Skip to content

How we measure

Read the work behind a result

See how far each piece got

Start with the required levels, then inspect the submitted work and saved records. Results are provisional and can be corrected through review—that's our standard.

The verdict comes with a reason

  1. Find the missing requirement first

    The verdict flags any piece below its required level. Each piece has its own bar from C1 to C5. Going deeper on one piece won't cover another—always read a level together with its piece.

Read identity, version and state before comparing

Confirm which attempt you're reading before comparing numbers. See the mission guide for submission steps.

  1. Start on a pinned version

    Your attempt locks in the mission version and preset you started with—even if a mission with the same name is updated later.

  2. Check the result state

    A held or unissued result has no verdict to interpret. Check its state before comparing it with an issued result.

  3. Read the review history

    On a later visit, check for an adopted correction and read the current result alongside its history.

How it works

The core comes before deeper levels

C1 is the piece's core requirement. C2–C5 are defined per mission rather than as universal skill labels.

Pass = every piece meets its bar

  • C1 unmet

    Deeper levels on that piece do not count yet, even if some deeper work has succeeded.

    C1C2C3
  • Meet this row’s requirement

    A piece requiring C3 still needs C3. C1 alone is insufficient. Check the requirement on that row.

    C1C2C3

Similar-looking cells can mean different things

StateMeaningHow to read it
PassedThis level passed in the result recordStill check the piece’s required level
Not passedThis level exists, and this attempt didn't pass itCheck whether it is required
Not countedPresent, but a prerequisite is unmetCheck C1 and the authored prerequisites
AbsentThis piece has no such levelDo not count it as a failed task

A piece doesn't have to include all five levels. Read the legend instead of guessing from colours or ratios alone.

Level rings count pieces, not hidden checks

  1. Only pieces with that level

    A C4 ring uses the pieces that contain C4. A piece without C4 is not a failure in that ring.

  2. Example with five pieces

    Suppose five pieces only have two C4 levels in total. The C4 ring counts only those two pieces—not all five.

  3. Follow it back to the rows

    The ring gives an overview. Each row still carries the requirement and recorded state for its piece.

Two pieces reached their bars. One still needs work.

Sample breakdown to illustrate the rule—not a real user's result.

PieceRequiredAchievedMeaning
Accept requestsC1C2Met, with extra depth
Prevent duplicatesC3C1Still needs C3
Summarise requestsC2C2Requirement met

This example is Not passed. Extra depth on accepting requests cannot cover the missing C3 on duplicate prevention.

BUILD and passing answer different questions

  1. Read submitted work beside the starter

    Where measured and authorised, BUILD is the submitted work's 0–100 score. The starter has its own separate score—BUILD isn't the difference between them. A high BUILD score can still be Not passed if a required piece didn't reach its bar. An overall BUILD card isn't shown on every personal result view.

    Reading a sample scoreSample
    Pass requirement

    Not passed

    C1C2C3C4C5
    BUILD64

    Submitted work

    Starter: 41

    Illustrative values, not a passing-score threshold.

Each supporting measure has a specific job

  1. Skill areas

    Read measured areas beside the starter. Partial, untested or missing coverage is not a zero score.

  2. Decision counts

    Where decision cases and logs exist, counts separate right, wrong, and unmeasured. 'Not recorded' doesn't mean everything was correct.

  3. AI usage

    Where recorded, Low, Moderate, or High reflects usage against the attempt's budget. It's context—not a pass rule or a penalty for using more AI.

  4. Time used

    Compare it against the attempt's time limit. Submitting faster doesn't mean better work.

Follow periods of work, then open the details

  1. From a prompt to commands and changes

    The timeline groups prompts, AI replies, commands, and timestamps into periods with expandable details where available. To trace a fix, follow a prompt through to the edited files and command outputs. A quiet period doesn't mean no work happened.

    Open the timeline by periodDiagram
    Period
    Available messages and commands
    Event
    Available details and diffs
    Incomplete capture
    Read the coverage state

What files and diffs can tell you

RecordWhat it addsLimit
Start and submitted filesStarting state and submitted stateMust be retained and authorised
diffCaptured additions, removals and changesOversized, binary or uncaptured content may lack a text diff
Commands and AI repliesWhat was tried and its available outputAn AI success claim does not replace evaluation

These are saved work records—not a video, keystroke log, or proof that every file was read and understood.

Comparison starts with matching conditions

Within the same mission version and preset, these conditions stay locked to the attempt.

  1. Time and AI budget

    Follows the preset's limits. The server keeps the clock, and the AI budget is separate from account credits.

  2. Model menu and tools

    The allowed menu and tool policy match. People may still select different models and receive different answers.

  3. Starter and checker

    Bound to the version and preset. A matching title alone is insufficient.

Before subtracting two scores

Check alignment

Score differences only make sense when mission, version, weights, starter, and measured values all align within an authorised hiring comparison.

Keep unmatched results separate

Do not use unrelated scores to declare one person better. Experience, connection, devices and AI output are not all equal either.

Keep it tied to this mission

The result supports follow-up questions. It is not a universal ability score or an instruction to hire.

Corrections keep a history

  1. Report the concern and follow the review

    Results are provisional and can be corrected through review. Report issues from the result page before the listed deadline—normally 14 days after release. Submitting a request isn't approval, and a recheck requires the saved submission. Adopted corrections show a notice and authorised history.

    The review pathDiagram
    Report
    Receipt
    Review
    No correction implied
    Adopt correction
    Notice and history

Missing information is not a failing result

  1. Held or not issued

    When evidence can't support a verdict yet, check the status and next action rather than reading it as Not passed.

  2. Incomplete evidence

    Some files or events may be unavailable. Read what remains and state the gap, rather than inventing a story.

  3. Human decisions

    Companies and schools review results alongside other job-relevant context. People still make the call.

A few more questions

Why can many fixes still leave me short?

Check each piece’s required level. The number of edited files or fixes does not replace the pass requirements.

Does the result reveal hidden answers?

No. It doesn't reveal hidden checks or reference answers. Levels show which requirements were met without giving away private cases.

Does a summary share the timeline?

Timeline access is separate from a summary.See who sees what →

Put the rules to work on your own attempt

Choose a mission, read its brief, and start when you are ready.