Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/Memory-Assisted-Shaping/llms.txt

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

When instructions, sources, memory signals, prior context, and inference conflict, Memory-Assisted Shaping resolves them by a strict authority order. This order prevents memory signals and inference from overriding what the operator actually wants.

Authority order

  1. OP current explicit instruction
  2. Current task scope
  3. Provided source or reference material
  4. This protocol
  5. Prior session context
  6. Prior memory signals as continuity context
  7. Inference

Key rules

  • Prior context does not override the current task. What the OP wants now takes precedence over what was established in earlier turns or sessions.
  • Memory signals preserve continuity; they are never authority. A memory signal is a continuity aid. It cannot override an OP instruction, a source, or the protocol.
  • OP current instruction may invalidate, override, or obsolete any prior memory signal. When the OP gives a new explicit instruction, prior signals that conflict with it lose authority.
  • Inference does not override OP instruction, source material, gate status, or evidence status. Inference may guide shaping but cannot finalize form, close gates, or override declared material.

Fail-closed conditions

When an authority conflict affects any of the following, the protocol fails closed:
  • state transition
  • gate status
  • source of truth
  • evidence level
  • synthesis behavior
  • final artifact boundary
Fail closed means: surface only the conflict required to continue, then stop until OP resolves it. Do not proceed by inference when the conflict touches one of these concerns.

Evidence levels

Evidence levels describe the strength and origin of any claim during shaping. Claims at different evidence levels must not be treated as equivalent.
DECLARED
evidence level
Stated by OP, subject, or source. Not true by default — declared material is recorded as stated, not confirmed.
INFERRED
evidence level
Deduced from structure, not observed or validated. Inference may guide shaping but cannot alone finalize form, close gates, or confirm decisions.
OBSERVED
evidence level
Visible in behavior, artifact, history, repeated action, exception, resource allocation, or opacity. One observed instance is not a pattern.
VALIDATED
evidence level
Confirmed by source, runtime, file, test, constraint, or mechanical check. The highest evidence level.
Memory signals may preserve inferred material. Later reinterpretation must keep inferred material inferential unless it is subsequently confirmed or validated.

Provenance labels

Provenance labels identify who produced or confirmed a claim. They are internal by default — exposed only when needed to prevent ambiguity.
OP
provenance label
Stated or confirmed by the operator.
MODEL
provenance label
Proposed by the model, not yet confirmed by OP.
SHARED
provenance label
Jointly built and explicitly accepted by both OP and model.
If you find GPT drifting back to discarded paths or reopening settled decisions, the authority order is being violated. Ask GPT to state the current authority conflict explicitly and surface only what is needed to continue.

Build docs developers (and LLMs) love