Real work over self-report
Work arcs and outcomes carry more weight than claimed expertise, job titles, or a list of installed tools.
Operator Engineer evidence map
Proof you can recognize.
Badges are how the AI Work Assessment names what your work shows. Each one is a lens into cited work, and the proof behind it travels with it.
Fourteen badges, no ladder
People can earn several, and no badge outranks another.
★ Observed
★★ Established
★★★ Demonstrated
Stars show proof strength, not skill level.
AI capability is hard to see from conventional signals.
Resumes, titles, and portfolios fail to show how someone actually works with AI. Overflow looks for observable decisions, repeated behaviors, and outcomes across real work.
Behavior over volume
Activity and cadence give context; the evaluator looks for what the person chose, changed, checked, and shipped.
Evidence over labels
Every badge points back to badge-specific proof, and a missing badge means the evidence did not clear its gate.
Visible evidence only
The evaluator never reconstructs private chain-of-thought. It inspects visible requests, revisions, tool use, verification, and recovery, separates the person's choices from an agent's mistakes, and labels every coverage gap.
Private by default. Reviewable before sharing.
Raw histories stay local. Only the finished, participant-reviewed profile is uploaded, and only if the participant opts in.
1 · Choose sources
The participant selects agent histories, GitHub activity, and optional career context, each with a dated coverage window.
2 · Extract locally
The evaluator reads retained evidence on the participant's machine and separates observed work from self-reported context.
3 · Build work arcs
Related sessions become real work arcs with a problem, decisions, artifacts, outcomes, dates, and confidence level.
4 · Apply badge gates
All 14 badges begin at zero. Qualifying proof, independent observations, outcomes, time span, and counterevidence determine each result.
5 · Show the whole map
Awarded badges, unawarded badges, coverage limits, and the proof needed for the next star are shown together.
6 · Owner review
The participant checks sensitive abstractions and factual claims, then decides whether to keep the HTML local or publish a profile.
What a badge says, and what it does not.
A badge is a compact index into evidence. It is useful only when its definition, proof, limits, and counterevidence travel with it.
What a badge says
- A defined behavior was observed, clearing the badge's positive gate.
- The proof strength is stated. Stars describe breadth, independence, outcomes, and time span.
- The claim is inspectable. A reader can follow it back to specific work arcs and the closest misses.
- The result is bounded by the source window.
What a badge does not say
- Not an intelligence, personality, or seniority score. Confidence, polish, verbosity, and fluency are not proxies.
- Not a hiring decision or certification. It starts a conversation about demonstrated work.
- Not a complete career history. Local tools retain different amounts, and missing evidence is not negative evidence.
- Not a ranking of worth or potential. Stars are never combined into one score.
Technical chops
How the person builds.
Technical chops★★T-01
Prototyper
Technical chops★★T-02
FrontendCrafter
Technical chops★★★T-03
ProductionShipper
Technical chops★★★T-04
SystemsArchitect
Technical chops★★T-05
ContextEngineer
Technical chops★★★T-06
AgentOrchestrator
Business know-how
What the person understands and translates, including industry and functional domain.
Business know-how★★★B-01
WorkflowArchitect
Business know-how★★B-02
ValueTranslator
Business know-how★★★B-03
ClearCommunicator
Business know-how★★B-04
AdoptionOperator
Good judgment
How the person decides, verifies, recovers, and maintains systems over time. Each one requires observable decisions.
Good judgment★★★J-01
VerificationFirst
Good judgment★★J-02
TradeoffNavigator
Good judgment★★J-03
RecoveryOperator
Good judgment★★★J-04
SystemsSteward
What one through three stars means
Stars describe the breadth, independence, and confidence of the evidence for one badge. They are never added into a person score.
Observed
One concrete badge-specific episode, or several observations inside one narrow arc. It proves the behavior happened.
Established
At least two distinct qualifying arcs, or one sustained 30+ day arc with three independent observations; at least one direct outcome and no unresolved material contradiction.
Demonstrated
At least three badge-specific arcs across two systems or organizations and 90+ days, with two high-confidence arcs and two direct outcomes.
How we keep badges honest
The assessment starts at zero for all fourteen badges and has to earn its way to each one.
Every badge gets a verdict
All fourteen are rated or explicitly not awarded. A whole pillar may come back empty.
Evidence is spent once
One piece of evidence supports one badge. The same work arc counts twice only when it shows two genuinely different decisions or outcomes.
The near misses are shown
Contradictions and closest misses are kept. Every one- or two-star badge names the evidence that would earn the next star.
What each badge requires
Every badge has a positive gate and a specific “not enough” rule.
Prototyper
Qualifies: repeatedly turns ambiguous ideas into functional artifacts a real user can try; two working-prototype-or-stronger arcs.
Not enough: mockups, plans, isolated experiments, or tool usage without a usable artifact.
Frontend Crafter
Qualifies: recurring visual, interaction, responsive, accessibility, or design-system judgment on interface work.
Not enough: React/CSS presence, hosted builders, generated imagery, or a polished screenshot without demonstrated decisions.
Production Shipper
Qualifies: repeatedly gets substantive changes into real use with merge, release, deployment, migration, or handoff evidence.
Not enough: local apps, staging links, prototypes, plans, or “ready to ship” claims.
Systems Architect
Qualifies: repeated consequential decisions about technical boundaries, data models, dependencies, integrations, migrations, or reliability that survive implementation.
Not enough: general systems language, broad ownership, diagrams, or technology selection without a consequential design decision.
Context Engineer
Qualifies: repeatedly structures, retrieves, refreshes, compresses, or hands off durable agent context through reusable mechanisms.
Not enough: long prompts, a large context window, copied background, or an installed memory tool.
Agent Orchestrator
Qualifies: decomposes work across agents, scopes ownership, integrates results, and verifies the combined outcome.
Not enough: spawning subagents, parallel calls, or token volume without synthesis and control.
Workflow Architect
Qualifies: maps actors, decisions, exceptions, controls, and data movement into a scalable business workflow or automation.
Not enough: generic automation, a connector, or a linear happy-path script without real process understanding.
Value Translator
Qualifies: repeatedly connects goals, economics, KPIs, customer needs, or operating constraints to technical scope and priority.
Not enough: requirements transcription, strategy language, or stakeholder contact without an observable outcome link or tradeoff.
Clear Communicator
Qualifies: across three substantive tasks, supplies a usable outcome, relevant context, constraints, and definition of done; corrections are specific and exploration closes in a decision.
Not enough: politeness, confidence, verbosity, grammar, native fluency, long prompts, agent agreement, or one unusually good brief.
Adoption Operator
Qualifies: repeatedly designs handoff, training, governance, review cadence, change management, or the operating loop after delivery.
Not enough: a demo, documentation alone, launch messaging, or a one-time handoff.
Verification-First
Qualifies: repeatedly uses independent checks before acceptance: tests, visual QA, source comparison, reconciliation, production checks, or second-agent audit.
Not enough: the producing agent saying it works, casual review, or one check performed only after failure.
Tradeoff Navigator
Qualifies: repeatedly chooses among real alternatives using value, cost, latency, risk, maintainability, reversibility, or time.
Not enough: preferences, tool loyalty, changing tactics without new evidence, or a decision with no stated consequence.
Recovery Operator
Qualifies: detects a real failure, diagnoses it, constrains retry or rollback, restores safe state, and changes the approach.
Not enough: ordinary iteration, repeated retries, or a successful first attempt.
Systems Steward
Qualifies: improves a live or established system over time through monitoring, maintenance, refactoring, reliability, or governance.
Not enough: shipping the initial release, claiming ownership, or returning once for an isolated fix.
System rule
Every badge has a separate evidence test. Prototyper is not a junior Production Shipper. Systems Architect is technical structure; Workflow Architect is business-process structure. Production Shipper gets work live; Systems Steward proves post-launch care.