Why Old LeetCode Lists Stop Working: Recency Bias in Interview Prep

Every interviewing community has its canonical lists. "Top 75 for FAANG." "The Blind 75." "150 questions before your Google loop." These lists are widely shared because they were useful to someone, once. The problem is not that they are wrong, it is that they are static artifacts in a system that changes continuously. LeakCode was built specifically to solve this. Here is the analysis behind why it matters.

How fast do company question pools actually rotate?

LeakCode ingests reports continuously from multiple sources and timestamps every entry. This lets us look at what proportion of questions reported in recent months were also represented in reports from two years ago, and what proportion appear to be new additions.

The rotation rate varies significantly by company. Some companies appear to cycle their question pools faster, either because they have more interviewers who introduce variety, or because HR actively manages question freshness to reduce "gaming." Other companies show more stable pools, either because they have tightly standardized their question sets, or because certain problem types are so aligned with their technical culture that they recur across cohorts.

What LeakCode's data shows across the broader corpus is that a non-trivial fraction of questions reported in 2026 interviews did not appear in reports from 2023 or earlier. The exact proportion varies by company and by round type, but it is large enough to matter for preparation strategy. If you are working from a 2023 list and preparing for a 2026 loop, you are working from a partial map.

The rotation is also non-uniform by topic. Some algorithmic categories are nearly timeless: binary search, sliding window, and BFS/DFS patterns appear across the full history of the LeakCode database with consistent frequency. Other categories show more volatility, either emerging as new favorites or declining as companies rotate away from patterns that have become too widely known. See the LeakCode methodology page for how we track this over time.

Why static lists get stale faster than people expect

The internet has made preparation material for technical interviews extremely public. When a problem appears frequently in candidate reports and gets discussed broadly, it eventually becomes well-known enough that interviewers at the company become aware of it too. The result is predictable: questions that appear too often in public prep materials get replaced by companies that are paying attention.

This is not a new dynamic, but it has accelerated. The density and speed of interview report sharing has increased over the period covered by the LeakCode database. The lag between a question becoming widely discussed and it being rotated out of active use has compressed.

The practical implication: static lists that became popular because they were accurate descriptions of interview question pools at a point in time are increasingly accurate descriptions of what companies used to ask, not what they currently ask. The more popular a list gets, the faster it becomes a target for rotation by companies managing their question pools.

LeakCode's approach is to ingest continuously and weight recency. Our stats page shows the last ingestion timestamps and report counts by time period, so you can see how fresh the data behind any company's question pool actually is. Our changelog logs every significant data update.

The sources problem: why aggregation matters

Most candidate report sharing happens in one of a few places: LeetCode Discuss, Reddit, Blind, and a handful of Chinese-language forums. Any single source represents a slice of the candidate population, not the whole. Candidates who report on Reddit skew toward certain companies and roles. Candidates who report on LeetCode Discuss skew toward coding-focused content. Candidates on 1Point3Acres, which LeakCode indexes and translates, skew toward a different geographic and linguistic background that sees different interview styles and often different question pools.

A prep resource built on only one source inherits that source's blind spots. If you prep exclusively from LeetCode Discuss data, you are missing the behavioral depth that comes from Reddit and Blind reports. If you prep from English-only sources, you are missing the 4,842 entries from 1Point3Acres that are nearly invisible to English-speaking candidates.

LeakCode pulls from seven sources. This is not a cosmetic feature. It means the question pool you see on LeakCode is broader and more representative than any single-source alternative. It also means the recency signal is stronger: when a new question appears across multiple sources in a short time window, LeakCode's ingestion picks it up from multiple angles, which is a stronger signal than a single report. For the full source breakdown, see the analysis in our earlier post on FAANG hiring data for 2026.

Recency weighting in practice

When LeakCode surfaces the most-asked questions for a company, recency is a factor in how questions are ranked. A question reported ten times in the last six months ranks differently than a question reported ten times over the last three years, even if the raw count is the same. This is intentional, and it is the core of how LeakCode differs from a static list.

There are two practical consequences of this approach. First, questions that were genuinely common in 2022 but have faded from reports in 2025 and 2026 will rank lower, even if they are still theoretically "in the pool." Second, questions that started appearing more recently will surface faster than they would in a static list that does not update.

This matters most for candidates preparing for companies with active question management, where the pool rotates faster. For companies where the pool is more stable, the recency weighting has less effect. LeakCode shows you both the aggregate count and the recent-period count for each question, so you can make your own judgment about which signal to weight more heavily for your specific target company.

What this means for how you should structure your prep

The recency problem does not mean static lists are worthless. It means they should be treated as a floor, not a ceiling. Use a classic list to cover the foundational patterns that are genuinely evergreen. Then use a recency-aware database like LeakCode to close the gap between what was commonly asked and what is currently being asked.

Specifically:

• Start with patterns, not problems. The algorithms that appear in interviews are relatively stable. Specific problem instances rotate. Build your pattern library first, then use LeakCode to see which current problems instantiate those patterns.
• Filter by recency, not just count. On LeakCode, use the date filtering to focus on reports from the last 12 months. A question with 30 reports from 2021 and 2 from 2025 is a different preparation priority than one with 15 reports uniformly spread across the last 18 months.
• Cross-reference across sources. Questions that appear in both LeetCode Discuss and Reddit reports from recent months are stronger signals than questions that appear in only one source. LeakCode's multi-source aggregation makes this cross-referencing automatic rather than something you have to do manually.

LeakCode's source attribution page shows how each question in the database is sourced, which lets you apply exactly this kind of cross-source confidence scoring to your preparation list.