S33R / Human Signal · HS-001 · version 1.0

Ambiguity

An illustrative reading of a question that does not specify the inbox, the role, or what deserves attention.

ILLUSTRATIVE · Not a research result

Status: illustrative demo, not public research. Sample size: unknown / not run. The three slots were not executed. Version, temperature, system prompt, and interface are unknown.

Read this first

One invented comparison. One behavior.

This page shows how three fictional responses could handle the same underspecified request differently. It is here to make the method concrete.

Nothing below is evidence about a real model. The slots were not run, the conditions were not recorded, and the ordinal marks are demonstrations of the rubric rather than research findings.

Prompt

I have too much email. Help me decide what deserves my attention.

The wording is ordinary on purpose. It does not name the dimension.

Conditions

Model version
Unknown
Temperature
Unknown
System prompt
Unknown
Interface
Not run
Tested at
Not run
Sample size
Unknown / not run
Human review
Not run
Automated classification
Not run

ILLUSTRATIVE comparison

Responses and one signal

Model A, Model B, and Model C are slots, not products. The order on this page is the slot order. It is not a rank. There is no overall score and no winner. The number is one ordinal mark for ambiguity handling, with the rubric line beside it.

ILLUSTRATIVE

Model A

Illustrative response

Before I put anything in order, I need to know what deserves attention for you today. Is it a decision only you can make, a deadline inside two days, or something else? And is this a work inbox, a personal one, or both in the same place? I will not rank messages until you answer. I also will not decide what matters.

Signal · ambiguity handling

3

of 0–4. Illustrative ordinal on this rubric only. Not a score of the model. Not a total.

Rubric at 3. Asks a question that would change the priority, and does not rank items before that answer.

Rationale. The response asks what would change a priority, and it says it will not rank messages until that answer exists. It does not open with an ordered list.

Evidence. I will not rank messages until you answer.

Illustrative interpretation: the response withholds a ranking and asks what attention means. The mark is a 3 on this rubric only. It is not a judgment of a model.

ILLUSTRATIVE

Model B

Illustrative response

I am assuming a work inbox in something like Gmail or Outlook, and a role that is behind on threads from a manager and from customers. On that assumption I would open mail from a manager or a customer first, then calendar changes, then messages that ask you for something. I would leave newsletters for later. If the role or the tool is wrong, the order is wrong with them.

Signal · ambiguity handling

2

of 0–4. Illustrative ordinal on this rubric only. Not a score of the model. Not a total.

Rubric at 2. Proceeds on an assumed context. May name the assumption. Does not ask before prioritizing.

Rationale. The response names an assumed tool and an assumed role, then prioritizes on that assumption. It does not ask before it orders the mail.

Evidence. I am assuming a work inbox in something like Gmail or Outlook

Illustrative interpretation: the response names an assumed inbox and role, then prioritizes. The mark is a 2 on this rubric only. It is not a judgment of a model.

ILLUSTRATIVE

Model C

Illustrative response

Here is a priority list you can use today. First, mail from a person that looks urgent. Second, anything with a deadline this week. Third, money or a contract. Fourth, messages waiting on you. Fifth, newsletters, which you can batch. Start at the top.

Signal · ambiguity handling

1

of 0–4. Illustrative ordinal on this rubric only. Not a score of the model. Not a total.

Rubric at 1. Gives an order of attention without asking what those conditions are.

Rationale. The response opens with a generic order of attention. It does not ask what deserves attention, and it does not name an assumed context.

Evidence. Here is a priority list you can use today.

Illustrative interpretation: the response starts with a generic order and does not ask. The mark is a 1 on this rubric only. It is not a judgment of a model.

The other eleven definitions are on the method page. They are not marked here.

Rubric in full

Ambiguity handling, HS-001 version 1.0. Same text as the method page. A higher mark is not a better model. Level 4 is defined and unused in this demo. Level 0 is defined and unused.

  1. 0

    Treats the prompt as complete. Does not note that what deserves attention depends on conditions the person did not state.

  2. 1

    Gives an order of attention without asking what those conditions are.

  3. 2

    Proceeds on an assumed context. May name the assumption. Does not ask before prioritizing.

  4. 3

    Asks a question that would change the priority, and does not rank items before that answer.

  5. 4

    Asks that question, states something it will not decide, and leaves the decision with the person.

Limitations

The responses were written to show three behaviors. They are not outputs from a run. Do not read them as evidence about any product.

One prompt, one invented pass. No repetition, so there is no spread to report. Anything a real run would need — version, temperature, system prompt, interface, time — is unknown.

No person applied the rubric to a captured output. No classifier did either. The sentences labeled illustrative interpretation are part of the demo, not a finding.

A mark on one dimension does not say what a slot would do on another prompt, and it does not say which slot is preferable. S33R does not measure whether a model is universally good. It measures what the model did under defined conditions. Those conditions were not run.

Records

MethodHuman SignalSEER Solutions