After two or more years out, the first thing gone is your hands, not your judgment. Standard library names, argument order, the shape of a sort key: that layer starts rotting within months, and it is also the cheapest layer to rebuild. Systems intuition, the thing you were most afraid of losing, moved the least. The six weeks below repair the list in decay order, fastest first, because everything near the top hides whether the things below it are actually broken.
The decay ranking, fastest to slowest
This is the order things actually go for someone who stopped writing production code, whether the break was caregiving, health, a long layoff, a visa wait, a startup that died, or a deliberate stop. Rank 1 is gone first and repaired fastest. Rank 8 is probably still intact.
| # | What decays | How fast | The test that proves it is back |
|---|---|---|---|
| 1 | Standard library and syntax recall: the heap import, a deque, a frequency map, a sort key with a tiebreak, a 2D array whose rows are not the same object | Months. First to go, and the loudest in a live round | Ten primitives typed from a blank file, under a minute each, no autocomplete |
| 2 | Working in a bare editor: no linter, no test runner, no jump to definition, nothing underlining the paren you forgot | Under a year. Your setup was doing more than you noticed | Twelve easy problems, eight correct on the first run |
| 3 | Pattern triggers. You still remember what a monotonic stack is. What went is the reflex that fires it from the wording of the input | One to two years | Naming the pattern out loud before typing, on 12 of 15 problems |
| 4 | Talking while typing, in front of a stranger who is grading you | Decays with any time out, and for a lot of people it was never there | No silence over 20 seconds in the first eight minutes of a recorded session |
| 5 | Concentration stamina across a four-round day | Goes with any long break from full-time work, faster if the break involved broken sleep | Two 45-minute rounds back to back with a ten-minute gap, no drop in the second |
| 6 | Ecosystem currency: versions, defaults, the tool everyone switched to while you were out | Real but shallow, and it feels far worse than it is | Three sentences on what changed in your stack since your last commit |
| 7 | Debugging instinct and code reading | Very slow. You will surprise yourself here | Read a stack trace in a language you have not touched in two years and find the real frame |
| 8 | Systems intuition and trade-off judgment: why the queue is there, what breaks at ten times the load, which consistency you can give up | Last and least. Two years does not remove it | Nothing. Stop auditing this one and go fix rank 1 |
The ordering matters more than the list. Cost to restore rises as you go down, and every row masks the row beneath it. If you cannot type a heap in under a minute, you cannot tell whether your pattern recall is gone or whether you just ran out of clock at minute nine. People coming back from a break tend to diagnose themselves at rank 8, decide they have lost the ability to engineer, and stop applying. They were failing at rank 1.
Six weeks, in decay order, with one target per week
Five sessions a week, 45 to 90 minutes each. Every week carries a single number, and you do not advance until you hit it.
Week 1: twenty primitives, no problems at all
Write a list of twenty one-liners in your interview language and type them daily from blank. Sort tuples by the second field descending. Frequency map with a default. Min heap of pairs. Deque used as a queue. A grid of zeros where the rows are not aliased. Set difference. Integer division on a negative number. Iterate a map in value order. Split, strip, join. The library binary search, and what it returns on a miss. Target: the whole list in under 20 minutes, zero lookups, by Sunday.
Week 2: bare editor, first-run correctness
Twelve easy problems in a plain text editor, autocomplete off, no assistant open. Do not run the code until you believe it is right, then run it once and record pass or fail. You are not measuring speed this week, you are measuring whether you can still hold a program in your head without a red squiggle. Target: 8 of 12 correct on the first run.
Week 3: triggers before code
Fifteen problems, mixed easy and medium. Before typing anything, write one sentence naming the property of the input that should fire a pattern. Sorted input plus a pair condition fires two pointers. Longest run under a constraint fires a sliding window. Nearest greater element fires a monotonic stack. Unweighted shortest path fires BFS, never DFS. Target: 12 of 15 triggers named correctly before the first line of code.
Week 4: the voice
Five sessions, 35-minute timer, screen recording on, talking aloud from the first second to an empty room. Review only the opening eight minutes at 1.5 speed and count silences longer than 20 seconds. Narrate the dead ends: saying you are checking whether sorting breaks the index requirement is a sentence, staring at the screen is not. Target: zero silences over 20 seconds by Friday.
Week 5: the gap answer and the currency page
Two jobs this week. Deliver the gap script below out loud, standing up, until it lands in about 40 seconds without notes. And write one page on what changed in your stack since your last commit, sourced from release notes rather than opinion threads. Target: ten clean spoken deliveries, and three specific changes you can name unprompted.
Week 6: stamina, in one sitting
Two mocks with actual humans, plus one simulated day: two 45-minute coding rounds back to back with a ten-minute gap, then a behavioural round, without leaving the desk. This is the week that catches the failure nobody plans for, which is round three going badly because round two used everything you had. Target: the sequence completed once, with rounds one and three graded the same.
Two adjustments. If you changed language during the break, week 1 becomes two weeks and the plan is seven. If your break was under a year, skip weeks 1 and 2 outright and open at triggers.
What to say about the gap
The coding interviewer will usually not ask. Recruiters ask, hiring managers ask, and the question is not really about your reason. They are checking three things: whether the break is over, whether it will recur, and whether you are current enough that ramp-up looks normal. Answer those three, in that order, and stop.
Three beats, about 40 seconds:
- Dates and one clause of cause. No defence, no apology, no second reason. Offering two reasons reads as neither one being real.
- One dated artifact. Not "I kept my skills sharp". A named thing with a month attached: a repo, a course you finished, a short contract, an old side project you ported to a current runtime.
- The pivot. It is settled as of a named month, here is the work you want next, and that is why this role.
I was out from March 2024 to May 2026, caring for a parent full time. That ended in May. Since then I rebuilt my old scheduling service on the current runtime and shipped it, which is the repo in my application. I want to go back into backend work on a team that owns its services end to end, which is why I applied here.
Four things not to do. Do not lead with the gap in a coding round, because that round is 35 minutes and you would be spending them on your own biography. Do not inflate the break with a consulting entity that had no clients, since background checks generally confirm dates and titles. Do not volunteer a diagnosis, a relative's condition, or your application count. And do not call yourself rusty, because the interviewer has not concluded that, and week 1 is what fixes it anyway.
Where this plan, and this tool, are wrong for you
The decay ranking assumes you are returning to roughly the work you left. Switch specialisation at the same time, backend to machine learning say, and the order inverts: rank 8 is not intact for you, because systems intuition in the new area does not exist yet, and it becomes the expensive item instead of the free one. This plan will not do much for that case.
If the break was health related and stamina is the genuine constraint, treat week 6 as optional. Plenty of companies will split a loop across two days if the recruiter is asked, and asking is not a red flag.
On tooling, the honest limit runs against us. Inside a week 2, 3 or 4 session, Code Copilot is the wrong tool, and opening it while the timer runs destroys the only unfaked measurement you have. It belongs after the timer stops, when you want the optimal solution with the intuition, a dry run and the Big-O in one place. The remaining facts, plainly: OfferPilot AI runs on Windows 10 and 11 only, with no macOS, Linux, mobile or browser version. The overlay is excluded from screen capture at the operating system level, so it does not appear in Zoom, Teams, Meet, OBS or screenshots. That covers screen sharing and screen recording, and nothing past them. It does nothing about a webcam, a lockdown browser, a proctor in the room, eye tracking, or a phone on a stand pointed at your desk, and where an assessment forbids outside help the six weeks beforehand are the only part that counts. It also cannot answer the gap question. Nobody can produce a dated artifact on your behalf.
Measure tonight
Open a blank file, start a timer, and type the twenty primitives. Whatever number comes back is your real baseline, and it is usually far better than two years of dread suggested. Then put the weeks in the calendar. When you reach week 5 and the behavioural rounds start to matter, the company Q&A generator builds a role-specific bank, including the version of the gap question that particular company tends to reach for. The free tier is five minutes a day and takes no card.
Related guides
- Dyslexia Coding Interviews: Cutting the Reading Bottleneck
- Technical Interviews in English as a Second Language: 40 Phrases
- Non-Native English Speaker Interview Tips: Recovery Phrases
- US Interview Culture for International Candidates: Calibration
- Panic Attack During an Interview: What to Say in the Moment