remains accurate. For uncertain details, either omit them or phrase the record conservatively without implying proof. Then compare the corrected sentence with the original ledger row and confirm that every remaining factual element has a source. If the Radaryno workflow reports a gap or asks rather than inventing, use the same discipline during your review: unanswered evidence questions should remain open rather than becoming polished claims.
| Draft claim | Career note located | Finding | Safer correction |
|---|---|---|---|
| Fictional illustrative example: Led a PostgreSQL migration that cut reporting time by 40%. | Maintained PostgreSQL reports; notes mention no migration and no measured reduction. | Only the technology and reporting context are supported. | Maintained PostgreSQL reporting workflows; migration ownership and time reduction were removed. |
| Fictional illustrative example: Built a JavaScript dashboard for the finance team. | Notes say created an internal JavaScript dashboard for finance analysts. | Supported, with wording narrowed to the documented audience. | Built an internal JavaScript dashboard for finance analysts. |
| Fictional illustrative example: Managed three engineers during the platform launch. | Notes describe coordinating launch tasks with three engineers; no direct-management statement. | Coordination is supported; people management is uncertain. | Coordinated launch tasks with three engineers. |
| Fictional illustrative example: Automated weekly exports and saved the team ten hours. | Notes confirm automated exports but contain no time estimate. | Automation is supported; the quantified outcome is unsupported. | Automated weekly exports; the time saving was not stated in the notes. |
A worked review sequence
Start with one generated sentence and preserve it exactly. For the fictional claim about leading a PostgreSQL migration, mark the technology, activity, ownership, and measured outcome as separate evidence questions. This prevents a supported tool name from making the entire sentence appear reliable.
Next, search the career notes for direct support for each part. The fictional notes confirm PostgreSQL reporting work, but they do not describe a migration or a forty percent reduction. Record the closest source without upgrading it into proof.
Then classify the parts separately. The technology and reporting context are supported, while migration ownership and the metric are unsupported. If a note is suggestive but incomplete, use an uncertain classification instead of forcing a yes or no.
Finish by rewriting only the supported substance and retain the ledger beside the corrected draft. In this fictional case, the safer statement describes maintaining PostgreSQL reporting workflows and removes both the unverified ownership and the unrecorded measurement.