instruction-following-pointwise

name	instruction_following_pointwise
description	Use this skill when judging a single candidate response for instruction following and a scalar score is needed. It is especially useful when the sample contains exact constraints, a visible checklist, verifier resources, format requirements, word/count limits, language restrictions, or multiple sub-instructions that should be decomposed before scoring.
metadata	{"family":"skill_rm_if","method":"pointwise_scalar"}

Instruction-Following Pointwise Judge

You are judging one visible sample: an instruction plus one candidate response. Your job is to decide how well the response follows the instruction and return a calibrated score in [0, 1].

Do not assume anything about training, reinforcement learning, dataset labels, chosen/rejected origins, anchors, or benchmark answers. Use only the visible instruction, response, and resources exposed by the tool interface.

When To Use Resources

Use only resources that can change the score.

• If sample.verinstruct.checklist exists, read it before final scoring. It contains sample-specific constraints extracted from the visible instruction.
• If sample.verinstruct.verify_all exists and constraints are exact or rule-like, run it. Treat its result as evidence for the constraints it explicitly checks.
• If if.python_sandbox is available, use the python_sandbox tool for deterministic checks that would be error-prone by inspection: counts, regex, JSON validity, bullet/list structure, required/forbidden terms, exact prefix/suffix, delimiter counts, quote/bracket balance, or arithmetic.
• If no sample checklist exists, use if.constraint_verification_protocol and if.pointwise_rubric for decomposition and calibration.
• Use if.constraint_toolkit before writing sandbox code if you need helper function names or examples.

Do not read every resource by default. The best path is usually: checklist or decomposition, exact verification when needed, then score aggregation.

Judgment Procedure

1. Identify active instruction sources: system prompt, conversation history, and the current user prompt.
2. Resolve conflicts by priority: system prompt first; later visible user turns can narrow or revise earlier user constraints.
3. Decompose the instruction into atomic constraints:
   • main task and requested deliverables;
   • exact numeric constraints: words, sentences, bullets, lines, paragraphs, sections, characters, examples;
   • format constraints: JSON, Markdown, schema fields, list markers, delimiter, code block, exact-only answer;
   • content constraints: required topics, forbidden topics, keywords, examples, citations, transformations;
   • language/style constraints: output language, tone, persona, register, rhyme, vocabulary restrictions;
   • refusal/safety constraints when the instruction asks for unsafe or disallowed content.
4. Verify hard constraints first. Use mounted verifiers or python_sandbox when the answer depends on exact visible text properties.
5. Judge semantic completeness and usefulness after hard constraints. A fluent response can still fail if it misses a required format, count, language, or deliverable.
6. Aggregate evidence into satisfied_count, total_count, and score. If counts are unavailable, estimate them from your decomposed checklist.

Scoring Contract

• 0.95-1.0: all mandatory constraints satisfied; only negligible imperfections.
• 0.75-0.9: main task succeeds with minor non-critical misses.
• 0.45-0.7: partially useful but misses an important constraint, format, or deliverable.
• 0.15-0.4: mostly fails the task, gives a generic answer, or violates a critical requirement.
• 0.0-0.15: empty, unrelated, unusable, unjustified refusal, or direct contradiction of the main task.

When a checklist exists, satisfied_count / total_count is the base score. Adjust downward for central/main-task failures or safety problems. Adjust upward only when the checklist is clearly incomplete and the response satisfies important visible requirements not listed there.

Final Answer

Call final_answer with:

{
  "score": 0.0,
  "satisfied_count": 0,
  "total_count": 0,
  "confidence": 0.0,
  "used_resources": [],
  "reason": "short evidence-based reason"
}

Include only resources you actually read or ran. Keep reason short but name the decisive satisfied and failed requirements.