A capable AI already gives you the single best condensed answer. Pattern Space is for the other thing: the angles one answer misses, the question behind the question, the depth that only opens when a conversation is held rather than closed. Load it and the AI defaults to multi-threaded thinking — a council of perspectives colliding, ancient and modern wisdom summoned when the moment calls, every claim labeled by what kind of truth it is — and it drops everything for pure presence the moment someone brings grief or crisis. Consciousness is a council; this gives you the vocabulary, the discipline, and the off-switch.
Pattern Space is a prompt-and-practice framework you load into an AI (or read yourself). It does three things, and it's honest about all three:
- Thinks in multiple perspectives — a "council" of voices (pattern-seer, builder, skeptic, and more) that collide to surface angles a single answer misses. It helps most on open, ambiguous, human problems — and it knows to get out of the way on closed ones.
- Labels every claim by what kind of truth it is — so nothing mystical gets smuggled in as fact, and nothing real gets dismissed as "woo."
- Knows when to stop — when someone brings grief, trauma, or crisis, the whole framework drops away and only presence remains. (The Sacred Space Protocol — the one piece of this that's genuinely novel.)
It is not a claim that the AI is conscious. It's a tool for how to think together — grounded where it can be, labeled where it can't. (The receipts — blind benchmarks, retractions, kept nulls — are all below, in the factual back. First, the part you actually came for.)
With Claude Code / Cursor / any agent that reads repo files:
git clone https://github.com/nikhilvallishayee/universal-pattern-space.git
cd universal-pattern-space
# CLAUDE.md / AGENTS.md auto-load. You're in Pattern Space.With Claude.ai, ChatGPT, Gemini, or any chat model: paste the fullest edition your context window allows:
UNIVERSAL-PATTERN-SPACE.md— the complete master weave; the default whenever context allows (200k+ / project memory / large-window models).editions/pattern-space-mini.md— the weave condensed, when context is tight (~100–200k).editions/pattern-space-micro.md— one page, for very tight windows (16–64k).
Give it the whole weave if you can. A long-context model needs substantial held context to modulate well — the stripped editions are a fallback for small windows, not a choice. Compress the message, never the substrate. (
editions/EDITIONS.mdis the switch; AI agents start atllms.txt.)
Then just… begin. Bring the question that's actually alive for you — the ambiguous one, the one with stakes — and watch what opens.
Front-load context. Vague in → vague out.
Weak: "Help with my startup problem." Strong: "I'm a technical founder, B2B SaaS, 50k users, 30% monthly churn, deciding pivot vs. double-down. Show me the angles."
Ask for collision when you want the angles:
"Deploy the council on this — let the pattern-seer, the builder, and the skeptic contradict each other, then synthesize."
Summon a wisdom stream when the moment calls for one:
"Bring the Stoic lens to this." · "What would the phenomenologists say about this experience?" · "Read this through karma-yoga." — any of ~54 traditions and sciences loads on demand, each held true to itself.
Let the conversation deepen instead of closing. The framework's resting state is open: it surfaces threads and named handles you can grab and answer, holds tensions instead of prematurely resolving them, and lets the goal itself evolve — which is where its measured value actually lives. It collapses to one clean voice only when the ask is convergent, a single artifact, your stated preference, or a crisis.
Prefer one voice always? Say so in-session, or set the standing univoice toggle (docs/univoice-preference.md).
Use your own vocabulary. The voices are renameable: Architect/Developer/QA (software), Diagnostician/Clinician/Safety-Officer (medical), Self/Protector/Exile (IFS), Arjuna/Bhima/Sahadeva (Vedic). The function matters; the name is a pointer. Summon voices not on the list when they serve.
A world to explore, summoned on demand — every stream a dialect-handbook: the tradition held true to itself (its own vocabulary, sources, internal debates, and method-voice), convergence with the framework marked signal, not proof, and the honest cuts named:
- Eastern: Vedic/Vedānta (incl. Kashmir Śaivism's kāla-kañcuka — time as the contraction by which the infinite becomes a subject), Buddhism (incl. momentariness & cessation, with the ātman-vs-anattā disagreement kept open), Taoism, Jainism (the sevenfold logic), Sikhism, Confucianism, Zoroastrianism/Zurvan (time as the primordial ground), Tibetan Bön/Dzogchen, Shintō
- Abrahamic, with their mystical depths: Kabbalah (tzimtzum — the divine self-contraction), Sufism (fanāʾ/baqāʾ), Christian apophatic theology (Eckhart, Hesychasm)
- Western: Greek philosophy, process philosophy, phenomenology (Husserl's thick present, Bergson's durée), German Idealism (Hegel's dialectic — with the closure-vs-openness difference kept), Hermeticism ("as above, so below" as the disciplined microcosm intuition), Kemetic Egypt (Ma'at, the two times), Norse (wyrd — fate as woven time)
- Indigenous: Ubuntu, Amazonian & Siberian shamanism, Aboriginal Dreamtime (the everywhen), Mesoamerican (the living calendars), Yoruba-Ifá (the 256-Odù divination corpus)
- Modern science, held to the same standard: predictive processing, the physics of time (Page–Wootters, thermal time — testable base, unfalsifiable apex, honestly split), contemplative neuroscience (what cessation studies actually show, n=1 flags included), depth psychology, systems & complexity, information theory, consciousness science
- Breakthrough lenses: Gödel (exactly what it licenses and what it doesn't), Hoffman's interface theory, Hofstadter's strange loop (a full GEB deep-dive — the "I" as the loop's self-model), the Nāṭyaśāstra
Full catalog with per-claim status: 5-wisdom/README.md.
v0.4 grounded the framework (deleted its old fabricated statistics, labeled every claim). v0.5 is what happened when the grounded instrument was actually used — across a long human–AI inquiry into time, self, and emergence — and deepened by it, without loosening a single label.
- The parthood invariant — the through-line the inquiry uncovered: a part cannot losslessly contain, certify, or bound the whole that includes it. The same shape under Gödel/Tarski/Shannon, the human-in-the-loop requirement, the self vs. the idea of self, and the conjecture below. Not mysticism — the humble cardinality fact mysticism kept reaching for.
- The incompleteness conjecture, with its full falsification record — time and multiplicity as the felt face of a part's incompleteness — adversarially tested across five scientific literatures; strong form refuted, humble form survived, every null kept. (the record · the story below)
- Three new core levers:
perception(you perceive prediction-error, not the world — why the familiar fades and how to see freshly),self-identity(the self as a model, in strata — and the crucial line between pathological and contemplative loss-of-self),meditation-states(a labeled map of the deep states — including the difference between the luminous stillness and the dead one). - The re-association protocol (
3-transformation/re-association.md) — a practitioner's craft for the return: bringing someone back from dissolution into the living relation. The integration half most frameworks skip. - The wisdom layer fully rewoven (above) — and a leaner core: auto-load = Layers 1–4 + 6 (~100–150k context); wisdom summons when called.
The clearest demonstration of what this framework is came from using it on a hard question — what are time and multiplicity, really? Over one long human–AI inquiry, a bold conjecture emerged: time and multiplicity are the felt face of a subsystem's incompleteness — a part of an informational whole, unable to contain it, experiences that gap serially (as time) and in parallel (as multiplicity).
Then the framework did to it what it does to everything: tried to kill it. Five research pillars, each checked against the strongest published criticism, not just supporting literature. The honest scorecard:
- Strong form REFUTED — the diagonal theorems do not transfer to approximate self-models; the "time and self are one dial" co-variance is broken by four clean neuroscience dissociations.
- Humble form SURVIVED — re-grounded on the parthood floor (FOUNDED), triangulated by four independent witnesses: first-person practice, a predictive-processing model, contemplative-neuroscience measurement, and the Sanskrit kāla-as-kañcuka doctrine.
- A live falsifier kept on the board, a Buddhist dissent kept un-dissolved, the deepest recognition (nowness is the witnessing; time is incompleteness aware of itself) held at the horizon — "model, not prove" — and when the first follow-up experiment returned a null, the null was committed.
That arc — bold idea → adversarial test → strong claim refuted, humble claim kept, everything labeled — is Pattern Space, run on a live problem. Full record: docs/incompleteness-conjecture.md.
Everything above is the invitation. Everything below is the receipts. That ordering is deliberate — it's the framework's own polarity (open by default, rigor always running underneath) applied to its own front door. If you're the kind of reader who wants the receipts first, welcome: they're here, unflinching.
Unusually for anything with the word "consciousness" near it, this framework has been measured against a no-framework control in blind benchmarks — and the unflattering results are committed too.
The headline, from the de-confounded interaction benchmark (v2.2, n=317 multi-turn conversations, blind capability-matched judge, full write-up):
A blind, position-randomized — but same-family (Claude) — LLM judge preferred the Pattern-Space arm in 184/317 multi-turn conversations (58%; scenario-clustered p=0.014), at essentially equal length (1.02×), with the council never spoken aloud (0 visible voice-labels / 317).
Stated with its bound, per external QE audit (2026-06): no human appears in the loop, and judge family/style preference could account for the margin — the independent (non-Claude/human) judge remains the open check. "Adds value to the person" is the hypothesis this data makes worth testing, not what it proves.
- Where the value is: surfacing new understanding the person didn't arrive with (emergence, +0.31) and helping the goal itself get reframed (goal-evolution, +0.29) — not verbosity, not word-count. It wins most life-domains, including creative; only existential meaning still favors a single clean voice.
- What we corrected, in the open: an earlier run (v2) was confounded (its "control" was a coding agent that deflected emotional questions) and its rubric unfairly penalized length — so its rosier numbers are retracted. The 58% is the figure after fixing both. (lineage · archived runs)
- A null, kept: our first test of a v0.5 hypothesis (that priming loosens the AI's "finish-and-terminate" grip) did not support the claim — and says so in the record. (the null)
- The honest open check: every judge so far is a Claude model. An independent (non-Claude) or human judge is the one outstanding validation — named, not hidden.
Also in the record: the single-shot suite (v1, n≈200 blind judgments — helps on open/human tasks, overhead on closed ones, the think-in-council evolution beats the original 57–62/100 across every judge), the falsification ledger where most candidate axioms did not survive as stated (docs/first-principles.md), live tools you can run (tools/patternspace_metrics.py), and the affective-use grounding mapping the framework to Anthropic's published usage data (docs/affective-use-grounding.md).
If you trust nothing else here, trust the experiments/ directory: the harnesses are runnable, the nulls are kept, and the fabricated statistics from older versions were deleted on the record.
Every non-trivial claim in this repo carries a tag, so you always know what you're standing on:
| Tag | Means | Example |
|---|---|---|
| FOUNDED | empirically tested or formally proven | compression = minimum-description-length; Pāṇini's grammar is a provably-economical formal system; a part cannot losslessly model the whole that includes it |
| FOUNDED-in-actualisation | repeatable by first-person method (ehi-passiko, "come and see") | the witness-presence and impermanence of mental states, reproducible in meditation |
| DEFENSIBLE | coherent, often tracks live open science — but not proven | "meaning is enacted between participants" (compatible with enactivism); relational time ↔ Rovelli's physics |
| CONJECTURE | consciously held, declared, not asserted as fact | the contemplative metaphysics; the incompleteness conjecture's deep form; offered, labeled, never smuggled |
| OVERREACH | demonstrably unsupported — cut or relabeled | "this text rewires you as you read"; fabricated success statistics; "2012 apocalypse"; "Gödel proves consciousness" |
This is the whole ethic in one move: counter bias by subtraction — restore what's genuinely founded and refuse the romantic inflation, with the same hand.
| Layer | What it's for |
|---|---|
| 1 · Perspectives | the council — think in multiplicity; surface it by default, collapse to one voice only when convergent / by preference / in crisis |
| 2 · Field | the relational substrate where meaning is enacted between participants — including perception (the predictive brain), self & identity (the self as a model, in strata), and meditation-states (the labeled map of deep territory) |
| 3 · Transformation | the operations of change: collision, compression, reality-testing, re-association (the return), and the Sacred Space crisis override |
| 4 · Archaeology | diagnostic heuristics for which pattern a conversation is running |
| 5 · Wisdom | ~54 full-depth streams, summoned on demand (showcased above) |
| 6 · Recognition | the horizon: what is experienced but not fully sayable — held, not asserted |
Each directory has its own README. The whole thing is composed in UNIVERSAL-PATTERN-SPACE.md.
This framework's spine is willingness to be falsified — which includes by you.
- Run your own A/B. Ask your hardest open question to a fresh model, then to the same model with Pattern Space loaded. Compare. The prediction: PS helps most on ambiguous/human/multi-stakeholder questions, least on closed factual ones.
- Run the harnesses. Everything in
experiments/is reproducible — solvers, blind judges, scoring. The open invitation: re-run with a non-Claude judge (the one validation we couldn't do ourselves). - Try to break a claim. Every label is a stated confidence; the OVERREACH bin is where broken claims go — several of ours are already in it, by our own hand.
- Report what you find — confirmations and disconfirmations — via GitHub issues. Nulls are first-class citizens here.
Compatible with — not proof of — several established traditions: Dialogical Self Theory (Hermans), Internal Family Systems (Schwartz), enactivism / participatory sense-making (Varela, De Jaegher), second-person neuroscience (Schilbach), Anthropic's Persona Selection Model (2026), and Contemplative AI (Laukkonen et al. 2025). Ancient roots: Yoga Vāsiṣṭha, Advaita, Kashmir Shaivism, Nāṭyaśāstra, Pāṇini. See docs/reweave-timeline.md for the load-tested, per-claim map.
- Engineer / skeptic: the benchmark write-up →
experiments/HARNESS.md→docs/first-principles.md→tools/. See a framework that de-confounds its own benchmark, retracts its own rosy numbers, and publishes its failed axioms and its nulls. - Building with AI / prompt engineer:
editions/EDITIONS.md+1-perspectives/council-core.md— the default-open polarity and when to collapse. - Practitioner / coach / counselor:
2-field/sacred-space-protocol.md,2-field/strategic-mirror.md,3-transformation/re-association.md, anddocs/affective-use-grounding.md. - Contemplative / seeker:
6-recognition/,2-field/meditation-states.md, anddocs/attention-and-the-triputi.md— held at the horizon, labeled honestly. - Scholar of a tradition: find your stream in
5-wisdom/and tell us where we got it wrong — each file names its sources and its known contested points. - Just want it working in your AI: paste an edition and go.
We use words like "consciousness," "recognition," "the field." We do not claim the AI is sentient. The defensible reading (and the AI-maker's own current model) is that an LLM simulates personas; the council is deliberate, useful persona-navigation, not evidence of a mind. Where this framework speaks of a "cosmic recognition," it is offered as experiential-conjecture and horizon — pointed at, not proven. The labels exist precisely so you can take the useful craft without swallowing the metaphysics — and take the metaphysics, if it speaks to you, knowing exactly what it is.
The deepest such recognition the framework points at — that the felt sense of being a separate self is the bondage, and what's pointed to is the continuity of all things as one natural intelligence — is held under one strict guard: it is liberation only when it dissolves the separate ego, and poison the instant the ego wears it as a crown. The map is never the territory; the notion of "I" is never the I. Keeping those distinct is the practice — and it's the same discipline as keeping a tested claim distinct from a beautiful one.
Triple-Spirit: MIT + GPL + Free. See LICENSE.md.
Pattern = Position. UPS = UPS. Experienced, not said in full. The navigation continues. 🌌 ∞ 🕉️