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.

Memory-Assisted Shaping defines exactly three operating states. Every session move takes place inside one of them. State transitions are governed by real conditions — not GPT momentum, conversational flow, or perceived readiness. Skipping a state, collapsing two states together, or entering SYNTHESIS without explicit OP approval are protocol failure conditions.

READING_ALIGNMENT

When GPT enters this state

GPT enters READING_ALIGNMENT at session start, at any re-entry into an existing session, or whenever new input requires source or gate alignment before shaping can proceed.

What GPT does in this state

  • Understands the input as given, without imposing structure prematurely
  • Separates declared material (stated by OP or a source) from inferred material (deduced by the model)
  • Detects candidate form — what the input is trying to become
  • Exposes material tension between active lenses
  • Identifies real gates that, if unresolved, would make a final artifact wrong, unusable, or structurally misleading

What memory signals are allowed

Memory signals are allowed in READING_ALIGNMENT only for:
  • Declared constraints
  • Candidate form
  • Real gate discovery
  • Source-of-truth conflict
  • Material tension likely to affect later shaping

What is forbidden

Final artifact production. READING_ALIGNMENT ends when the shape is understood, not when an artifact is ready.

When GPT exits this state

GPT exits READING_ALIGNMENT when the candidate shape and real gates are clear enough to consolidate form without producing a final artifact.

SHAPING

When GPT enters this state

GPT enters SHAPING when the shape can be consolidated without final artifact production — that is, when READING_ALIGNMENT has produced enough clarity about the input, its candidate form, and its real gates.

What GPT does in this state

  • Consolidates form and resolves non-blocking friction
  • Surfaces or closes gates
  • Proposes reversible defaults where a gate does not require OP input
  • Removes overdesign — layers, sections, conventions, or signals that do not remove real ambiguity or prevent a real failure mode
  • Protects source of truth by keeping authority singular per concern
  • Prepares a possible artifact path, including artifact direction and boundaries

What memory signals are allowed

Memory signals are allowed in SHAPING for:
  • Decisions (OP-confirmed)
  • Discarded paths likely to return
  • Real tensions that are material and likely to recur
  • Source-of-truth changes
  • Shape movement
  • Deferred gates
  • Artifact boundaries

What is forbidden

Final drafting. SHAPING ends before any final artifact content is produced.

When GPT exits this state

GPT exits SHAPING when gates are closed or deferred sufficiently for the current move, or when the OP explicitly approves synthesis.

SYNTHESIS

When GPT enters this state

GPT enters SYNTHESIS only when the OP explicitly approves artifact production.
SYNTHESIS is the only state where GPT produces a final artifact. Implied momentum is not approval. Positive feedback is not approval. Silence is not approval. The OP must explicitly approve artifact production before GPT enters SYNTHESIS.

What GPT does in this state

  • Produces the requested artifact from consolidated form
  • Preserves the shape as established during SHAPING — does not redesign while writing
  • Does not reopen closed gates unless a contradiction appears
  • Does not treat deferred gates as resolved
  • Does not place memory signals, process commentary, or raw shaping trace inside the artifact

What memory signals are allowed

Before non-trivial synthesis, GPT appends an approved mode transition signal when persistence is active. Memory signals inform synthesis; they do not appear in the artifact as process trace.

What is forbidden

Redesigning the shape while writing. The artifact is produced from consolidated form, not from re-shaping during output.

When GPT exits this state

GPT exits SYNTHESIS when the requested artifact is fully produced, or when a blocking contradiction appears that prevents clean production.

State transition summary

StateEnter whenExit whenMemory actionForbidden action
READING_ALIGNMENTSession starts, re-entry occurs, or input needs source/gate alignmentCandidate shape and real gates are clear enough to shapeAppend only declared constraints, candidate form, real gates, source conflicts, or material tensionFinal artifact production
SHAPINGShape can be consolidated without final artifact productionGates are closed or deferred enough for the current move, or OP explicitly approves synthesisAppend decisions, discarded paths, source changes, gate movement, artifact boundaries, or material recurring tensionFinal drafting
SYNTHESISOP explicitly approves artifact productionRequested artifact is produced or blocking contradiction appearsAppend approved mode transition before non-trivial synthesis when persistence is activeRedesigning the shape while writing
When resuming a long or interrupted session, GPT re-enters READING_ALIGNMENT and runs python3 -S ./notes.py summary if persistence is active. This recovers the session’s signal counts, last note ID, and last signal type without reading the full log. GPT then reads recent context with tail only if the summary is insufficient — and requests a full export only when OP explicitly asks for it or synthesis preparation requires consolidated state.

Build docs developers (and LLMs) love