Graph Source
LIVE · CURRENT SESSIONMachine State
Process State
Prediction Residue
I/O Signal
Recent Observation Frames
| Sequence | CPU | GPU | RAM | Temp | Stress | Residue | QoS |
|---|
Target process --%
Used -- GB
Temperature --°C
Latency 0 ms
— Pulse modulation
··· Residual feedback
– – Agent directive
ACQUIRING
Confidence 0.00
Active Dataset
0 B persisted as JSONL observation frames.
Persistence Contract
Every interval stores machine and workload state, selected action, unresolved residue, measured outcome, and rolling metrics.
Saved Sessions
| Session | Samples | Bytes | Last sample | Target |
|---|
After each recording: Learn All → Graph Blobs summarizes inactive sessions, calibrates Futurist skill, stores a compact graph blob, and deletes heavy raw JSONL. The active session is protected.
No session compacted in this run.
Latest Frame
Waiting for first observation frame…
Iteration Memory
Cortex Boundary
Pressure Transduction
Net Vector Pressure
Custom Iteration Graph
Select an iteration and any two normalized signals.
Learned Iterations
| Iteration | Version | Samples | Raw reclaimed | Transduction | Slack |
|---|
Discoveries
No compact learning iteration loaded.
Know Thy System
Envelope
Pressure Risk
Skill vs Persist
Stress H=5
RAM H=5
Governor Ability
Memory Guards
Futurist Channels
Waiting for multi-horizon forecast…
Waiting for telemetry.
Concurrency
Batch Size
Token Budget Scale
Retrieval Depth Scale
Development Stage Gate
Bounded gain writes remain disabled unless explicitly enabled in configuration. Agent-policy mode exposes recommendations; the external agent chooses whether to obey them.
Adaptive Tuner
Waiting for finalized evidence.
Tuning Evidence
Finalized samples · confidence 0.00 · applied NO
Machine-Readable Directive
{}Machine-Readable Tuning Suggestion
{}Average Throughput
Flow Stability
Thermal Oscillation
Prediction RMSE
GPU Energy
Energy per Token
Oscillation Coherence
Futurist Governor
Aggregate Contraction
Trigger Density
Homeostatic Slack
Pressure Momentum
Recovery Balance
Resource Coupling
Baseline / Candidate Comparison
--
--
INCONCLUSIVE
Select two saved sessions. Use the same workload, model, ambient conditions, and duration.
Metric Boundary
Intervention value requires a comparable baseline and controlled run. PulseFlow deliberately reports it as unavailable until a baseline/candidate pairing exists; it does not manufacture a performance claim from correlation alone.
Replay Laboratory
Replay recalculates controller traces against recorded observations. It is a tuning screen, not a causal substitute for a live A/B benchmark.
Timeline
Replay Frame
No replay loaded.
Candidate Report
No candidate replay run.
Runtime and Persistent Event Ledger
| Time | Kind | Message |
|---|
Bounded Live Tuning
The control sliders in the right rail update a bounded runtime configuration. Values are clamped by the backend. Change one parameter at a time and use replay plus controlled live trials before claiming an improvement.
Safety Boundary
PulseFlow does not alter clocks, voltage, fan curves, firmware, GPU power limits, or system power plans. Active control is restricted to a process explicitly launched or attached by the operator.
ARIA Verification
Run scripts\ARIA-Verify.ps1. It compiles Rust, runs unit/integration tests, checks the UI action contract, parses browser JavaScript, validates schemas, and confirms that each declared button maps to a handler and backend route.
Operator Quick Start
Start: enables frame recording and, when a target process exists, bounded process modulation. Pause: stops modulation while telemetry remains live. Reset: opens a new recorded session.
Quiet / Balanced / Performance: select the controller setpoint. Begin with Balanced. Change one slider at a time, press Apply Parameters, and compare saved sessions in Analytics.
Recommended learning order: Recorder, Analytics, Replay, Shadow, Bounded Adaptive, Agent Policy. Do not promote a later stage until the earlier evidence gates pass.
System Interlink Verification
Current Configuration
Loading…