One entry point for both domains. Select an observation adapter, load its source, and stream the Human and Agent measurements through the same CCSM and L/R/G architecture.
Aether observation adapter
Conversation observation adapter
Select an episode from the JSON corpus. Embedded trajectory cues are masked before measurement; ground-truth annotations remain evaluation-only. The timeline uses message turns, not invented seconds.
Aether mode · choose a trajectory CSV, then stream.
Separate online HDMIC engines for User and Agent. Both see the same event clock and maintain independent Local, Regional, Global attractors, adaptive parameters, error histories, and plots. View the full machine for either participant, or compare both.
Choose CSV or demonstration.
30
User · Interaction Dynamics
—
Awaiting input
Agent · Interaction Dynamics
—
Awaiting input
Composite score is a causal heuristic built from changes in SMC, coupling, Shared EB, drift, velocity, and turning when available. Its components include exploration, continued fluid action, reorganization and clamping. Missing measurements are omitted and disclosed; this is not an independently validated consciousness or cognition metric. Select “Raw” for exact original-data comparisons.
Aether CSV note: 50 ms samples include inactive intervals. Flat early signals and zero coupling can be genuine data; regulatory scores are withheld during measured inactivity.
Independent observation clock · processing status
Awaiting recording · Human engine pending · Agent engine pending
Optional: add handoff and intervention markers
Human and Agent participation / engine progressSource events marked on timeline when available
The horizontal axis is source recording time, not wall-clock playback duration. Sampling speed and optional L/R/G pacing can make elapsed wall time different. Native engines may lag; their results are retained in chronological order.
Compare the complete conversational measurement histories of two conditions without concatenating independent episodes. Phase 1: descriptive, observation-only batch analysis.
Mean episode trajectories; normalized progress smooths event structure and is not elapsed time. Differences are easier to inspect in the contrast plot below and in actual per-turn episode traces in the exported report.
B − A · Difference in mean score (centerline 50 = equal; visually magnified ×2.5)
Regulatory composition · Traditional SMC
Episode-level occupancy of inferred environmental unclamp (−1), stable flow (0), and mental/interactional unclamp (+1). Pause inference is not verified reflection.
Interaction and collaboration dynamics · Layer A / Layer B
Layer A: sustained interaction regimes inferred from masked conversational behavior. Layer B: candidate onset events, counted once per onset. These are research heuristics, not independently annotated ground truth.
Layer A · Sustained regimes (mean share of turns)
Layer B · Interaction event onsets (per 100 turns)
The three plots display different kinds of evidence. Event rates use the count per eligible turns; no native HDMIC Layer A/B classifier or independent ground-truth validation is implied.
Events and measurement coverage
Run comparison to populate.
Conversation-level inspection
Run analysis before inspecting an episode.
Condition
Conversation
Turns
Score
Coupling
Coordination
Reorganization
Event F1*
Status
Awaiting A/B analysis
Scope: Independent episodes are analyzed with the same masked conversation adapter. No native L/R/G learning, frozen landscape transfer, or real-time forecasting is claimed for fast batch mode. Composite score and Traditional SMC are derived descriptors, not external validation. Synthetic event F1, where reference annotations exist, compares heuristic event detections with synthetic labels. Paired comparisons require defensible matching; unassigned corpora do not establish causality.
Experimental protocol · v4.52
Experiment Section
Four-block timed observation, condition tracking, measurement summaries, and exportable reports.
NO RUN YET
Experimental Run
No experimental runs recorded in this session. Configure a protocol and press Start. Observation capture follows the PEM measurement stream.
Current PEM is a recording-analysis instrument. Experiment conditions are not automatically imposed on the underlying AI/drawing system. Until actuators are connected, comparisons are descriptive, not randomized causal effects.
0 / 4 blocks completed
Condition
Samples
Human active
Agent active
Interaction
Coherence
Coupling
Drift
Waiting for experimental run.
Metrics are within-block descriptive means from newly received measurements. Missing values are unavailable, never zero-filled. Session Reset does not erase completed experiment snapshots. No causal inference is implied by condition labels.
Experiment Controls
Current condition—
Time remaining—
Block0 / 4
Total duration2:00
New observations0
Trial cacheReady
Start recording playback before or after beginning the protocol; a block collects new observations only. Pausing the experiment stops its elapsed time, not PEM's underlying input or learning engines. Surveys are optional manual observations.
PEM v4.57 · Live research index
Interaction Score
Exploratory organization index · time-varying, not externally validated quality or co-creation achievement
Click the score chart to examine trends, recurring motifs, and label intervals.
Overall InteractionEmergenceTransitionTrajectory
Scores describe measurable temporal organization, not artistic quality or causal evidence. The formula is fixed during the run and uses causal observations. Native L/R/G identifiers are reported separately and are not imputed into the score when unavailable.
Temporal event observatory · v4.51
Interaction Event Timeline
Overall Interaction Score through time. Markers show descriptive event hypotheses and native attractor events; click a marker or its card for context.
Waiting for observations
Hover over an event marker or card to preview its stats. During replay, the first crossed event pauses playback and opens the full detail window; close it and press Play to continue. The playhead replays already analyzed observations and leaves the learning engines unchanged. Click the score trace to seek, or click a colored event marker for its context.
Events will appear as streaming observations and native L/R/G results arrive.
Composite events are detected from the Overall Score; interaction-dynamics labels are provisional rules; L/R/G transitions are native state-ID evidence, not independently annotated semantic events. Delayed engine results appear when available.
Interaction Event · temporal context
Selected event
Temporal Scope · Measurement interpretation
Metric
Latest computed value
—
Operational definition · calculation
Measurement boundaries
Interpreting this curve
Temporal evidence · recorded history
Select a point on the plot to inspect its value and temporal context.
Temporal Scope · Interaction Score analysis
Interaction Dynamics Explorer
Drag across the chart to select an interval. Candidate boundaries and repeated score patterns are exploratory, not ground-truth event labels.
No interval selected
Candidate events are detected from changes in Overall Interaction. The three component curves supply dependent evidence about their character, not independent proof. Click an event band or card to inspect its history. Events from the Live Interaction Events panel can also open this context directly.
Open the explorer to detect composite event intervals.
Click a detected event to inspect its pre-event baseline, component evidence, recorded activity, and prior related patterns.
Selected interval · trend analysis
Select an interval to analyze its slope, variability, and corroborating measurements.
Triangulation across measurements
No selection.
Shared-input proxies are not independent confirmation. Raw activity/coupling and native L/R/G state changes are reported separately from score components. Optional event-file markers add another source.
Candidate recurrent score motifs
Run analysis to discover candidate patterns.
Label the selected event
No manually labeled intervals yet.
Participatory Sense-Making · temporal instrument
Sense-Making Observatory
Human, Agent and Participatory trajectories. Composite: +1 unclamping, −1 clamping, 0 pausing/waiting/hesitation. This is a provisional regulatory code, not validated understanding or creative quality.
Awaiting observations
Switch Calibrated / Legacy anytime for retrospective comparison. Threshold and strength stay fixed during this run; switch does not retrain HDMIC.
Layer 1b · First difference of raw interaction codes
Change per observation window: positive = code rises (including movement from −0.5 toward 0); negative = code falls. Not equivalent to mental unclamping.
Layer 2 · Regulatory evidence
● Mental● Environmental● Clamping● Inactive
Layer 3 · Cumulative SMC-EMoC
Human · Agent · Participatory mean; cumulative sum of 0.25 s window codes (turn-based for conversations)
Sense-Making Memory · cross-detector comparison
Traditional SMC × Calibrated EMoC
The same observed Aether windows are scored two ways. Traditional SMC is an interpretable three-state hypothesis; calibrated EMoC is a continuous evidence contrast. Differences are diagnostic—not independent evidence that either detector is correct.
Awaiting observations
━━ Traditional SMC (−1, 0, +1)━━ Calibrated EMoC (−1 to +1)● Disagreement (opposite signs or strong contrast)
Patterns are descriptive sequences of (Human, Agent) Traditional SMC codes. Counts are derived from nonoverlapping three-window sequences; recurrence is not evidence of learned memory or predictive usefulness.
Waiting for data
Pause → +1 in the Traditional operational code is an inferred mental/interactional unclamp, not verified reflection. Native L/R/G evidence may arrive after observation; comparisons reflect currently available telemetry.
Waiting for observations. Native model-change evidence can arrive after the observation clock; unavailable sources are not treated as verified changes.
Human · SMC—Awaiting stream
Agent · SMC—Awaiting stream
Participatory · relational view—Descriptive joint trajectory; not a new independent measurement
HumanAgentParticipatory (mean)
Lagged responsiveness · Human → Agent—Correlation of preceding Human and subsequent Agent changes
Lagged responsiveness · Agent → Human—Correlation of preceding Agent and subsequent Human changes
Divergence / convergence—Participant slope difference, not absolute raw SMC distance
Click the chart to select a moment, or use Explore selected interval to open the existing Interaction Scope event and motif analysis.
SMC contribution study · A/B/C evaluation
A: conventional activity, coupling and interaction signals. B: A + participant SMC slope/acceleration. C: B + native L/R/G state-change features. The prototype scores causal next-step *interaction-score change* prediction using online learning; it does not test independently annotated events, transfer, or creative quality. A/B/C are only comparable on the same eligible paired observations.
Not evaluated. No results are inferred before running the study.
Model
Features
MAE
Paired samples
Run study to populate
Cross-domain comparison must use equivalent coding and time-normalization. Participant slopes and reciprocity are exploratory proxies; associations do not establish causality.
v4.36 · Interaction dynamics observatory
Live Interaction Events
Causal onset hypotheses from the unfolding stream. Click an event to inspect the corresponding history in Interaction Scope.
Current interaction
Awaiting observations
Waiting
—
Overall interaction · 0–100
0 candidate events
Waiting for events.
Select a candidate to inspect its observed evidence.
Interaction classifications are provisional heuristics. The composite-event explorer independently groups Overall Interaction changes into intervals; its component curves are dependent evidence, not three independent confirmations.
Live temporal instrument · native L/R/G
L/R/G · Persistent Attractor Observatory
Direct Human and Agent engine outputs. State identifiers are local to each engine; numbers are not shared semantic categories.
Waiting for L/R/G engine observations
No engine samples
L · LOCAL CONFIGURATIONS
—
—retained attractors
Show L/R/G diagnostics
Waiting for observations
R · REGIONAL SEQUENCES
—
—retained attractors
Show L/R/G diagnostics
Waiting for observations
G · GLOBAL TRAJECTORIES
—
—retained attractors
Show L/R/G diagnostics
Waiting for observations
Scale-input audit: The native engine feeds the same normalized instantaneous value to all three machines. L/R/G differ in reference-window duration, thresholds, and update policies; they do not yet ingest separately aggregated scale-specific observation streams. Native counts are read directly from separate machine objects. “Observed max count” is not a cumulative creation counter.
Created = highest observed attractor count; retained = latest engine count; re-entries = returns to a previously seen non-adjacent state (descriptive only). Next-ID hit rate and Hellinger fit are shown as unavailable unless supplied as verifiable native metrics.
Temporal structure · formation and recurrence
Attractor Formation Timeline
Formation = first appearance of a new ID; consolidation = first 5 assigned samples; recurrence = return after exit; abandonment = prolonged absence after exit.
Observation clock—No observations yet
Selected-role emergence—Formations, consolidations, re-entries, and abandonments
Structural emergence timeline
Local active IDRegional active IDGlobal active ID▲ formation■ consolidation○ recurrence✕ abandonment
Recent emergence events
Waiting for processed observations.
Participatory Emergence Machine · Research Instrument
Interaction Measurement Observatory
Can a continuously measured model of participatory interaction reveal meaningful temporal and relational organization that output-centered evaluation leaves unmeasured?
Awaiting chronological observations
Episode provenance: load a conversation to inspect the source of each chart.
Research executive summary · live current-run evidence
Emergence · Transition · Trajectory
Start a recording or review a completed conversation to populate this summary.
1 · Emergence
—
Observed structure and recurrence
2 · Transition
—
Temporal changes and reorganization
3 · Trajectory
—
Dyadic continuity and baseline context
Scores are current-run experimental indices (0–100), not independent ground-truth accuracies. Held-out motif and regime validation requires a separate protocol and cannot be inferred from these signals.
01 · Dyadic coherence & coupling
What is happening in the relationship between the Human and Agent right now?
Awaiting streamed observations
Primary relational signals: interaction coherence, recorded coupling, and coordination, with disruption as a contrastive indicator
Simultaneous exploratory measures, not exclusive or externally validated state labels
ExplorationReorganizationFluid action
03 · Temporal organization
What temporal organization has emerged?
Awaiting streamed observations
Existing Local, Regional, and Global repertoires from the two original learners
Human local state IDAgent local state ID
04 · Evaluation benchmarks
How do measurable outcomes compare with a simple baseline?
Choose a causal baseline by clicking its card. Compare matched errors against the lagged-engine estimate; measure online forecast coverage separately from offline retrospective evaluation. Engine estimates are fitted current-value outputs lagged one step, not dedicated next-step predictions.
Online learning · on-time coverage—Evaluated before next observation
MAE · lagged engine—Waiting for paired predictions
MSE · lagged engine—Waiting for paired predictions
Each baseline predicts sample t using only data through t−1. All evaluation uses matching sample indices; baselines never peek at future observations.
Awaiting measured observations
Conversation mode · optional research toolsConversation observation & reference evaluationExpand for corpus batch analysis, learning curves, and reference annotations.▼
Load the original HDMIC longitudinal JSON. Select a conversation; only message text (with embedded trajectory-cue tags masked) enters the observation adapter. Ground-truth phases and events remain separate for retrospective evaluation. Time is measured in message turns, not seconds.
Corpus evaluation · background prequential learning
Default background mode ingests observations without animating each episode or waiting for iframe engines. L/R/G remains available in single-conversation playback; batch L/R/G is explicitly unavailable unless exhaustive engine mode is selected. L/R/G outputs still processing are marked partial (not full analysis). Check “Wait for full L/R/G analysis” for slower exhaustive per-episode processing. Batch learning: a separate causal online predictor updates after each observed message. Its weights transfer across conversation boundaries when enabled; L/R/G engines still restart per episode. The learner never accesses synthetic labels. Disable for isolated-episode comparison.
Batch visualization:Background computation · charts remain stable during processing. View a completed episode afterward.
No completed conversation to inspect.
Online predictor rolling MAEPersistence rolling MAERolling predictive skill (right scale)
Cumulative time axis: observed messages across all episodes. Vertical markers show conversation boundaries. Loss is computed prequentially (predict before seeing each target, then update). This simple lexical-interaction predictor is separate from the native L/R/G learners.
Load a conversation corpus, then click Run Corpus Batch. Live episode and engine progress will appear here.
Episode
Turns
Engine samples
MAE
Baseline MAE
Mean coupling
Mean reorganization
Event P / R / F1*
Online MAE
Status
* Experimental event detection uses only the masked text adapter's heuristic event proxy and the synthetic reference event annotations; it is not the trained HDMIC event detector. Scores are retrospective, per-turn, exact-match micro metrics. Causal L/R/G forecasts remain separate from ground-truth labels. Batch engine delivery is accelerated/offline, not on-time real-time coverage.
No conversation corpus loaded. Aether CSV mode remains available below.
Awaiting conversation observations.
Ground-truth comparison becomes available as playback progresses. Synthetic annotations are not independent human judgments.
CCSM · domain-independent measurement schema
Co-Creation Score Suite
Four separate measurement families; no unvalidated global score. Displayed traces are experimental operationalizations. Unavailable semantic annotations remain unavailable, not fabricated.
A · Cognitive dynamics
Sense-making & regulation
Human changeAgent changeReorganization
Causal CSM/selected-column derivatives and regulatory proxy. Reflection/projection appear in the EMOC panel below.
B · Interaction dynamics
Coordination & participation
Recorded couplingCoordination proxyActivity balance
Observed participation and coupling versus derived coordination. Negotiation requires event/semantic coding.
C · Domain dynamics
Actions & shared activity
Any participant activeSignal changeAction variation
Domain-independent action and environmental-change proxies. Quality, task progress, and consequences need independent annotations.
D · Collaboration dynamics
Joint development & repair
Joint activityDisruption proxyRepair proxy
Offers, acceptance, elaboration, surprise and verified repair remain unmeasured without event labels or discourse coding.
05 · Enactive feedback-loop configurations
How are the Human and Agent organized within the four feedback relations?
Quadrants: upper-left Reflection / Integration · upper-right Projection / Anticipation · lower-left Action / Consequences · lower-right Affordance Exploration. Participant loop estimates use individual activity and sense-making derivatives only; shared coupling, drift, and embodiment are reserved for dyadic measurements. Similarity is a heuristic comparison, not evidence of shared cognition.
Human / User
Awaiting streamed observations
Loop percentages and emergent state will appear during streaming.
Agent / Aether
Awaiting streamed observations
Loop percentages and emergent state will appear during streaming.
Measurement contribution matrix · established evaluation and proposed extensions
Evaluation dimension
Common output / task-focused measures
Participatory interaction measurement
Evidence boundary
Output quality
Correctness, relevance, error
Contextualize when responses occur in evolving coordination
Output labels not supplied
Task outcomes
Completion, success, satisfaction
Temporal pathways toward outcomes
Independent task ratings needed
Multi-turn behavior
Context consistency and progress
Continuous individual and dyadic trajectories
Heuristic measurements; not independently calibrated
Existing Local / Regional / Global attractors and transitions
Formation and recurrence are algorithmic
Prospective value
Accuracy and task performance
Comparative dynamics forecasting against persistence
Comparable horizons and denominators required
Research objective: establish observability, describe temporal organization, and test incremental explanatory or prospective value. Claims of better human–AI interaction measurement require independent validation; these indicators are experimental.
Engine diagnostics · original User and Agent HDMIC streams, attractors, adaptive regulation, and plots
Participatory Emergence Machine
Single-channel HDMIC prediction of the combined User + Agent Sense-Making Curve · Online Local / Regional / Global emergence
Nicholas Davis, Ph.D.
Participatory Sense-Making · default streaming signal
How is the interaction organizing itself? Which familiar attractor, regime, or trajectory is it approaching or re-entering? How will its combined User + Agent Sense-Making Curve develop next?
Default corpus: synthetic demonstration, 1,200 chronological turn events. One prediction channel = cumulative (User SMC + Agent SMC) / 2. User/Agent curves and raw interaction codes remain inspectable in uploaded/exported CSV. The machine's existing HDMIC L/R/G logic is retained. This is not a measurement of real conversations.
Latest User SMC —Agent SMC —Combined SMC —Combined RIC —
(no file loaded)
(no Yahoo data loaded)
(choose a preset or type a custom symbol)
(Change column/strategy → dataset restarts)
10
Initializing adaptation
L/R/G effective parameters pending.
Automatic tuning is causal and slow. Inputs above display the current Local effective parameters in Adaptive mode. Frozen Adaptive preserves the last learned settings; Manual restores the user-entered settings. Every sample is still observed.
Anomaly: 0.00
Prediction: 0.00
Active Attractor: –
# Attractors: 0
Attractors L/R/G:
0 /
0 /
0
Participation (EMA): 1.00
Decision advantage (EMA): –
Regime F1: –
Regime F1 L/R/G:
– /
– /
–
State entropy: –
Model horizon: – samples
(~– s)
Mean horizon: – samples (~– s)
Macro horizon: – samples
(~– s)
Mean macro duration: – samples
Next regime shift in: – samples
(~– s)
Local next shift: – samples (~– s)
Regional next shift: – samples (~– s)
Global next shift (meta): – samples (~– s)
Predicted shift severity: –
Regime transitions: –
Regimes L/R/G:
– /
– /
–
Regime stats L/R/G: count
– /
– /
–, F1
– /
– /
–
Next regime: –
Decision advantage (EMA): –
MAE: –
MSE: –
MAE@24: –
MSE@24: –
MAE@48: –
MSE@48: –
MAE@96: –
MSE@96: –
MAE@192: –
MSE@192: –
MAE@336: –
MSE@336: –
MAE@720: –
MSE@720: –
Skill L/R/G: – /
– /
–
Stress L/R/G: – /
– /
–
Active scale: –
Trading: Position FLAT,
Equity 1.000,
Equity $ 10,000.00,
Local regime –,
Exp. return 0.00000