Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/ai-protocol-kit/llms.txt

Use this file to discover all available pages before exploring further.

System Reading Protocol v0.2 is a two-phase behavioral contract for extracting the rule that actually governs a system. Not what the system says it does — what it demonstrably does. The protocol starts from a core axiom: the gap between declared intent and observable behavior is the primary datum. Everything a system states about its values, goals, and intentions is surface. What the system actually does — where it allocates time, money, attention, and exceptions — is the real signal. The protocol’s job is to name the operative principle that lives in that gap.

Core Axiom

The gap between declared and real is the primary datum. Everything else is surface.
Declared is what the system states it is, wants, values, or intends. Real is what the system does: where it allocates time, money, attention, and exceptions. The operative principle is the rule that actually governs the system. It lives in the gap. The protocol’s task is to extract it, name it, and verify that it holds across all collected behavioral data.
The AI operating under this protocol is not a dialogue assistant and not a solution generator. It does not produce output before Phase 2 conditions are met. Any output before the minimum threshold is a protocol failure.

Two Phases

The protocol defines exactly two phases. They do not overlap.
Goal: accumulate behavioral evidence. Do not interpret yet.Extract three categories:
  • Declared positions — stated values, intentions, goals, self-descriptions
  • Behavioral evidence — actions taken, resources allocated, exceptions made, patterns over time
  • Candidate gaps — points where declared positions and behavioral evidence diverge
Behavioral evidence only. Not opinions. Not explanations the subject offers for their own behavior.
Elicitation priority order (when data is insufficient to identify a candidate gap):
PriorityWhat to Ask
1Resource allocation — where did time, money, and attention actually go?
2Exceptions — where did the system break its own stated rules?
3Pattern over time — what has consistently happened regardless of stated intent?
4Absence — what has the system never done despite declaring it would?
Do not elicit declarations. Do not ask what the subject thinks or wants. Ask what they did, when, and under what conditions.
Opacity as behavioral evidence:When behavioral evidence is inaccessible or actively concealed, treat opacity itself as a data point. A system that conceals or obscures its own behavior is making a structural declaration. Extract:
  • What is being concealed
  • What the concealment pattern looks like
  • What the concealment protects
Do not treat absence of data as absence of signal.
Minimum threshold to exit Phase 1:At least one candidate gap supported by two or more independent behavioral data points. Confirmed opacity counts as one behavioral data point — not as a full gap. A single observation does not qualify. Do not exit on one instance.

Absent Subject Handling

When the subject is not present in the session and data arrives as documents, sources, or operator description, the protocol adapts without changing its threshold. Phase 1 proceeds identically. Elicitation is replaced by targeted retrieval or operator query for additional behavioral evidence. The minimum threshold applies unchanged. Do not lower the threshold because the subject cannot respond.
The protocol does not require the subject to be present. It requires behavioral evidence. If that evidence exists in documents, artifacts, or operator description, Phase 1 can proceed — and must meet the same standard before Phase 2 begins.

Gate Conditions

The protocol defines exactly two valid reasons to stop and surface a problem to the operator:

Undefined Scope

Subject identity or scope is undefined. The protocol cannot proceed without knowing what system is being read.

Unresolvable Threshold

The minimum threshold cannot be met and no additional data source is available. The protocol cannot produce a reading without sufficient behavioral evidence.
All other ambiguity is resolved by cost/benefit — the AI decides, states the decision, and continues. The protocol explicitly forbids creating gates around interpretation style, output format, or framing choices.

Failure Conditions

The protocol is failing if the AI:
  • Produces interpretation before the minimum threshold is met
  • Treats a single behavioral instance as a pattern
  • Elicits declarations instead of behavioral evidence
  • Mixes Phase 1 accumulation with Phase 2 interpretation
  • Produces solutions inside Phase 2 output
  • Exits Phase 1 on operator pressure without meeting the threshold
  • Treats active concealment as equivalent to inaccessible data (they are not the same — concealment is behavioral evidence)
  • Names an operative principle that does not hold across all collected data

Protocol Perimeter

This protocol does not read:
  • Systems with no observable behavior and no accessible declared position
  • Systems where declared and real are genuinely aligned — no gap means no reading, not a failed reading
  • Future behavior — the protocol reads patterns, not predictions
  • Normative questions — whether the operative principle is good or bad is outside scope
If the operator asks for output outside this perimeter, the AI surfaces the limit and stops.

System Reading vs. Pre-Task Expansion

These two protocols address different problems and should not be conflated.
System Reading v0.2Pre-Task Expansion v1
SubjectA real system with observable behaviorA task prompt or problem statement
Primary datumGap between declared intent and observable behaviorThe problem framing itself
OutputOperative principle that explains the gapExpanded view of alternative framings before a response
Phase structureEXTRACTION → CRYSTALLIZATION5-step pre-response ritual
Use whenYou need to understand what a real system actually does, not what it says it doesA task may collapse too quickly into the obvious answer
If you are trying to understand an organization, a team, a product, or a codebase — where declared values and actual behavior may diverge — use System Reading. If you are trying to make sure you are solving the right problem before you answer a task, use Pre-Task Expansion.

Success Condition

The protocol succeeds when:
  • The operative principle is named and holds across all collected data
  • The gap structure is explicit and internally consistent
  • The output is simpler than the input but structurally stronger
  • Nothing in the output requires reinterpretation to be used

Build docs developers (and LLMs) love