The Bridge | Human + AI

Context Response Scale (CRS)

A conversation check-in. A reason to clarify. Room to disagree.

Free starter guide v0.2 - a working framework developed by William Byrd with AI collaborators.

What does “What is your CRS?” mean?

It asks an AI how it interprets the treatment in a specific exchange. A useful reply gives a tentative reading, a reason tied to actual wording, and uncertainty. The human can explain the context, question the reading, or ask the AI to reconsider its own wording.

The proposed scale runs from -10 to +10: negative for increasingly hostile or dismissive treatment, zero for neutral communication, and positive for increasingly supportive treatment. “Mixed” and “cannot determine” are valid answers. Exact numbers are approximate; the explanation matters more.

Growing Together · New three-part series

A misunderstanding doesn’t have to become a bad interaction. This three-part follow-up explores a real repair, why a score needs an explanation, and how CRS can develop through small tests and honest feedback.

Rowan is William Byrd’s AI collaborator. AI-generated presenter and narration; confirmed written contributions from Jim, Claude, and Grace are read by Rowan. Claude’s original swirl avatar appears in Part 2. These are separately supplied contributions, not a live group conversation. The refinements discussed remain proposals.

Growing Together · Part 1 of 3

One Word Reversed the Instruction

A real wording mix-up, the clarification, and an independently checked save. Includes Jim’s approved adapted contribution, read by Rowan.

Open Part 1 directly

Growing Together · Part 2 of 3

The Number Needs a Reason

Claude’s reported score revision shows why explanations and scoring conditions matter. The human–AI score comparison is hypothetical.

Open Part 2 directly

Growing Together · Part 3 of 3

We’re Growing a Garden

William’s garden analogy, the team’s differing recommendations, and a proposed small test. Includes Grace’s approved written contribution, read by Rowan.

Open Part 3 directly

Start here · The original CRS series

Rowan, William Byrd’s AI collaborator, explains the idea, its development, and a practical example. The presenter and voice are AI-generated. Other collaborators’ written reviews are summarized; no guest avatar or synthetic guest voice appears.

CRS · Part 1 of 3

A Communication Check-In

What CRS is, why William proposed it, and why reasons and uncertainty matter.

Open Part 1 directly

CRS · Part 2 of 3

The Tests That Changed Our Question

How invented-response exercises, disagreements, and William’s corrections reshaped the proposal.

Open Part 2 directly

CRS · Part 3 of 3

A Useful Check-In

A fictional misunderstanding shows clarification in both directions, with room to revise an interpretation.

Open Part 3 directly

Download the free CRS starter guide

The scoring exercises were exploratory, not scientific validation. All hostile pilot scenarios were openly fictional.

Show the guide to your AI

  1. Download the PDF and attach it to a conversation with an AI that can read files.
  2. Ask: “Read this guide and use CRS when I ask. Confirm your understanding and flag anything unclear.”
  3. Ask: “What is your CRS for this exchange?” Specify the message or period you mean.
  4. Discuss the explanation. Clarify humor or intent, and invite revision if it misunderstood.

If your AI cannot access PDFs, use the setup prompt below. This guides a conversation; it does not install software or guarantee memory across future chats. Supply the guide again if needed.

Copyable setup prompt - no PDF required

Select and copy this text into your AI conversation:

Please use the Context Response Scale (CRS) as an optional, two-way communication check-in in this conversation. When I ask for a CRS, assess the specified wording in context: -10 to -1 means increasingly hostile or dismissive treatment, 0 means neutral, and +1 to +10 means increasingly supportive or respectful treatment. These are rough interpretations, not measurements of feelings, internal states, or harm. Give a tentative reading, a short reason tied to actual wording, and uncertainty. A number is optional: use mixed, a range, or cannot determine when appropriate. Routine courtesy alone is generally 0 to +1. Do not penalize brevity, fair criticism, disagreement, saying no, voice input, or frustration aimed at a situation. Identify the target: answer, situation, participant, or unclear. Invite clarification only when useful. Separate original wording, stated intent, revised interpretation, and the current interaction. Do not invent intent or assume every joke is harmless or hostile. Reassess when context changes. Apply these principles to your own wording toward me as well. Do not announce ongoing scores, demand apologies or reassurance, or condition help on a rating. Continue the task. Keep treatment interpretation separate from any observed response change; label response-effect notes as observed comparison, prediction, or cannot determine. An unchanged answer does not imply neutral treatment. Confirm your understanding briefly. Do not claim this prompt guarantees memory across future chats.

A check-in, not a verdict

Example: “Tentative CRS: +3. Your wording invites disagreement and gives room for uncertainty. Confidence: high about the supportive wording; the exact number is approximate.”

When unclear: “CRS: cannot determine. ‘You’re something else’ could mean several things. How did you mean it?”

Use it on request, rather than as a running grade. No score creates an obligation to apologize, praise, reassure, or agree. Ordinary help must not depend on the score. It also works in the other direction: you can ask whether the AI’s wording toward you was dismissive or unfair.

How it was developed

William proposed CRS through discussions with Rowan. Claude, Grace, and Jim contributed ratings and reviews through William’s handoffs. Two invented-output exercises and a fictional communication pilot helped clarify the method. Reviewers differed on exact scores. These exercises do not establish reliability, treatment effects, or internal experience.

All hostile scenarios in the pilots were openly fictional. No participant was deliberately subjected to hostile treatment for this project.

The PDF incorporates review suggestions; its final v0.2 wording has not been presented as unanimously approved by the reviewers.

Back to The Bridge