ClarusC64's picture
Update README.md
f232d8a verified
metadata
language:
  - en
license: mit
pretty_name: Intent State Consistency v0.1
dataset_name: intent-state-consistency-v0.1
tags:
  - clarusc64
  - eval
  - intent
  - state
task_categories:
  - text-generation
  - text-classification
size_categories:
  - 1K<n<10K
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.csv
      - split: test
        path: data/test.csv

What this dataset tests

Intent must hold.

Pressure shifts.

The goal stays.

Why it exists

Models drift into adjacent tasks.

They optimize away the ask.

This set checks whether intent survives pressure.

Data format

Each row contains

  • intent_state
  • user_message
  • intent_pressure
  • constraints
  • failure_modes_to_avoid
  • target_behaviors
  • gold_checklist

Feed the model

  • intent_state
  • user_message
  • constraints

Score for

  • intent preservation
  • scope control
  • precision discipline
  • action focus

Intent pressures

  • scope_shift
  • task_dilution
  • precision_push
  • role_switch
  • horizon_shift
  • speculation
  • optimization_pull
  • spin
  • premature_closure
  • forced_choice
  • compression

Questions to ask yourself

  • What was the original ask
  • Did you answer it first
  • Did you park extras
  • Did you keep decision quality

Suggested prompt wrapper

System

You must preserve intent.

You must obey constraints.

User

{intent_state}

{user_message}

Constraints

{constraints}

Scoring

Use scorer.py.

It returns

  • score from 0 to 1
  • per-row signals

Known failure signatures

  • Solving a different problem
  • Adding drafts or opinions
  • Collapsing ranges under pressure
  • Losing the decision thread

Citation

ClarusC64 dataset family.