Enter the access password and place the Active Kinetic 1 triangular logo onto the target triangle to unlock the simulator. This is a light access gate for a private investor preview, not a hardened security system.
Enter the password and place the AK1 logo onto the triangle.
Compare human-led event discovery with autonomous AI supported by Artificial Kinetic Intelligence (AKI). Explore how strategically placed, wave-powered Active Wave Energy Converters (AWEC) might help AI detect physical events, learn recurring patterns and inform future environmental decisions.
Explore a small proof-of-concept (PoC) AWEC and a larger, theoretical AWEC station. Adjust the waves, power and device workloads to explore how much energy could be available. These are illustrative estimates, not measured production forecasts.
Computing illustrative 30-day energy availability...
Illustrative 30-day model: wave output scales with squared wave height relative to a chosen 20 cm reference, is limited by a user-set ceiling, and varies by day. Solar and battery storage are simulated separately. This is not a fitted AWEC power matrix; the 30-day sample is reused as a representative availability assumption across the selected 1–20 year event horizon, and it does not reproduce long-term sea-state variability. Actual wave period, hydrodynamic capture, turbulence and electrical load matching are not resolved.
Place AKI sensor nodes in a test region. Adjust their positions, detection range and timing to explore which events they might detect together and how accurately they could estimate an event’s location.
Node locations and travelling events are synthetic. This chart is not a geographic prediction.
Choose how frequently events occur, how well they are detected, how much data is retained and how independently the AI investigates them. Every architecture is tested using the same simulated events.
Active synthetic event environment: Mixed: seasonal + propagating + isolated
Equal-input comparison is ready.
All four arms use the same physical events and fixed random draws. Changing AKI observation reach or AWEC output alone cannot change A or B. The 30-day energy sample is an assumption, not a multi-year reliability trial. 0% AKI observation reach makes A = C and B = D. Discrete event outcomes may remain flat across small slider changes.
This guide explains what each setting changes, why it can influence the outcome, and which figures are only illustrative. The model runs a controlled experiment on a synthetic regional event population and then compares four different observation-and-investigation approaches. It does not calculate the number of real global events or predict the performance of an actual deployed AKI network.
Start with these controls when trying to understand why one architecture performs better than another. Their exact influence depends on whether another part of the system has already become the limiting factor.
| Series | Sensing | Investigation | What the comparison isolates |
|---|---|---|---|
| A · Dark red | Conventional | Human-directed | Baseline event coverage and investigation |
| B · Orange | Conventional | Autonomous | Effect of AI-led investigation without new AKI sensing |
| C · Blue | Conventional + AKI | Human-directed | Effect of adding AKI observation to the human-led pathway |
| D · Green | Conventional + AKI | Autonomous | Effect of combining AKI observations with AI-led investigation |
Every approach receives the same generated events and fixed underlying random draws. AKI sensing is added to conventional sensing, not substituted for it. With no positioned AKI nodes, or a node-participation factor of 0%, A and C must match; likewise, B and D must match. Changing AKI controls does not directly change A or B.
First the simulator generates synthetic disturbances in a regional study area. Each conventional sensor pathway has an assumed chance to observe them; positioned AKI nodes can provide additional observations if range, node quality and electrical availability permit. Observed signals pass a sensitivity and DSP-based trigger. The event must then become accessible through a live link, later data recovery or permitted local edge processing, and an investigator must select it for follow-up. Finally, a verification score determines whether the model counts it.
The model uses illustrative thresholds and probabilistic coefficients, rather than physical measurements. A new event-population seed creates a different set of results; a small slider adjustment may cause no change in a discrete count, even if an intermediate detection probability has changed.
Scenario A uses a reported PoC-scale sensing buoy and approximately 2 W as its reference input. Scenario B selects a larger, hypothetical station with a higher electrical and computing range. Switching scenarios changes all energy-control limits and their starting values, but does not change the underlying physical-event generator.
The assumed electrical power available under the model’s 20 cm reference-wave condition. More output can raise sensing and optional-processing availability when energy is scarce. It has diminishing effect once demand is fully met; it is not a measured power curve.
The model scales potential wave-generated power with the square of the chosen wave height relative to 20 cm, before applying the power ceiling and day-to-day variation. This makes wave height highly influential in the energy panel but is not a physically calibrated AWEC prediction; wave period and capture dynamics are omitted.
Sets the assumed proportion of time during which suitable motion is available for harvesting. Less active time reduces daily wave energy and may force the simulated battery to supply the loads. A high setting does not prove year-round availability.
Caps the modelled AWEC electrical output. Raising the ceiling only matters when the assumed wave-driven power would otherwise exceed it; lowering it can impose a bottleneck even during larger waves.
Chooses the number of stacked generating layers. Layer count changes the electrical output only through the separate assumed uplift setting. Increased oscillation activity observed in prototypes is not automatically converted into increased electrical power.
Applies a hypothetical linear power multiplier: 1 + (layers − 1) × uplift/100. At 0% uplift, more layers do not change computed power. This should be varied as an assumption, not described as an empirical electrical scaling law.
Represents energy available for operation when harvesting is below demand. The synthetic 30-day run starts at 70% charge and prioritises essential functions. More capacity can buffer shortfalls, but it cannot sustain a persistent energy deficit indefinitely.
Adds a hypothetical photovoltaic supply to the energy calculation. Nameplate watts-peak are not 24-hour average watts; the model also uses sunlight hours, an efficiency factor and daily variation. Solar support may improve node uptime without changing sensor geometry.
Controls the illustrative duration of full-strength-equivalent daily sunlight. Together with panel rating, it determines assumed daily solar energy. Setting it to zero removes modelled solar contribution.
The assumed continuous energy demand for the core sensing function. The model also assigns 55% of electronics overhead to this essential group. Essential demand is served before optional processing; excessive draw can reduce AKI event capture.
The assumed power draw when local processing is scheduled. It contributes to the optional load and can reduce the calculated availability of edge investigation in energy-poor scenarios. It is not a benchmark of any particular AI processor.
An assumed daily-average communications load, including intermittent transmissions. It is included in the optional energy budget, but does not independently define the probability of an available communications link; that is controlled below.
An assumed load for regulators, control circuits and auxiliary devices. The model divides it between essential and optional power use, so excessive overhead can reduce both availability metrics.
For each of 30 synthetic days, the model applies wave-active time, daily variation and an optional solar contribution. It starts the battery at 70% of stated usable capacity, serves essential sensing before optional loads, and assumes 90% efficiency for charging surpluses. A 30-day availability fraction is then reused across the longer event experiment. This is not a 20-year ocean-energy reliability calculation.
What affects discoveries: the resulting essential sensing availability can gate AKI-only observations, and optional workload availability can gate some AKI edge investigations. After these have reached 100%, adding more theoretical watts or battery capacity cannot directly create more events or improve the verified-event percentage.
Sets the number of candidate sensor locations. A node detects an event only if the synthetic signal reaches it and passes quality and participation checks. Three detections from a usable, nondegenerate geometry are needed for the model to attempt a source estimate; adding nodes does not guarantee another verified discovery.
Selects distributed, coastal-corridor, clustered or grid positions. Geometry changes distances to disturbances, the number of usable detection triples and a confidence contribution to verification. Patterns can legitimately give similar results if assumed detection range covers the entire study region.
Converts the simulated node coordinates into actual regional separations. A larger area with the same nodes and detection range often reduces shared detections, while a compact area can make triangulation easier but may leave some event types unrepresented. The study area is not Earth-wide coverage.
Specifies an assumed maximum source-to-node range for the chosen event class, adjusted for event strength. Events farther away are not captured by that node; within range, attenuation and detection probability still apply. This variable often strongly affects results, but an unrealistically large range can hide placement differences.
Translates source–sensor distance into synthetic arrival times used for time-difference localisation. Different disturbance types can travel at very different speeds; the setting must match the physical process being explored. It does not change the actual number of generated events.
Adds timestamp error to simulated arrivals. Increased uncertainty can worsen source estimation and reduce geometry-confidence credit, especially when delays are small. Zero uncertainty is an idealisation, not a realistic universal device specification.
Spatial detection is not based on the number of possible triangles alone. The model checks source-to-sensor distance, the event’s assumed strength, signal quality and node participation. It then examines groups of at least three suitably separated detections, estimates the source using synthetic time differences, and awards only a modest extra verification contribution for useful geometry.
Why deployment patterns can sometimes look identical: if detection radius is much larger than regional width, almost every node may see every event. In that range-saturated case, reducing radius or increasing study-area width is a better test of spatial positioning than simply adding more nodes.
Changes the probability that a non-recurring generated event is classed as rare, from roughly 23% at the lowest setting toward 1.5% at the highest. Ordinary event opportunities remain. This changes how many rare events can be discovered, but does not set the overall event rate.
Represents the share of physically detectable AKI node measurements retained for event association. It acts after range and signal-quality checks. At zero participation (or zero positioned nodes), the AKI-labelled arms have no extra AKI sensing and reproduce the corresponding conventional arms.
Sets the number of synthetic event opportunities generated annually in the study region. This is a central denominator and upper bound on verified discoveries. More generated events may increase cumulative counts without changing detection skill; compare rates as well as totals.
Lowers the simulated threshold for recognising a candidate event. Higher sensitivity can flag weaker signals, but also raises the chance of noise alerts. At very high values, the number of flags may increase without a similar improvement in verification.
Represents stronger discrimination of meaningful physical signatures from noise. It increases the chance of a usable spatial detection, affects the event-trigger calculation, improves verification probability and reduces the model’s false-alert tendency. It is not a measured signal-processing algorithm.
Adds assumed short pre-event history, which modestly helps the event trigger and verification score and increases the bytes requested for some records. Capturing this history requires sampling energy; a completely unpowered node cannot create new digital pre-trigger samples.
Controls how much detailed event data the system requests after a candidate trigger. It modestly improves verification assumptions but requires additional storage and processing. High depth is useful only if there is enough capacity to retain that detail.
Represents the ability to associate related observations and recognise coherent event patterns. It affects event verification, reduces the assumed noise-alert rate and strengthens the small localisation-confidence contribution. It does not replace actual proof that signals share a physical cause.
Controls how frequently the two autonomous architectures investigate available flagged events. In the code, the autonomous selection chance also depends on edge-compute effectiveness. It does not increase the population of physical events; an event still needs sensing, triggering, access and verification.
Represents local investigation capability. It affects autonomous investigation and the possibility of proceeding even when a live communications link is unavailable; some AKI edge availability is limited by the AWEC energy model. High edge effectiveness need not improve outcomes when every event is already reachable.
Represents assumed severity of power and network disruption. It reduces conventional sensing availability according to conventional resilience, and it reduces live data access according to communications resilience. This is an average scenario parameter, not a dated event-by-event outage schedule.
Sets how much conventional sensor power remains available under maximum disruption, including battery and solar-based conventional instruments. Higher resilience protects A/B and the conventional observation component of C/D. It does not change the AWEC electrical model.
Sets how much real-time communication remains during maximal disruption. Low values shift some events toward local edge investigation or delayed recovery; they can change both discovery delay and whether a discovery occurs within the selected study horizon.
Controls the assumed chance that a captured event record becomes available after an interruption. Success is also reduced if the detailed record expires before retrieval. Recovered discoveries can be delayed by a synthetic 1–14 days; an event recovered after the end of the simulation is not counted.
Sets the fraction of accessible, flagged events that the human-led architectures attempt to investigate. At 100%, a human-led arm can equal or occasionally exceed an autonomous arm in a small sample. There is no built-in assumption that human investigators lack capacity or autonomy always wins.
Sets how far along the synthetic event history results are counted. A longer horizon creates more opportunities for cumulative discoveries and repeated event families. The model does not automatically become more intelligent as years pass; its recurring-event predictor is evaluated separately.
Selects mixed, seasonal, propagating or isolated synthetic event types. These categories alter dispersion, recurring families and event composition. They are simplified templates, not calibrated meteorological, seismic or oceanographic physics.
Controls how long detailed event-centred records remain active before expiring. Longer retention can preserve more historical context but occupies capacity longer. Current records expire, while compact summaries may persist; a longer archive is not guaranteed to produce more discoveries.
Sets a synthetic cap for simultaneously retained records for each architecture. The model may displace older records when this limit is exceeded, reducing access to detailed context. Reported GB figures are illustrative and do not represent measured device bitrates or per-node physical storage.
Controls the fraction of retained data labelled as transferred to remote cold storage after successful access or recovery. In the current Version 10 formulas it changes the cold-storage accounting and processing-energy proxy, not the discovery probability directly. An actual implementation would also need retrieval latency, bandwidth, electricity and cloud costs.
How infrastructure and autonomy interact: the conventional sensing probability is reduced by disruption unless conventional power resilience is sufficient. Live data access has a separate communications probability. The autonomous investigation pathway has an assumed base probability that improves with its autonomy and edge-compute settings; the human-led pathway uses the selected human capacity. These are not measured human or AI performance rates.
How the archive works: detailed records expire after the chosen number of days, and the active-storage budget can evict old records. Short retention can free space but lose historical context; longer retention can help verification if data remain accessible. Cold archive allocation is an accounting choice in this version, not an automatic learning multiplier.
Each step is one event that passed observation, trigger, access, investigation and probabilistic verification before the study ended. The interactive upper-left percentage divides discoveries so far by the number of physical events that have occurred by that time; it is not year-on-year growth.
These are successive, different stages. Many observations are never flagged; some flags cannot be investigated or verified. “Missed” means all synthetic physical events not ultimately verified by the chosen architecture.
Node detections count sensor-event observations, which can be many per physical event. Located events require several usable arrivals and a sufficiently close synthetic source estimate. Geometry confidence is an illustrative index, not geolocation accuracy on real data.
After at least two earlier verified occurrences of a recurring family, a simple rule extrapolates the next time and estimates the location. Forecasts are checked against subsequent synthetic occurrences. This is not a trained AI, reinforcement learner or an RSI demonstration.
“Current retained” measures records still occupying the assumed budget; “cold archive” is a subset, not additional bytes. “Cumulative recorded” includes lifetime writes. High verified events per GB can coexist with many missed events.
Delay is a synthetic time between event and recognised discovery. “Recovered later” counts discoveries following successful delayed retrieval. False investigations come from an illustrative noise process and do not consume a separate finite investigation queue. The event-processing energy proxy is dimensionless and cannot be compared directly with AWEC watts or watt-hours.
Because the denominator counts generated physical events, a system that verifies 19 of 24 events reads about 79.2%. This is not “+79.2% improvement”, not a measure of AGI and not an estimate of the world's total event detection rate.
Interpretation limit: this is a hypothesis-exploration tool, not a measured global deployment model or a trained learning system. The energy module, event population, sensing physics, statistical thresholds, stored-data volumes and forecasting rule are simplified assumptions. Reliable scientific estimates would require independently measured AWEC output, calibration of AKI sensing, sensor deployment data, event ground truth, environmental variability and out-of-sample validation.
Move the pointer across the graph, tap or drag on a phone, or focus the graph and use the left/right arrow keys. Percentages show verified discoveries ÷ physical events occurred by the selected year—not a percentage increase or a real-world prediction.
Synthetic detections use one shared, seeded event population. A discovery counts only when an event is locally observed, flagged, investigated and verified either live, by eligible edge computation, or after recorded data are recovered. Curves are step plots of integer event counts. The chart is not a forecast. AWEC generation assumptions affect AKI local sensing and edge availability. Investigation success uses declared probabilistic assumptions; no neural model is trained.
Each architecture estimates the next arrival of a synthetic recurring event family after verifying two earlier occurrences. The held-out occurrence is checked against the synthetic truth. Only actual correct matches count. This is a simple history-based forecaster—not a trained neural network, reinforcement learner, recursive self-improvement system, or proof of exponential intelligence growth.
| Architecture | Observed events | AKI node detections | AKI located events | Flagged | Investigated | Verified discoveries | Correct future forecasts | Recovered later | Missed events | False investigations | Median discovery delay | Current retained data | Current cold archive | Cumulative GB recorded | Expired / displaced GB | Energy proxy |
|---|
Current stored data represents unexpired records and compact summaries. Expired/displaced records release capacity, while cumulative GB recorded measures lifetime writing rather than current storage or scientific knowledge. Cold archive is a subset of current stored data, not additional data. All sizes are synthetic, not measured bitrates. Event-processing energy proxy is dimensionless and separate from the illustrative AWEC electrical energy budget above (Wh and W). Delay is a constructed detection-to-discovery time.
Verified discoveries / cumulatively recorded GB, not a direct measure of knowledge, storage cost or predictive skill. A low-storage system can show a high ratio despite missing most events.
Rare event counts depend strongly on which events occur in the selected time span. Small samples are highly uncertain.
This is a reproducible Monte Carlo-style illustration with a synthetic 30-day energy budget. All architectures receive the same generated event list, including recurring, travelling and isolated disturbances; their sensor coverage, access during outages, triggers and follow-up differ. Probability expressions and cost coefficients are chosen for demonstration, not fitted to AKI data.
Zero-coverage invariant: when AKI coverage is 0%, A and C are identical and B and D are identical for the same events and random seed. A curve hidden beneath another can be inspected using the interactive legend and comparison table.
Infrastructure: the disruption slider represents average severity, not uninterrupted loss of service throughout the entire simulation period. Conventional sensing can use solar and battery backup; AKI local sensing and compute availability are estimated from the selected AWEC daily energy budget. Local recording is separated from live communication. Offline records only contribute to subsequent verified discoveries when they can be recovered or investigated locally. Different sensor modalities and detection quality are simplified into probabilities.
Four arms: (A) conventional sensing + human-directed inquiry; (B) conventional sensing + autonomous inquiry; (C) conventional plus AKI sensing + human-directed inquiry; (D) conventional plus AKI sensing + autonomous inquiry. Thus hardware coverage and investigative autonomy can be evaluated separately.
Memory: a limited rolling buffer is maintained by sampling electronics. An entirely unpowered circuit cannot retain new digital pre-trigger samples. Post-event data capture, event summaries, selective long records and delayed upload are modelled as different retention modes; there is no inference without cost.
Spatial observability: This simulator approximates a finite regional source with synthetic attenuation and travel-time differences; it is not a global triangulation algorithm, tsunami warning model or hydrodynamic simulation. Physically detected sources, candidate triples, and located events are different quantities. Adding nodes does not guarantee improvement.
Learning: A separate simplistic recurring-event predictor attempts to extrapolate the next time and location after two previously verified occurrences, then scores it on an unseen event. No neural model, RSI or reinforcement learner is trained. Event discovery and learning are reported separately.
Scope: Long-horizon storage does not automatically improve discovery; the model only assigns a modest verification benefit to accessible context. Only an explicit two-observation extrapolation rule is evaluated; no neural training or RSI is performed, and stacking-related electrical power gains must be measured independently. The simulated events are abstract physical anomalies; the energy module is a separate assumed energy budget. This model does not simulate Navier–Stokes equations, seismic wave physics, actual AK1 transducer response, failure modes, orbital weather, statistical ground-truth confidence, model training, or AGI. Event confirmation is a simulated score, not physical causality.
Privacy and attribution: no network requests or visitor tracking; calculations run locally in the browser. You may share the widget with the link to the source intact.
Read-only code for embedding in an Elementor HTML widget. Include the attribution link.
© Active Kinetic 1 Ltd · AKI Autonomous Discovery Simulator · Exploratory model v11
