Meta SDE onsite has long had a reputation for being the most "mechanical" big-tech loop — pattern-matching coding plus a clean SD framework was usually enough. The 2025–2026 cycle pass rate slipped, though, because Meta dialed up the Self-Reflection and Cross-functional Collaboration behavioral signals. Plenty of algorithmically strong candidates fail behavioral, not coding. This review covers each round in order — Coding x2 → System Design → Behavioral — with the real pacing and the scoring cues we see.
The Loop at a Glance
| Round | Time | Content | Scored on |
|---|---|---|---|
| Coding 1 | 45 min | 2 algorithm problems (medium + medium-hard) | Correctness, communication, code quality |
| Coding 2 | 45 min | 2 problems (one with follow-up) | Same + optimization mindset |
| System Design | 45 min | One product-level design | Framework, tradeoffs, depth |
| Behavioral | 45 min | 4–6 BQs + Self-Reflection | Leadership / conflict for E5+ |
E5 and above usually adds a second SD round or a People Management round.
Coding Rounds: Topic Mix and Pacing
Topic frequency (community)
| Topic | Share | Reps |
|---|---|---|
| Array / String | 30% | LC 71, 408, 680 |
| Tree / Trie | 20% | LC 199, 314, 938 |
| Graph / BFS / DFS | 15% | LC 1091, 1162, 314 |
| Heap / Priority Queue | 10% | LC 215, 973, 1091 |
| Interval / Simulation | 10% | LC 56, 759, 986 |
| Random / Math | 8% | LC 528 and variants |
| Other | 7% | – |
Standard 45-minute pacing
0–3 min Clarify + restate with example
3–10 min Brute-force, complexity, surface optimization
10–25 min Optimal solution, narrate while coding
25–35 min Self-driven test cases / edge cases
35–45 min Follow-up (always shows up)
Most candidates lose points on the follow-up: they relax after the first problem and run out of runway. Reserve 10 minutes per problem for follow-up.
Three things that lift the coding score
- Drive clarification — "Are negatives allowed?" / "Can input be empty?"
- Narrate while coding — every line gets one sentence of "why"
- Volunteer test cases — don't wait to be asked
System Design: Meta's Eight-Step Method
Meta SD problems lean product-shaped — News Feed, Instagram Stories, Live Comments, Notifications. Eight steps:
- Clarify — DAU / QPS / data scale / read-write ratio
- API — list core endpoints up front
- Data model — entities, fields, keys / indexes
- High-level architecture — 3–5 components
- Storage choices — MySQL / Cassandra / Redis / Memcached, with reasons
- Walk through 1–2 critical paths in depth
- Tradeoffs / bottlenecks — CAP, hot keys, fan-out modes
- Monitoring + recovery — metrics, gradual rollout, rollback
Tradeoffs Meta cares about specifically
- Fan-out on read vs write — the classic News Feed tradeoff
- Push vs pull vs hybrid — must come up for high-fanout users
- Cache invalidation — Memcached is Meta's workhorse; you must articulate TTL / lease / mcrouter
- Sharding strategy — user_id hash vs range, and the business consequences
Behavioral: Self-Reflection Is the Newest Trap
Since 2025 Meta has surfaced a Self-Reflection category in BQ:
- "Tell me about a time you got harsh feedback. What did you do?"
- "What weakness came up in a review? How did you address it?"
- "Tell me about a project that didn't go well — what would you do differently?"
Strong candidates fail because the answers stay too "professional" — action only, no emotion or introspection. Meta wants:
| Signal | Cue |
|---|---|
| Vulnerability | Acknowledge the emotional reaction ("It stung at first...") |
| Concrete change | Specific behavior change with a timeline |
| Pattern recognition | Link the reflection to later work |
Meta BQ staple bank
- "Tell me about a time you disagreed with your manager"
- "Tell me about a time you pushed back on a stakeholder"
- "Tell me about a time you mentored someone"
- "Tell me about a project you owned end-to-end"
- "Tell me about delivering under unclear requirements"
- "What's your biggest weakness?" → Self-Reflection
One STAR story plus one Reflection story per item is enough coverage.
3-Week Prep Plan
| Week | Coding | System Design | Behavioral |
|---|---|---|---|
| W-3 | Run Meta tag top 100 once | DDIA highlights + News Feed | Draft 8 STAR stories |
| W-2 | One verbal mock per day | 2 full SD mocks | Record and replay BQs |
| W-1 | 4 timed mocks | 2 SD mocks + debrief | 1 BQ mock |
Real Failure Patterns We've Seen
- Pattern 1: Both coding problems correct, but candidate stayed silent the whole time → "communication weak."
- Pattern 2: SD framework was clean but candidate couldn't articulate cache invalidation → lean no hire.
- Pattern 3: Self-Reflection answered with "I don't really have a true weakness" → instant reject.
In our VO assistance flow we mock the full loop, record-and-review, and grade each round at the hire / no-hire level. For pricing and slots, ping WeChat Coding0201.
FAQ
Is the E4 onsite the same shape as E5?
E4 runs the standard 4 rounds; E5 typically adds a second SD round or a People round. E5 BQ weighs Leadership signal more heavily.
Do I have to grind the Meta tag list?
Strongly recommended — repeat rates are high and the top-50 tag list hits ~60%. Don't grind only tags though, the follow-ups frequently leave the tag.
Must I cite Memcached in System Design?
Not mandatory, but Meta runs on Memcached + TAO internally. Being able to say "why not Redis" wins points.
STAR or SAR in behavioral?
STAR is fine, but Meta weights Reflection specifically — close every story with "what I learned and how it changed me."
Preparing for Meta / Google / Amazon SDE onsite?
oavoservice tracks E3–E6 Meta coding, system design, and BQ recalls every cycle. Mentors come from front-line Meta SWE / EM teams and offer full-loop mocks, SD framework drilling, BQ story polishing, and Self-Reflection templates as part of our VO assistance.
👉 Add WeChat: Coding0201 — Get the Meta SDE onsite prep package.
Contact
Email: [email protected]
Telegram: @OAVOProxy