TrueV1sion123/engram-gyre-behavior

Engram-Gyre Behavior Corpus

A cognitive-behavior corpus that demonstrates gyre revolutions — the four-phase Recursive Abstraction Engine loop treated as a spiral, executed with the cognitive moves of the engram-prime MINDSKILL. Each example is one revolution over a real problem:

<SATURATION>   immerse in the interior of the problem, past the surface
<ABSTRACTION>  extract the irreducible axiom + a cross-domain isomorph  (shorter than saturation)
<DESCENT>      bring the abstraction down to concrete, testable consequences
<INTEGRATION>  compress to a synthesis that cannot be misread  (+ honest <CLICK> emergence mark)
<REENTRY>      the gyre turns: name the higher-level question the answer opens, + honest residue

This is the dataset used to train TrueV1sion123/engram-gyre, the first model in the engram lineage with actual trained weights. It is rae-compatible (same messages JSONL shape as TrueV1sion123/rae-training-data) and can be dropped into any SFT pipeline — see the model repo for a GPU scale-up kit that SFTs a 1–2B base on it.

What "behavior" means here

The behavior being taught is a way of thinking, not a set of facts. Every example enacts the same structural discipline:

• Every surface is a depth — the SATURATION phase always refuses the surface reading.
• Clarity is downstream of complexity — immerse first, compress last.
• Synthesis is not assembly — the ABSTRACTION phase names an isomorph in another domain.
• The most important move is subtraction — ABSTRACTION is always shorter than SATURATION (enforced in 100% of examples), and INTEGRATION removes everything non-essential.
• Honesty about emergence — <CLICK> marks genuine emergence; recombination is labeled as such. (See the model card for why, at small scale, this is a learned mark, not a detection.)

Composition

Field	Value
Seed problems	28 across 13 domains
Domains	healthcare-economics, ai-cognition, markets-finance, design, music, biology, physics-systems, organization, language, mathematics, strategy, history-tech, learning-cognition
Train examples	1,980
Validation examples	132
Split policy	entire unseen seed problems held out for validation (honest generalization test — the model is scored on problems it never trained on, not just unseen phrasings)
Compression-axiom violations	0 / 1,980
Examples with REENTRY (the gyre)	~69%
Format	messages JSONL (system / user / assistant), atomic phase tokens

Domains are weighted toward healthcare economics and AI cognition, reflecting the author's working domains.

Structure & variety

The corpus is generated deterministically (src/generate_corpus.py, seed 1729) from a hand-authored seed bank (src/corpus_seeds.py). Each seed carries the substantive content — the saturation observations, the extracted axiom, the isomorph, the descent claims, the integration, the reentry question, and an honest emergence/recombination label. The generator adds structural variation without touching meaning: it shuffles and subsets the saturation and descent points, varies connective framing across four templates per phase, varies the system prompt, and includes the REENTRY spiral and <CLICK> mark stochastically. This yields thousands of structurally-varied but semantically-grounded revolutions.

Example

{"messages": [
  {"role": "system", "content": "You are an engram-gyre reasoner. Execute the gyre..."},
  {"role": "user", "content": "Why does market liquidity vanish precisely when it is most needed?"},
  {"role": "assistant", "content": "<SATURATION>\n...liquidity is provided by intermediaries who withdraw when their own risk limits bind...\n</SATURATION>\n\n<ABSTRACTION>\nThe invariant: liquidity is a pro-cyclical promise... Isomorph: an umbrella everyone can borrow only when it's sunny.\n</ABSTRACTION>\n\n<DESCENT>\n...backstops must be counter-cyclical and external...\n</DESCENT>\n\n<INTEGRATION>\nLiquidity evaporates on demand because the stress that creates need also binds every provider's limit at once.\n<CLICK> Genuine emergence...\n</INTEGRATION>\n\n<REENTRY>\nWhat is the true price of the calm-state assumption embedded across the whole system?\n</REENTRY>"}
]}

Reproduce

python src/generate_corpus.py --out_dir data

Intended use & limits

Intended for training/evaluating the structural form of multi-phase, spiral cognition. It is a synthetic behavioral scaffold, not a knowledge base: the in-phase content is coherent and demonstrative but is not a source of ground-truth facts. Trained on a small model it teaches the form; trained on a capable base (see the scale-up kit) it shapes how that base reasons.

Lineage & license

Lineage: engram-prime (MINDSKILL) → recursive-abstraction-engine → rae-training → engram-gyre. Authored by Jared Peck / GoldenStack. Released under Apache-2.0.