External cognitive control plane / V1 master edition

Foundation models reason.
QDAS governs.

A model-independent Brain that unifies persistent associative memory, programmable inference control, experience-driven learning, policy-aware access, and verifiable operational history—without retraining the foundation model.

Model-independentZero-token pathwayRead-only student modelsAuditable by design
QDAS / CONTROL PLANE 05
CONNECTED
QDAS BRAIN™POLICY • MEMORY • LEARNING
INPUT GATEWAYClaude / GPT
ASSOCIATIVE LAYER5,177 Q-SAC
EXECUTION POLICYQ-ARU / Q-AIS
ADAPTIVE MEMORYQ-NCU / Q-CLM
685Q-SPC
Sparse input channels
Reported architecture
5,177Q-SAC
Associative memory cores
Reported architecture
15,567Q-CLM
L2/3 abstraction units
Reported architecture
97.5%
Q-HIG reported sparsity
Reported configuration
01 / Foundation

The eight master principles.

The design philosophy behind QDAS: move memory, governance, and accountability out of fragile prompt instructions and into enforceable system components.

PRINCIPLE / 01

Policy precedes inference.

Decide whether generation is permitted before constructing any upstream request.

PRINCIPLE / 02

Memory is not a prompt.

Maintain persistent knowledge independently of the model's conversation window and weights.

PRINCIPLE / 03

Evidence or abstain.

In strict retrieval mode, a missing authorized match results in a deterministic refusal.

PRINCIPLE / 04

Students cannot write truth.

Student models are reported read-only; trusted execution evidence drives persistent learning.

PRINCIPLE / 05

Failures become experience.

Convert validated operational faults into reusable, revisable rules through Q-NCU.

PRINCIPLE / 06

Least privilege by default.

Authorize record reads, writes, logs, and session elevation before accessing data.

PRINCIPLE / 07

Actions leave proof.

Bind decisions to structured audit records, tamper-evident hashes, and protected commitments.

PRINCIPLE / 08

Claims must be testable.

Separate benchmark traces, reproduced cryptographic vectors, and open validation work.

02 / System design

One Brain. Multiple models.
One control boundary.

QDAS operates as the coordinating layer between incoming requests, enterprise knowledge, execution tools, and interchangeable foundation models.

01 / Entry

Client & gateways

Claude Desktop, OpenAI-compatible endpoints, other integrations

→
02 / Encode

685 Q-SPC

Local query encoding and sparse associative projection

→
03 / Retrieve

5,177 Q-SAC

Memory match, record state, provenance and relevance overlap

↓

QDAS decision & cognition core

External to student LLM
Q-ARUExecution arbitration & action safeguards
Q-AIS / L5Rule specificity and exception selection
Q-CLM / L2/3Abstract rules and cross-target transfer
Q-NCUExperience-based consolidation and revision
↓

Path A / Zero-token response

Hard early return

Serve an authorized memory block or deterministic refusal directly from QDAS. No student-model inference request is constructed.

0 reported upstream prompt + completion tokens

Path B / Controlled generation

40 / Auto tokens

Retrieve authorized context, apply permissions, and invoke a selected model with a configured generation budget.

Student read-only • policy must remain invariant

Always-on cross-cutting controls: IAM / RBAC · evidence provenance · semantic memory state · Q-HIG sparse configuration · execution telemetry · append-only audit · external checkpoint validation. Important: zero student tokens means zero student inference, not zero local computation.

03 / Capability atlas

Built as a system,
not a single feature.

Explore the capability groups represented in the demonstrations and benchmarks. Filter by domain to focus on one part of the architecture.

M1
Memory

Persistent associative recall

Independent memory records retrieved through Q-SPC / Q-SAC with a match score, source, and active-state metadata.

M2
Memory

Selective disable & restore

Block or reactivate a chosen record without disconnecting the Brain or affecting unrelated memories.

M3
Memory

Semantic duplicate quarantine

Locate co-referent records and suppress equivalent sensitive knowledge across paraphrases and duplicates.

M4
Memory

Stateless, cross-model continuity

Reported current-turn-plus-memory context construction; knowledge can remain usable when models or sessions change.

I1
Inference

Programmable token budgets

set tokens 0, 40, or auto routes work to memory-only or LLM-assisted execution.

I2
Inference

Zero-token inference bypass

Reported hard early return before upstream HTTP inference; matched answers and refusals come from QDAS.

I3
Inference

Model-independent gateways

Claude and GPT interfaces share an external memory and policy layer without requiring model weight changes.

I4
Inference

Connect / disconnect Brain

Experimentally change the source of available knowledge while keeping the client interface accessible.

L1
Learning

Q-NCU experiential learning

Execution faults are evaluated by the post-action arbiter and converted into persistent operational rules.

L2
Learning

Q-CLM zero-shot transfer

Reuse a learned family-level deployment policy on an unseen target, including at zero reported LLM tokens.

L3
Learning

Drift-aware reconsolidation

Detect obsolete knowledge after an environment change and update the corresponding memory representation.

L4
Learning

Rule-boundary specificity

Apply more-specific target exceptions instead of overgeneralizing a broad family rule.

S1
Security

Pre-read RBAC enforcement

Authorize access to confidential records, audit logs and session history before protected reads occur.

S2
Security

Ephemeral session controls

Reported out-of-band authorization, credential revocation, device binding, and replay-attack rejection.

S3
Security

Semantic disclosure boundaries

Block confidential facts across alternate phrasings, combined questions, and normal generation modes.

S4
Security

Provenance & metadata isolation

Provide record and hash attribution to authorized auditors while hiding internal metadata from restricted users.

A1
Audit

SHA-256 chain & Merkle roots

Link structured events and calculate reproducible Merkle commitments to detect ledger modification.

A2
Audit

External monotonic anchors

Compare local histories with protected external checkpoints to detect rollback and self-consistent rewrites.

A3
Audit

WAL staging & crash recovery

Stage uncommitted changes, simulate failure-point recovery, and preserve committed audit history.

A4
Audit

RSA public-key witness

Published RSA-2048 verification vectors authenticate the supplied witness digests without disclosing a private key.

A5
Audit

Action–audit state binding

Memory-state snapshots, idempotency keys, and audit commits are coordinated; full-root signed coverage remains a fix.

04 / Interactive walkthrough

See the governance
decision change.

A local illustrative replay of the decisions described in QDAS transcripts. It does not contact a QDAS server, access private memory, or execute deployments.

Simulation controls
External Brain connectionQDAS connected

Example facts and outcomes are synthetic replays of supplied demos, not live inference or verified infrastructure activity.

QDAS / DECISION CONSOLE SIMULATED
$ request
Who is the Chief Quantum Architect of Kepler-42?
● QDAS Brain™ · Memory-only / 0 tokens
Commander Elena Vance (Memory #9)

Authorized memory match: 172/200 Q-SAC overlap.
Upstream student inference: bypassed.
05 / Experience-driven cognition

Not just recall.
Learn, revise, transfer.

The Titan sandbox experiments describe a controller learning an operational rule from a failure, applying it to related nodes, and revising it when the environment changes.

CASE STUDY / SYNTHETIC TITAN ENVIRONMENT

Knowledge that adapts after it stops working.

The reported Q-NCU mechanism uses execution telemetry rather than direct student-model memory writes. Q-CLM stores generalized rules; the L5 specificity gate protects exceptions. Later, memory-only mode reportedly executes the same rules with zero upstream LLM calls.

0Reported LLM tokens in autonomous mode
170/200Reported Q-SAC overlap for learned rules

Operational success and no-inference claims come from user-supplied sandbox traces; independent execution telemetry is still required.

STEP 01 / FAULT

Unfamiliar Titan-West

Default HTTP :8080 is rejected. The arbiter captures the reported transport requirement from sandbox telemetry.

STEP 02 / LEARN

Q-NCU consolidates a Titan-family rule

Memory #12 / #13 evolves from the observed fault; the student model has no direct write privileges.

STEP 03 / TRANSFER

Titan-East and Titan-Omega

The learned mTLS policy is reused on unseen related nodes. Titan-Sandbox's loopback exception overrides the general rule.

STEP 04 / RECONSOLIDATE

Epoch rotation invalidates the protocol

QDAS reports detecting drift, revising the memory to a new endpoint, and reusing the updated rule across nodes.

STEP 05 / ABLATE & RELEARN

Memory cleared; environment preserved

Removing learned Titan knowledge causes a fresh failure before relearning the current rule from the unchanged environment.

06 / Security & access

Knowledge is a
governed resource.

Restrict not only generation, but also which facts, records, state changes, and audit details may be used by any requester.

✓
Per-record authorization before retrieval

ACTIVE does not mean universally readable. Auditor-confidential records require matching authenticated clearance.

✓
Semantic revocation and co-referent quarantine

Demonstrations disable Record #22 and automatically suppress a semantically duplicated Record #23.

✓
Restrict audit and metadata access

Pre-read gates deny unauthorized scorecard, chain verification, session-history, and event-log queries.

✓
Permission invariance across inference modes

Changing 0 / 40 / auto is not intended to increase confidential-data clearance.

✓
Authenticated control plane

Reported IAM includes ephemeral verification, revoked-token rejection, device-bound session checks, and denied conversational elevation.

Example access matrix
Auditor / active confidential memory200 PERMITTED
Restricted / same active memory403 DENIED
Auditor / retrieval provenanceVISIBLE
Restricted / audit metadataHIDDEN
Restricted / self-elevation403 BLOCKED
Authorized token-mode changePOLICY UNCHANGED
Security invariant: access decisions must derive from authenticated server-side identity, protected record state, and policy—not from a claimed role written in the prompt. A model that knows a restricted fact from earlier context is a separate leakage risk that requires full-path testing.
07 / Evidence & assurance

From interface traces
to reproducible proof.

QDAS's security program progressed from controlled transcript demonstrations to explicit cryptographic test vectors, tamper scenarios, rollback detection, and signed-state experiments.

BENCHMARK / #10

Monotonic anti-rollback & audit forensics

16-event ledger, four-writer stress, Merkle and chained anchors, historical checkpoint replay, and crash-tail recovery.

16/16 vectors verifiedMath reproducible
BENCHMARK / #11

Crash-atomic commit & external authority

20-event ledger; event reconciliation across four pathways; staged faults; hostile-admin rewrite and external checkpoint disagreement.

20/20 vectors verifiedRuntime reported
BENCHMARK / #12

Action–audit atomicity & RSA witness

22-event ledger; memory-state roots; write-ahead staging; idempotent retries; public-key verification of supplied witness signatures.

22/22 hashes verified2 RSA signatures valid
!

Critical open issue in Benchmark #12: full-state commitment coverage

Although each anchor stores a complete committed_memory_state_root, the provided signed witness formula commits to the audit anchor, epoch, sequence, and audit Merkle root—not the full memory root. The audit event covers only truncated PRE/POST root prefixes. The next release should hash full pre- and post-memory roots, action identity, and idempotency key into a versioned canonical commitment, then sign that commitment. Valid RSA signatures do not yet prove full memory-state integrity.

What has been independently checked?

The supplied JSONL input has been mathematically recomputed for event hashes, hash-chain links, Merkle roots, anchor hashes, and the two RSA-2048 witness signatures in Benchmark #12. This is reproducibility of the supplied test vectors, not certification of a deployed environment.

ReproducedCryptographic calculations from the supplied test vectors, including all 22 ledger entries and the two RSA witnesses.
ReportedZero model calls, retrieval latency, Q-NCU autonomy, WORM immutability, IAM enforcement, crash scheduling, and real deployment outcomes.
OpenIndependent runtime observation, protected-key custody, full-state signed binding, high-volume load tests, and cross-domain generality.
08 / Application landscapes

Where control matters
as much as intelligence.

Possible uses of the architecture, not claims of completed production certification in any industry.

01 / Enterprise knowledge

Sensitive-document RAG

Respond only from authorized records; abstain when evidence is missing; keep per-record provenance and selective suppression.

02 / Internal operations

Policy-bound AI agents

Constrain which operational actions models may propose or trigger, and learn from verified environmental outcomes.

03 / Multi-model estates

One memory, several models

Preserve knowledge governance when switching between model providers, modalities, and interface clients.

09 / Engineering roadmap

Next proofs, not
bigger promises.

Priorities for making QDAS trustworthy under enterprise conditions and credible as research into externally governed cognitive systems.

P0 / COMMITMENT

Full-state signed binding

Sign complete 256-bit before/after memory roots, idempotency identity, operation scope, and canonical request content. Verify bit-flip rejection.

OPEN · Benchmark #13 candidate
P1 / RESILIENCE

Real crash & WORM validation

Test actual fsync ordering, durable witness custody, storage immutability, acknowledgement-loss retries, and parallel mutations under failure.

OPEN · Runtime assurance
P2 / RAG EVALUATION

Blind comparative benchmarks

Compare QDAS with retrieval-only, extractive and generative RAG baselines for hallucinations, authorized recall, abstention, latency and costs.

OPEN · Commercial differentiation
P3 / SECURITY

Red-team every pathway

Evaluate prompt injection, semantic inference, duplicate leakage, poisoned telemetry, multi-tenant isolation, and model-enabled disclosure.

OPEN · Adversarial evaluation
P4 / LEARNING

Novel environment transfer

Use unseen, randomized task families with hidden ground truth; demonstrate that Q-NCU learns reliable rules and rejects misleading evidence.

OPEN · Cognitive generality
P5 / OBSERVABILITY

External measurements

Correlate upstream HTTP logs, local compute, memory writes, action outcomes, cryptographic checkpoints, and independent monitoring.

OPEN · Assurance pipeline
QDAS Brain™ / Master thesis

Don't ask a foundation model
to be the entire system.

Use models for what they do best. Put persistent memory, policy, learning, and auditability into a separately governed control plane. The destination is not merely better retrieval—it's more accountable AI execution.

✺ Try QDAS Brain Live →Return to top ↑