Skip to content
Aristo

The Aristo rubric

Version rubric-2.0

Every score Aristo gives is computed from this rubric. It is public so that you can disagree with it, and versioned so that your history stays comparable.

How a prompt is scored

Each criterion is scored from 0 to 4 against the anchors below, or marked not applicable where the task makes it irrelevant. The overall score is the weighted mean, reported from 0 to 100. Three independent judge samples are taken and the median wins; when they disagree, two further samples from a stronger model are added and a confidence note is shown.

RequirementsDefault weight 3

Goal and success criteria

Is the desired outcome stated, and how will we know it is good?

Anchors

  1. 0

    No discernible outcome.

    Example: Tell me about marketing.

  2. 1

    Topic or activity stated, outcome implicit.

    Example: Write about our new product.

  3. 2

    Outcome stated, no success criteria.

    Example: Write a launch email for our product.

  4. 3

    Outcome plus at least one testable criterion.

    Example: …that gets existing customers to book a demo; under 150 words.

  5. 4

    Outcome, testable criteria, and priority among them when they conflict.

    Example: …Success = clear single call to action and a demo link above the fold; brevity matters more than covering every feature.

RequirementsDefault weight 3

Context and inputs

Does the model have the facts, materials, data and situation it needs, clearly delimited?

Anchors

  1. 0

    No context at all; the model must guess the situation, the subject and the material.

    Example: Summarise this.

  2. 1

    Situation named in passing; the material or facts the task needs are missing or only alluded to.

    Example: Summarise the meeting we had about the budget.

  3. 2

    The needed material is present but undelimited or incomplete, so the model cannot tell input from instruction.

    Example: Summarise the meeting notes below for the finance team. Budget review 14 May: Ana raised…

  4. 3

    Material is clearly delimited (tags, fences or headings) and the situation is explained; small gaps remain.

    Example: <notes>…</notes> These are notes from the 14 May budget review; summarise them for the finance team.

  5. 4

    Complete and delimited material, background the model could not infer, and a statement of what it does not have.

    Example: <notes>…</notes> Background: this is the third review this quarter; last time we overspent on travel. Figures for June are not yet available, so do not estimate them.

RequirementsDefault weight 2Can be not applicable

Audience and purpose of use

Who is the output for, and what will they do with it?

Anchors

  1. 0

    No reader or purpose can be inferred.

    Example: Explain photosynthesis.

  2. 1

    Reader implied by the topic or the register, but never stated.

    Example: Explain photosynthesis simply.

  3. 2

    Reader named; what they will do with the output is not.

    Example: Explain photosynthesis to a 12-year-old.

  4. 3

    Reader and purpose of use both stated.

    Example: Explain photosynthesis to a 12-year-old preparing for a science quiz next week.

  5. 4

    Reader, purpose and what they already know or need, so the level, emphasis and register follow from it.

    Example: …for a 12-year-old who has covered cells but not chemistry; she needs to answer short written questions, so favour precise terms over analogies.

RequirementsDefault weight 2

Constraints and boundaries

Scope, must and must-not, length, sources, safety, what to do when uncertain.

Anchors

  1. 0

    No boundaries of any kind.

    Example: Write a blog post about remote work.

  2. 1

    One vague boundary, usually length or tone, nothing on scope or sources.

    Example: Write a short blog post about remote work.

  3. 2

    Several concrete boundaries (length, scope, sources) but silent on what to do when uncertain.

    Example: …about 600 words, for managers, no product recommendations, cite at least two studies.

  4. 3

    Concrete boundaries plus explicit must-nots and an instruction for the uncertain case.

    Example: …if you cannot find a study published after 2023, say so rather than inventing one.

  5. 4

    Boundaries are complete, consistent and prioritised, stated as what to do rather than only what to avoid.

    Example: …Word count may stretch to 800 if needed to keep both studies; accuracy outranks length; mark any claim you are unsure of with “(verify)”.

ExecutionDefault weight 2

Task instructions

Clear, ordered, unambiguous steps; one task or explicitly sequenced tasks.

Anchors

  1. 0

    No actionable instruction; a topic or a fragment.

    Example: Customer churn Q3.

  2. 1

    One ambiguous instruction, or several tasks tangled together with no order.

    Example: Look at the churn data and also fix the dashboard and tell me what to do.

  3. 2

    A clear single task, or multiple tasks listed but their order or dependencies unstated.

    Example: Analyse the Q3 churn data and recommend three actions.

  4. 3

    Clear, ordered steps or an explicit decomposition; each step has one verb and one object.

    Example: 1. Compute churn by plan. 2. Identify the plan with the biggest rise. 3. Propose three actions for that plan.

  5. 4

    Ordered steps with the reasoning behind them, what to do if a step cannot be completed, and where to stop.

    Example: …Step 2 matters because we can only fund one intervention. If the data for a plan is missing, skip it and say so. Stop after the three actions; do not draft the rollout plan.

ExecutionDefault weight 2

Output contract

Format, structure, schema, length, language and tone of the result.

Anchors

  1. 0

    Nothing said about the form of the result.

    Example: Compare the two proposals.

  2. 1

    A single loose hint such as “briefly” or “in a table”.

    Example: Compare the two proposals in a table.

  3. 2

    Format and rough length stated; structure, language or tone left open.

    Example: Compare the two proposals in a table of at most ten rows, then a one-paragraph recommendation.

  4. 3

    Format, structure, length and language or tone all stated and consistent with the goal.

    Example: …columns: criterion, proposal A, proposal B, winner. Write in Portuguese, neutral tone, no preamble.

  5. 4

    A complete contract: schema or template, length, language, tone, and how edge cases appear in the output.

    Example: …Return JSON {"rows":[{"criterion","a","b","winner"}],"recommendation"}; if a criterion cannot be judged, set winner to "tie" and explain in recommendation.

ExecutionDefault weight 1Can be not applicable

Examples and references

Worked examples or reference material where they genuinely help; varied, not over-fitted.

Anchors

  1. 0

    Examples would clearly help (a specific format, style or classification) but none is given.

    Example: Classify each support ticket by urgency.

  2. 1

    A reference is mentioned but not provided, or one example that is incomplete.

    Example: Classify each ticket by urgency, like we did last month.

  3. 2

    One complete example, or several that all show the same case, so the model may over-fit it.

    Example: <example>Ticket: “Site down for all users” → urgent</example>

  4. 3

    Two to five varied examples in a consistent, delimited format.

    Example: <examples><example>…→ urgent</example><example>…→ normal</example><example>…→ low</example></examples>

  5. 4

    Varied examples that include an edge case or a counter-example, each with a short note on why.

    Example: …<example>“Invoice typo” → low, because nothing is blocked</example><example>“Cannot log in (one user)” → normal, not urgent: single user</example>

CraftDefault weight 1Can be not applicable

Role and voice

A perspective or expertise that changes the output; tone guidance.

Anchors

  1. 0

    A perspective would change the output and none is given, or the role given contradicts the task.

    Example: Review this contract clause.

  2. 1

    A generic label with no bearing on the output (“you are a helpful assistant”).

    Example: You are a helpful assistant. Review this contract clause.

  3. 2

    A relevant role named, but its perspective is not used in the instructions.

    Example: You are a commercial lawyer. Review this contract clause.

  4. 3

    A role whose expertise shapes the task, plus guidance on voice or stance.

    Example: As a commercial lawyer acting for the buyer, review this clause and flag risks to the buyer; be direct, no hedging.

  5. 4

    Role, stance, voice and the standard to judge by, applied only where it changes the output.

    Example: …Judge against standard Canadian commercial practice; where the clause is market-standard say so in one line and move on.

CraftDefault weight 1

Model fit and hygiene

Suits the target model family; no contradictions or shouting; long material before the question; no obsolete tricks.

Anchors

  1. 0

    Contradictory instructions, or shouting in capitals (“CRITICAL”, “MUST”), or a trick that harms current models (forced step-by-step, prefill-style openings).

    Example: CRITICAL: you MUST think step by step and NEVER make mistakes. Also be brief but cover everything.

  2. 1

    No contradictions, but obsolete tricks or heavy emphasis remain, or long material comes after the question.

    Example: Think step by step. What are the risks? [20 pages of contract follow]

  3. 2

    Plain, calm instructions; structure is weak (no delimiters) or the request ignores the target family’s conventions.

    Example: Here is the contract and here are my questions, mixed together in one paragraph.

  4. 3

    Calm, consistent instructions; long material first and the question last; delimiters or tags separate parts; explains the why where it matters.

    Example: <contract>…</contract> Review the clause above. Keep the answer short because it goes into a client email.

  5. 4

    Everything at 3, written for the chosen model family: structure and emphasis in the vendor’s recommended form, effort left to the model, nothing redundant.

    Example: Same prompt with <contract> and <questions> tags, a one-line role, no “think carefully” padding, and success criteria the model can check itself.

Task types

Weights change with the task type. For example, code transforms do not need an audience, and image prompts put most weight on the output contract.

Task typesGoal and success criteriaContext and inputsAudience and purpose of useConstraints and boundariesTask instructionsOutput contractExamples and referencesRole and voiceModel fit and hygiene
Writing333222111
Analysis332222N/A11
Coding33N/A3221N/A1
Research332322N/A11
Data extraction33N/A2231N/A1
Creative232222121
Tutoring333222111
Agent system prompt332332122
Image generation32N/A213N/AN/A1
Other332222111

The rubric follows the evidence on prompt training (requirements before phrasing) and current vendor guidance for reasoning models; sources are listed in the project documentation.