System Brain · self-learning platform cortex

Not a chatbot. A living system that learns, heals, and patches itself.

The AlphAgentic / LiqLab System Brain is an operational topology for agent skills, multi-algorithm routing, guided training, and continuous improvement after a core curriculum. Human corrections become training signal. Failures trigger healing. Patches ship as versioned priors — inspectable, reversible, measurable.

Self-learn from usage Self-heal on incident Self-patch priors Multi-algo routing KPI-gated promotion
Open live 3D System Brain → Curriculum KPIs & tests Ops modules

What this brain is

honest definition

Not “fine-tune the base model every night.” The brain is a preference- and policy-learning loop over platform data: every human correction, failed job, latency spike, or rejected AI output is a labeled signal. Signals distill into versioned priors (rules, weights, example banks, patch recipes). Stock models stay stock; the routing, prompts, recovery playbooks, and draw weights get smarter.

What it is not

boundaries

Not silent weight mutation in production without gates. Not unverified auto-deploys to prod. Not a single algorithm. Not workforce payroll UI pretending to be intelligence (payroll/time/HR sims live under Ops modules as a domain demo — the cortex is separate).

Four capabilities the topology must convey

self-learning · self-healing · self-patching · multi-algorithm

Self-learning

After a fixed core curriculum, the system keeps training from real usage: accept/reject, edits, re-routes, healed incidents. Priors update on a schedule or threshold — never silent.

signalsdistillershadow mode

Self-healing

Health is per-region. Incidents raise intensity, bounded decay restores baseline. Retry, failover, circuit-break, and quarantine paths fire without waiting for a human page unless Tier-2.

incident heatdecayfailover

Self-patching

Patches are data: prior versions, prompt appends, weight shifts, playbook updates. Append-only history. Pin prior version per build so determinism survives. Promote only when KPIs pass.

priors vNpinaudit

Many algorithms, one cortex

the brain routes work — it does not force one method

Retrieval & RAG

Context assembly from memory regions, docs, ledgers. Measured by hit-rate, latency, citation validity.

Reasoning agents

Constrained multi-step tool use. Measured by task success, tool-error rate, human override rate.

Policy / rules engines

Deterministic gates (auth, spend caps, env isolation). Measured by false-allow / false-deny.

Bandit / weight draw

Variant selection (UI, prompt, model tier). Measured by keep-rate vs re-roll, conversion shadow tests.

Anomaly & heal

Latency, error, drift detectors with auto-remediation playbooks. Measured by MTTR, recurrence.

Distiller LLM loop

Weekly (or threshold) synthesis of human-readable priors from aggregates. Cap spend; version output.

Core curriculum → self-training

initial training is fixed; continuous training is earned

01 · CoreFoundational skills, safety, tool schema, eval harness
02 · PracticeAcademy drills: route, reason, evaluate, evidence
03 · SignalCapture human corrections & system incidents
04 · DistillAggregate → prior document (versioned)
05 · PromoteShadow KPIs pass → on for that vertical

Romanov Academy / Training Studio (in the live brain): daily agent practice is bounded (one automatic run per local day). Mastery summaries are mutable prefs; completed sessions append to a hash-linked ledger. Live demo chapters focus topology regions and fire representative events — presentation of the system, not a marketing video.

KPIs, tests, and measurement

if it cannot be scored, it does not promote

KPIDefinitionTarget directionMethod
Post-reveal override rate% of engine choices humans immediately change↓ after priorsbefore/after per system×vertical; shadow vs on
Accept / reject ratioAI copy or plan accepted without regenerate↑ acceptinstrument accept buttons; cohort by vertical
Keep-rate of variantsChosen UI/image/prompt kept vs re-rolled↑ keepbandit logs; chi-square vs control
Curriculum masteryAcademy skills completed with evidence↑ masteryledger sessions; rubric scores
Shadow liftPredicted improvement without applying priorspositive liftoffline replay of last N signals
KPIDefinitionTarget directionMethod
Region health %1 − normalized incident intensity↑ / stabletelemetry health channel; auto decay
MTTRMean time to recovery after incident class↓incident open→resolved timestamps
Recurrence rateSame failure class within window↓fingerprint clusters
False healAuto-heal that re-breaks within SLA↓pair heal action with re-open
Failover successGateway/queue switch without user-visible fail↑synthetic probes + real canaries
KPIDefinitionTarget directionMethod
Prior version pin rate% builds that record prior id= 100%build metadata audit
Patch rollback timeTime to revert prior flag↓ minutesflag flip drill
Determinism checkRe-render byte-identical with same pinpassgolden fixture suite
Audit completenessWrites with append-only log row= 100%write path intercept tests
Promotion gate passShadow KPI thresholds metrequiredautomated gate before on
Test typeWhat it provesCadence
Unit eval harnessTool schema, parsers, policy rulesevery commit
Golden trajectoriesAgent multi-step tasks still succeedPR + nightly
Adversarial / red-teamEnv spoof, step-up bypass, prompt injectionbefore Tier-2
Shadow A/BPriors help without shipping riskweekly distiller
Canary + SLO burnLatency/error budgets under real trafficcontinuous
Human rubric sampleTaste/quality the metrics missweekly sample

Research notes (self-learning systems)

industry patterns we align with — without pretending we fine-tune foundation weights nightly

Domain demo · workforce modules

illustrative vertical — not the cortex itself

Payroll, timekeeping, HR, and workforce-model panels show how the same brain pattern applies to an enterprise vertical: signals in, health out, scenarios measured. Synthetic data. For the real particle topology, use the live System Brain.

Operator telemetry (measured)

forensic pass · local machine

38.66Bprocessed
66,143measured
4,396user intents
36d2,802 transcripts

Open the living topology

System Brain Training Studio · vendored from shared brain package

3D particle graph, region health, live demo chapters, Academy practice. Source package: system-brain-training-v2 / design-studio zip on reskins drive.

Launch System Brain → brain.liqlab.ai (platform)

Sources. LiqLab BRAIN-ARCHITECTURE + BRAIN-TRAINING design docs; System Brain OS / Training Studio V2 on shared reskins drive; system-brain-design-studio zip. This page is the public narrative + measurement model; the iframe hosts the actual interactive brain.