Interview rubric
How Troubleshooting Interviews Are Scored: Rubric by Level (2026)
A troubleshooting interview is scored on 6 dimensions: triage and hypotheses (20%), systematic investigation (25%), mitigation before root cause (15%), systems, Linux and network depth (15%), post-incident thinking (10%), communication under pressure (15%). Each is rated 1 to 4 against anchors that rise with the level. At mid level, an outstanding score on systematic investigation means: "Picks the signal that splits the search space in half at each step and says why."
The dimensions and their weights
| Dimension | Weight | Solid (3) at mid level |
|---|---|---|
| Systematic investigation | 25% | Uses logs, metrics and traces to rule hypotheses in or out and bisects the cause. |
| Triage and hypotheses | 20% | Scopes blast radius first, checks recent changes and ranks testable hypotheses unprompted. |
| Mitigation before root cause | 15% | Mitigates first unprompted, by rollback, failover or traffic shift, then hunts the cause. |
| Systems, Linux and network depth | 15% | Explains resource limits, networking and service behaviour correctly under follow-up. |
| Communication under pressure | 15% | Stays structured, gives status updates and says who to inform and when, unprompted. |
| Post-incident thinking | 10% | Describes a blameless review with root cause, contributing factors and concrete actions. |
What changes between levels
On systematic investigation, the heaviest dimension, an outstanding mid-level answer "picks the signal that splits the search space in half at each step and says why", while an outstanding senior answer "finds the cause efficiently and explains which evidence was decisive and why". The same answer can score 4 at mid level and 2 at senior: senior candidates are expected to raise things unprompted that mid-level candidates may be asked about.
Every score, at every level
Triage and hypotheses (20%)
| Level | 1 · Poor | 2 · Borderline | 3 · Solid | 4 · Outstanding |
|---|---|---|---|---|
| Intern / new grad | Guesses a fix at random without asking what is broken. | Asks what is broken only after the interviewer prompts for symptoms. | Asks about symptoms when nudged and names one or two likely causes. | Asks about symptoms and recent changes unprompted and lists likely causes. |
| Junior | Guesses a fix at random without asking what is broken. | Names causes without first checking how many users are affected. | Asks about symptoms and recent changes and lists a few testable causes. | Checks scope and recent changes unprompted and ranks causes by likelihood. |
| Mid-level | Guesses a fix at random without asking what is broken. | Forms hypotheses but scopes blast radius only when prompted. | Scopes blast radius first, checks recent changes and ranks testable hypotheses unprompted. | Scopes quickly and picks the hypothesis that is cheapest to rule out first, with reason. |
| Senior | Guesses a fix at random without asking what is broken. | Triages methodically but fixes on the first hypothesis without considering others. | Scopes impact, sets severity and ranks hypotheses by likelihood and cost to test. | Spots a misleading symptom and redirects triage before time is wasted. |
| Staff | Guesses a fix at random without asking what is broken. | Triages the local service but ignores upstream or shared dependencies. | Triages across services and shared infrastructure and sets severity and roles. | Recognises a systemic or cascading failure pattern early and scopes the response to match. |
Systematic investigation (25%)
| Level | 1 · Poor | 2 · Borderline | 3 · Solid | 4 · Outstanding |
|---|---|---|---|---|
| Intern / new grad | Tries random changes with no way to tell if they help. | Checks logs only when the interviewer tells them where to look. | Checks logs or error messages when nudged and narrows down the cause. | Checks logs and basic metrics unprompted and narrows the cause step by step. |
| Junior | Tries random changes with no way to tell if they help. | Looks at signals only when told which one to check. | Uses logs and metrics to test each hypothesis in turn. | Uses logs, metrics and recent deploys unprompted and narrows the cause quickly. |
| Mid-level | Tries random changes with no way to tell if they help. | Uses signals but jumps between hypotheses without ruling any out. | Uses logs, metrics and traces to rule hypotheses in or out and bisects the cause. | Picks the signal that splits the search space in half at each step and says why. |
| Senior | Tries random changes with no way to tell if they help. | Investigates methodically but needs hints to use traces or compare healthy hosts. | Bisects across time, hosts and versions unprompted and compares healthy against unhealthy. | Finds the cause efficiently and explains which evidence was decisive and why. |
| Staff | Tries random changes with no way to tell if they help. | Investigates the local service but misses cross-service signals. | Investigates across services, correlating signals and coordinating parallel lines of inquiry. | Finds the cause efficiently and names the observability gap that slowed the investigation. |
Mitigation before root cause (15%)
| Level | 1 · Poor | 2 · Borderline | 3 · Solid | 4 · Outstanding |
|---|---|---|---|---|
| Intern / new grad | Never thinks about restoring service, only about the bug. | Considers a quick fix such as a restart only when the interviewer suggests it. | Suggests a restart or undoing the last change when asked how to restore service. | Suggests undoing the last change unprompted before digging into the cause. |
| Junior | Never thinks about restoring service, only about the bug. | Mentions rollback only when the interviewer suggests it. | Proposes a rollback or restart to restore service when asked about users. | Proposes rollback unprompted and checks it restored service before investigating further. |
| Mid-level | Never thinks about restoring service, only about the bug. | Mitigates only after being reminded that users are affected. | Mitigates first unprompted, by rollback, failover or traffic shift, then hunts the cause. | Picks the least risky mitigation with a reason and verifies recovery with a metric. |
| Senior | Never thinks about restoring service, only about the bug. | Mitigates but ignores the risk the mitigation itself carries. | Mitigates first, weighs the mitigation's own risk and verifies recovery. | Sequences mitigations to protect most users first and keeps evidence for the root cause. |
| Staff | Never thinks about restoring service, only about the bug. | Mitigates one service but ignores knock-on effects on dependents. | Mitigates across dependent services, protecting critical paths and preserving evidence. | Uses load shedding or degradation to protect the business and names the decision owner. |
Systems, Linux and network depth (15%)
| Level | 1 · Poor | 2 · Borderline | 3 · Solid | 4 · Outstanding |
|---|---|---|---|---|
| Intern / new grad | Cannot explain basic concepts such as processes, memory or DNS when asked. | Explains basic concepts with major errors. | Explains processes, memory, ports and DNS correctly at a textbook level. | Explains basics correctly and names a command or tool to check each one. |
| Junior | Cannot explain basic concepts such as processes, memory or DNS when asked. | Knows the concepts but cannot name a tool to inspect them. | Names the right tools, such as top, logs or curl, and reads their output correctly. | Uses tools correctly and explains what each output says about the cause. |
| Mid-level | Cannot explain basic concepts such as processes, memory or DNS when asked. | Knows tools but explanations break down on follow-up questions. | Explains resource limits, networking and service behaviour correctly under follow-up. | Explains kernel or network behaviour, such as connection limits, that explains the symptom. |
| Senior | Cannot explain basic concepts such as processes, memory or DNS when asked. | Solid on single hosts but shaky on distributed behaviour such as retries. | Explains distributed failure modes such as retries, timeouts and queue build-up correctly. | Explains subtle interactions, such as retry storms or GC pauses, and how to prove them. |
| Staff | Cannot explain basic concepts such as processes, memory or DNS when asked. | Solid on single services but shaky on cross-system behaviour. | Explains cascading failures across systems and the limits that trigger them. | Explains deep systemic behaviour and the architecture changes that would contain it. |
Post-incident thinking (10%)
| Level | 1 · Poor | 2 · Borderline | 3 · Solid | 4 · Outstanding |
|---|---|---|---|---|
| Intern / new grad | Has nothing to say about what should happen after the fix. | Says to be more careful next time, only when asked. | Suggests a test or check that would catch the problem next time, when asked. | Suggests unprompted a test and an alert that would catch the problem next time. |
| Junior | Has nothing to say about what should happen after the fix. | Suggests vague prevention such as more testing. | Names the root cause and a concrete fix plus a test to prevent recurrence. | Names root cause, a fix, a test and an alert, unprompted. |
| Mid-level | Has nothing to say about what should happen after the fix. | Names prevention steps only when asked, and they are vague. | Describes a blameless review with root cause, contributing factors and concrete actions. | Links actions to SLOs and alerting gaps and assigns owners and priorities. |
| Senior | Has nothing to say about what should happen after the fix. | Describes a review but actions only patch the immediate bug. | Identifies contributing factors beyond the bug and actions that reduce a class of failures. | Uses error budgets and SLOs to prioritise follow-ups against feature work. |
| Staff | Has nothing to say about what should happen after the fix. | Proposes fixes for one team only. | Proposes cross-team actions, such as process or platform changes, that prevent similar incidents. | Turns the incident into an org-level reliability programme with clear ownership. |
Communication under pressure (15%)
| Level | 1 · Poor | 2 · Borderline | 3 · Solid | 4 · Outstanding |
|---|---|---|---|---|
| Intern / new grad | Goes silent or panics and cannot explain what they are doing. | Explains investigation steps only when asked and loses track of what was tried. | Explains each step when nudged and stays calm. | Narrates each step unprompted, stays calm and summarises what is known. |
| Junior | Goes silent or panics and cannot explain what they are doing. | Explains steps only when asked, with no summaries. | Narrates steps calmly and gives a summary of what is known when asked. | Narrates calmly and offers a short status update unprompted. |
| Mid-level | Goes silent or panics and cannot explain what they are doing. | Narrates steps but never mentions informing users or stakeholders. | Stays structured, gives status updates and says who to inform and when, unprompted. | Gives crisp status updates with impact, actions and next update time. |
| Senior | Goes silent or panics and cannot explain what they are doing. | Communicates status but loses structure when new information arrives. | Keeps stakeholders updated on a cadence and stays structured as new information arrives. | Adapts updates to each audience and delegates clearly while keeping the thread. |
| Staff | Goes silent or panics and cannot explain what they are doing. | Communicates well technically but not to leadership or customers. | Acts as incident lead: sets roles, runs updates and speaks to leadership. | Leads the response calmly, sets roles, and keeps leadership and customers aligned. |
Get scored against this rubric
Thita's AI interviewer runs a live troubleshooting round and scores you against exactly this rubric, dimension by dimension, for the level you choose.
Practise a troubleshooting roundWhere this rubric comes from
These rubrics are researched: they follow the scoring methods big-tech companies describe publicly (behaviourally anchored ratings, bar raisers, leadership principles, named interview signals) and how their rounds are commonly run. No company publishes its actual scoring sheet, so the level-by-level wording is our own synthesis. It is the same rubric our AI interviewer scores every mock against.
Questions, answered
- How is a troubleshooting interview scored?
- A troubleshooting interview is scored on 6 dimensions: triage and hypotheses (20%), systematic investigation (25%), mitigation before root cause (15%), systems, Linux and network depth (15%), post-incident thinking (10%), communication under pressure (15%). Each is rated 1 to 4 against anchors that rise with the level. At mid level, an outstanding score on systematic investigation means: "Picks the signal that splits the search space in half at each step and says why."
- What is the difference between a mid-level and a senior troubleshooting answer?
- On systematic investigation, the heaviest dimension, an outstanding mid-level answer "picks the signal that splits the search space in half at each step and says why", while an outstanding senior answer "finds the cause efficiently and explains which evidence was decisive and why". The same answer can score 4 at mid level and 2 at senior: senior candidates are expected to raise things unprompted that mid-level candidates may be asked about.
- What does a score of 3 mean?
- Scores are relative to the level you are interviewing for. A 3 means solid for that level: the bar is met without help on the key points. A 4 means the candidate also raised the important points unprompted. A 2 usually means the interviewer had to lead the candidate to them.
- Is this the official rubric used by Google, Meta or Amazon?
- These rubrics are researched: they follow the scoring methods big-tech companies describe publicly (behaviourally anchored ratings, bar raisers, leadership principles, named interview signals) and how their rounds are commonly run. No company publishes its actual scoring sheet, so the level-by-level wording is our own synthesis. It is the same rubric our AI interviewer scores every mock against.