Does 2024 Interview Evidence Still Hold When You Rehire From the Pool?

Which parts of a 2024 interview file still hold?

Sort every line in the file into three piles. Still true: how the candidate reasons, what they know about the domain, how they explain a decision under challenge. No longer true by default: level, comp expectation, availability, what they now want. Unknown: how they work, because that is the part the last two years rewrote.

The first pile is bigger than the re-engagement advice implies. The World Economic Forum's Future of Jobs Report 2025 has employers expecting 39% of workers' core skills to change by 2030, down from 44% in the 2023 edition, with analytical thinking still the most sought-after core skill (roughly seven in ten companies call it essential) ¹. Most of what a good 2024 interview measured is in the part that did not move.

The second pile is what every silver-medalist playbook already covers, and it is the easy half: send the note, confirm they are looking, re-check the band. Nobody needs an article for that. The expensive mistake is treating the first pile's survival as proof that the third pile survived with it, and skipping to an offer on a file whose newest page is two years old.

Nothing sets an expiry date, and the federal guidelines say so outright. The Uniform Guidelines on Employee Selection Procedures (29 CFR part 1607, in force since 1978) state that "there are no absolutes in the area of determining the currency of a validity study," and that all circumstances, "including the validation strategy used, and changes in the relevant labor market and the job," decide when evidence is outdated ². The job changing is what dates your file. Not the calendar.

The first pile carries a caveat. A strong 2024 scorecard was never a strong prediction on its own, which is the same limit that makes candidates who interviewed brilliantly struggle in their first quarter. Reopening the pool does not weaken that evidence; it just does not strengthen it either. Whether assessment scores still predict performance once AI is in the workflow is the sharper version of the question, and it is worth reading before you decide how much weight the old file carries.

Why does 2024 engineering evidence age faster than finance evidence?

Because the change landed unevenly by field, and your candidate came from one field rather than from the average. Over September 2023 to February 2024, running into the year that pool was interviewed, the US Census Bureau's Business Trends and Outlook Survey put firm-level AI use at 3.7% rising to 5.4%, with sector rates running from 1.4% in Construction and Agriculture to 18.1% in Information ³.

Software sits at the fast end. In Stack Overflow's 2025 developer survey, 84% of respondents were using or planning to use AI tools in their development process, up from 76% the year before, and 51% of professional developers used them daily ⁴. Asked whether AI tools or agents had changed how they complete development work in the past year, 16.3% of respondents said to a great extent and 35.3% said somewhat ⁴. That is a majority reporting the work itself moved inside a single year.

The same survey is the reason not to overstate it. Agents are not yet mainstream by that survey's own reading: 14.1% of respondents used them at work daily, 37.9% had no plans to adopt them, and a majority either do not use agents or stick to simpler AI tools ⁴. So the honest position on a 2024 engineering finalist is that a live question opened up, not that they are behind.

What the split looks like on the file you are actually holding:

• Software engineering. Decomposition, test discipline and how they read unfamiliar code still read true. Whether they can supervise an assistant that opens a branch, edits nine files and writes its own passing tests is not in the file, because in 2024 nobody asked.
• Financial analysis. Memo craft, reconciling a segment figure to the filing, knowing which assumption moves the recommendation: all durable. What is missing is whether they open a source when a model hands them a confident number that supports the thesis.
• Legal operations. Playbook judgment and redline instinct hold. Retrieval-assisted drafting changed what the review step is for, and a 2024 note describes the old one.
• Marketing. Positioning judgment holds. Producing the artifact is now close to free, so 2024 evidence about output volume proves nothing about the standard they hold a draft to.
• Data and analytics. Statistical judgment holds. The new failure is a fluent explanation of a result the dataset does not support, restated more carefully each time you push.
• Healthcare revenue cycle. Denial-code knowledge holds. Whether they concede the claim an assistant will happily draft a persuasive appeal for is a different test.

If the role also changed between 2024 and now, that is a second clock running. Finding out what AI actually does in this role before you re-open the pool is the cheapest way to know which columns of the old scorecard are still measuring the job you are filling.

Run one current work act, not another five-round loop

One task, 45 to 60 minutes, in the role's own material, with AI allowed and disclosed. Put one decision in it that only checking settles: a figure the source packet contradicts, a claim the dataset will not support, a fixture that agrees with the bug. Then carry the 2024 file forward for everything it already answered, and re-ask none of it.

What makes the act worth an hour of a known-good candidate's time is that it produces evidence the old file structurally cannot. A 2024 interview captured someone describing how they check a confident claim. It could not capture them checking one, because the checking happens while the work happens. That gap is why the re-verification is a work act rather than another conversation about working style.

Keep it gradeable. Two reviewers should be able to fill the same columns separately and agree: which claim the deliverable rests on, what it was checked against, what was cited and plainly never opened, where a range was reported as a point estimate. "Showed good judgment with AI" cannot be scored twice the same way. "Reconciled the segment figure to the filing" is a yes or a no.

Budget it honestly against the loop it replaces. One session plus twenty minutes on their own submission is roughly a third of what a fresh finalist costs you, and the returning candidate is doing it having already passed the parts you are not repeating. The general version of this arithmetic is testing AI skills without adding another hour to the loop, and the constraint is the same one: the round has to get shorter, not longer, or the pool stops answering.

What do you owe a finalist you bring back?

Say what carried over and what did not, in the first message. A finalist who spent five hours on your process in 2024 and gets a form invitation to start again reads it as the last round having counted for nothing. Naming the shortened round, and why one piece is being re-checked, is the difference between a warm re-engagement and a cold application.

The wording that works is plain and short: the 2024 loop still stands for the interviews they did, the role has changed in one respect, and the re-check is one session on that respect. If they ask what changed, tell them. A candidate who hears that the way the job uses AI has moved since 2024 understands the request immediately, because it moved for them too.

Apply it to everyone you re-contact from that pool, not to the ones whose 2024 notes you happen to remember less fondly. A re-verification given to some finalists and skipped for others is two different processes with one job title on them, and it will be the first thing anyone asks about later.

One record point worth knowing before you reopen anything. Under 29 CFR 1602.14, the federal recordkeeping rule for Title VII, the ADA and GINA, application forms and other hiring records must be preserved for one year from the date the record was made or the personnel action, whichever is later ⁵. A 2024 file is past that federal floor, so whatever you still hold you kept by choice, and once it informs a 2026 decision it belongs to that decision's record too. State law and your own retention policy can run longer; your counsel owns that answer for your jurisdiction. The same decision taken in advance, about the people you are not hiring this quarter, is what a pipeline should hold before the role opens.

Write the carry-forward rule before the first outreach

One page, written before anyone gets a message, naming three things: which lines of the 2024 file carry forward, which get re-verified, and what the re-verification is. Written first, it is a process. Written after the first candidate underwhelms someone, it is a justification, and it will read as one to everybody who compares notes afterwards.

A usable version of that page is four lines. Interviews from 2024 stand and are not repeated. Situation facts (level, comp, availability, interest) are re-confirmed by conversation, not assumed. The working half is re-verified by one current work act, the same act for every returning finalist in this pool. The debrief on that act is the only new scorecard, and it does not overwrite the old one.

An old file plus one current session is two observations from two afternoons, taken two years apart. Neither of them is a reference conversation with someone who watched the candidate ship over months. If the role's title survived but its scope did not, the 2024 evidence was collected against a different job, and the currency rule points exactly there: changes in the job, not elapsed time, are what put a study out of date ².

The reason this beats both defaults is arithmetic. Skipping to an offer trusts a two-year-old read on the one thing that changed most. Running the full loop again spends a finalist's goodwill re-proving what you already proved, and across a pool of returning finalists it costs the weeks you reopened it to save. One act on the perishable half is the only move that treats the old evidence as what it is: real, partial, and dated in a specific place you can name.

See a sample report >