Articles · eBay
eBay SWE Intern OA 2027: Coding Questions, Scoring, and Interview Process
Prepare for the eBay SWE intern OA in 2027 with CodeSignal evidence, scoring and score-sharing guidance, coding practice, and later interview preparation.
The eBay SWE intern OA for 2027 raises two different questions: how to solve the coding tasks, and what the result means for your application. Recent historical internship reports point to CodeSignal, but this research did not verify an official eBay 2027 assessment specification or passing score. A platform maximum is not an interview guarantee.
Evidence checked September 7, 2026: Candidate examples below concern the Summer 2026 cycle. Provider documentation explains CodeSignal’s current assessment mechanics; it does not establish eBay’s next-cycle hiring policy. No two independent, same-cycle 2027 intern reports were verified.
Use this guide to identify your assessment, interpret scoring without inventing a cutoff, and practice implementation you can explain in later interviews.
Is the eBay intern OA on CodeSignal?
Candidate evidence: In a Summer 2026 eBay SWE internship thread, applicants describe CodeSignal, results out of 600, and an option to share a previous result. One participant reports 434/600; the original poster reports 600/600 without having heard back at the time of that comment. These are individual reports, not an employer threshold. Summer 2026 OA discussion.
Official employer context: eBay’s Emerging Talent page links students and recent graduates to opportunities, including a US internship openings route. It does not publish the 2027 SWE OA format or cutoff on that page. eBay Emerging Talent.
The defensible answer is therefore conditional: CodeSignal is supported by historical intern evidence, but your invitation determines the platform and assessment type. Do not replace the invitation with a third-party guide that names HackerRank, a fixed question count, or a deadline without matching your requisition.
Before starting, record the assessment name, expiry time and time zone, permitted languages, resource rules, and whether an existing result can be shared. “CodeSignal” identifies a platform; it does not automatically identify a particular certified assessment.
Four questions and 70 minutes: when that description applies
Official provider facts: CodeSignal’s General Coding Assessment documentation, updated August 2026, describes four questions and 70 minutes. Questions are accessible throughout the assessment and can be attempted in any order. The documentation requires coding inside the assessment rather than using an outside IDE. CodeSignal GCA structure.
That is a GCA specification, not a universal eBay promise. Confirm the assessment name and the instructions shown before the timer begins. A custom test, another framework, or a live interview should not inherit the GCA’s clock merely because it uses the same website.
Our preparation advice: Rehearse reading several unfamiliar problems before choosing an order. Identify the input size, required output, and a plausible baseline for each. Start where you can produce a correct implementation, then revisit questions with a clearer view of the remaining time.
Do not turn a practice heuristic such as “try question four before question three” into a rule. The actual problem may have different constraints. A task you recognize superficially can still demand more debugging than a less familiar task with a short, explicit contract.
Scoring: separate performance from selection
Official provider facts: CodeSignal’s certified Assessment Score ranges from 200 to 600. Its documentation describes base points and additional points for completing a module, with module bonuses reflecting assessment characteristics. The scale is not a percentage or percentile. Qualitative skill feedback is separate and intended for development. Understanding Assessment Score.
Our interpretation: You cannot calculate “questions passed divided by four” and call that the final score. Likewise, 500/600 should not be described as an 83rd-percentile result. A correct arithmetic fraction does not create a meaningful hiring statistic.
No official eBay 2027 passing score was verified. The candidate who reported 600 without an immediate response shows why a maximum score should not be treated as an automatic next-round invitation. That comment does not prove the candidate was rejected or reveal how eBay weighted the application.
Use three separate checks when reviewing your result:
| What you observe | What it tells you | What it does not establish |
|---|---|---|
| Assessment Score | Performance reported for that assessment | eBay’s cutoff, interview odds, or final decision |
| Result submitted or shared | The result-transfer action appears completed | That every recruiting review is finished |
| Recruiter interview invitation | A confirmed next step for this application | The format or timing of every later round |
This distinction also prevents unnecessary retesting. If the concern is an expired invitation or unclear submission status, another coding attempt may not solve the administrative problem. Check the actual assessment page and contact the appropriate recruiter or platform support with the error and deadline.
Can you reuse a previous CodeSignal result?
Official provider guidance: When a suitable new invitation allows it, CodeSignal supports sharing an existing result or taking the assessment again. Its documentation says a retake sends the new result to the requesting company even if it is lower; a company that previously received a result may retain that earlier result too. Sharing past results.
The Summer 2026 eBay thread includes a report of score reuse, but that is not proof that every future invitation will accept every old result. Check the offered options rather than assuming eligibility from another applicant’s screenshot.
Our decision framework: If sharing is available, compare your existing result with your current readiness and the remaining deadline. Do not assume that retaking automatically lets you send whichever score you prefer. If sharing is absent, resolve the reason before treating the missing button as a technical defect.
CodeSignal’s documentation identifies assessment eligibility details such as result age and proctoring requirements as relevant to whether an earlier result can be used. Cooldown and result eligibility. Your invitation’s options are more useful than an old internet rule about a fixed number of days.
Coding practice: validate a marketplace refund ledger
Original exercise—not a reported eBay OA question: Process payment and refund events for one order. Amounts are positive integers in cents. Each event has a unique ID, a kind, and an amount. An exact duplicate event is ignored, but reusing an ID with different data is invalid. A refund cannot exceed the remaining paid balance.
For this exercise, invalid events are rejected without changing the balance or remembering their IDs. Events arrive in the order they must be processed. These choices are part of the practice contract, not a claim about eBay’s payment systems.
class Ledger:
def __init__(self):
self.balance = 0
self.seen = {}
def apply(self, event_id, kind, cents):
if kind not in {"PAY", "REFUND"} or cents <= 0:
raise ValueError("invalid event")
payload = (kind, cents)
if event_id in self.seen:
if self.seen[event_id] != payload:
raise ValueError("conflicting event id")
return self.balance
change = cents if kind == "PAY" else -cents
new_balance = self.balance + change
if new_balance < 0:
raise ValueError("refund exceeds balance")
self.seen[event_id] = payload
self.balance = new_balance
return self.balance
Trace the sequence carefully. p1, PAY, 1000 leaves 1,000 cents. r1, REFUND, 300 leaves 700. Repeating that exact refund leaves 700, rather than subtracting it again. r2, REFUND, 800 fails and must leave the balance at 700.
Now submit r2, REFUND, 200. Under the stated contract it succeeds, leaving 500, because the invalid 800-cent attempt never recorded r2. This is the case that catches implementations which mark an event as seen before validating its effect.
Explain the invariant: after every accepted event, balance is nonnegative and each remembered ID has exactly one accepted payload. Compute the proposed balance before committing either state change. For ordinary dictionary operations, each event has expected O(1) processing time and the ledger uses O(u) space for u accepted unique IDs.
Test a refund before any payment, exact duplicates, conflicting IDs, a refund equal to the full balance, zero amounts, and recovery after a rejected event. The code assumes integer cents supplied by the caller. If the prompt instead accepts text amounts, parsing and validation become additional work; avoid silently converting decimal currency through floating-point arithmetic.
Turning practice errors into a better OA attempt
Our practice method: Classify each failure before doing another problem. A misunderstood rule needs a clearer contract. A timeout needs complexity analysis. A wrong answer on a repeated event needs a state trace. Typing faster does not solve all three.
Use the ledger to practice explaining a rejected approach. “I initially stored the ID before checking the refund balance. That made a failed event affect future valid input, so I moved the commit after validation” shows a specific correction. It is stronger than saying you “handled edge cases.”
During timed rehearsal, preserve a working version before making an optimization. Re-run the smallest examples that distinguish the old behavior from the new one. A solution that passes one visible sample can still fail on duplicate input, equality boundaries, or invalid state transitions.
After practice, write a short error note: the failing input, the broken assumption, and the test that now protects it. Revisit those notes before your next mock. Keep preparation examples separate from confidential or active assessment content.
Five eBay-tagged questions to practice
Use coding questions first; treat design as preparation for a confirmed later discussion rather than an assumed assessment component.
| Practice question | Skill to develop |
|---|---|
| Find the Longest Palindromic Substring | Test odd and even centers, substring boundaries, and deterministic output rules. |
| Determine Whether a String Is a Valid Number | Translate a grammar into exact acceptance and rejection cases. |
| Return the K Smallest Values from Sorted Arrays | Explain a heap frontier, preserved duplicates, and the cost of merging results. |
| Design a Shopping Search Bar with Autocomplete | Practice a bounded marketplace feature: prefix matching, latency, and updated suggestions. |
| Describe Your Most Technically Difficult Project | Explain your contribution, alternatives, and an observed technical result. |
For each coding exercise, practice both the implementation and a short explanation of why it works. A later interviewer can change a requirement; memorizing only the final code leaves you little room to adapt.
What happens after the OA?
Historical Summer 2026 reports: In a separate eBay intern thread, the original poster lists a December 8 OA and December 16 interview invitation. A commenter describes technical project follow-ups and a string problem, plus a behavioral round with scenario questions. Another reports a December 29 interview and a January 14 offer call. These are individual timelines, not an average or a 2027 service-level promise. Summer 2026 interview discussion.
Our preparation advice: Be ready to explain one project in depth, including the requirement, your implementation, a failure you investigated, and a tradeoff. For marketplace-oriented discussion, the ledger offers a useful extension: what changes when two workers process refunds concurrently? The in-memory class is not a thread-safe payment service; explain the need for an atomic persistent operation before adding infrastructure.
Keep your application ID, assessment submission date, and recruiter’s expected update together. Follow up after the stated window, and communicate any real competing deadline. Silence alone does not identify the reason for a delay.
Your next action should follow the evidence you have: verify an unclear invitation, prepare for a named round, or improve a diagnosed coding weakness. None requires inventing a passing score.