VitalGuard · Mycelium Sentinel
V5.7 · FIELD LAYER · EN / AR
Reference implementation of an offline observation node combining a table-driven finite state machine with an explainable decision layer
Apache-2.0 © mcorpai.org (Creator: ROK and Morgan J.) · zero-egress · zero-dependency
AUTO: browser language · fallback English
You do not need to read the source. The help tab explains what this tool is, how it is used, and what it cannot do.
Help and field manualEN / AR
Field ConsoleHUMAN ALERT LAYER
feedback_note
Signal SourceSOURCE
Synthetic input and the real microphone share the same decision path after feature extraction.
Envelope · Spectrum64 Hz / 256 SAMPLE
Top: four-second band-energy envelope. Bottom: 64-bin spectrum. No raw speech samples are retained.
State Transition LogDETERMINISTIC TRACE
Saving lives in the Las Vegas storm drains
An optional scenario in which a dry flood channel becomes a torrent within minutes. It steps through depth probes, the dual Bluetooth and ultrasonic mesh, the local siren, and the final last-will packet.
FLOOD SAFETYOFF
DUAL-LANE HEALTH0 / 9
WATER DEPTH0 cm
RISE RATE0 cm/min
ETA NEXT LEVEL—
Pending-validation simulation. It controls no physical radio, authenticated packet, water sensor, or public siren. The 1/15/30/45 cm and 10 cm/min thresholds are provisional and require site tests of velocity, geometry, debris, and escape time.
Scenario controlsDETERMINISTIC WATER FSM
Rain only raises readiness and sampling; it does not prove flooding. Upstream inflow can arrive without local rain, so water contact and rise rate govern the alert.
Dual-sensor brick clusterBOTTOM AND (TOP-A OR TOP-B) · 2/3
Bottom 1 cm plus top A or B wet for three ticks confirms one brick. One brick immediately warns people nearby; 2-of-3 independent bricks trigger network mesh and siren. Rapid rise, 30 cm, and 45 cm bypass quorum.
Learning may propose a quorum change but never changes life-safety thresholds automatically.
Channel cross-sectionPROBE ×3
CEILING BRAIN · BLE + ULTRASONIC
45 cm · LV1
30 cm · LV3
15 cm · LV5
1 cm · PREWARN
Recommended architecture: keep radio brain, antenna, and siren relay in a waterproof hub above expected water; run only sensor leads downward. A self-contained brick remains experimental until potting, mast, and tether validation. Never depend on submerged 2.4 GHz BLE.
Safety levels and actionLV ≠ TRUST
LV9DRY / READY
WATCH1cm · LOCAL
LV515cm · EVACUATE
LV330cm · ALL NODES
LV145cm · LAST WILL
Propagation channelsSIMULATED TRANSPORT
BLE MESH
ULTRASONIC
SIREN
HUMAN RELAY
Propagation here is a reference simulation of packet creation, hops, repeats, and channel selection. Physical nodes still require authentication, delivery acknowledgement, and radio-shadow testing.
Water-probe statusFAIL-SAFE VECTOR
An impossible upper-wet/lower-dry vector fails safe to an alert while recording SENSOR_INCONSISTENT. Radio silence alone is a device fault, not proof of water.
Last-will packetLAST WILL · VISION
Not generated yet.
At LV1 the node does not wait for quorum. It repeatedly sends evidence, depth, rise rate, sensor vector, and hop budget over every simulated channel, while disclosing that authentication is not implemented.
Event traceWHY · WHAT · NEXT
[BOOT] option inactive
Every line records observation, applied rule, and selected action. Receding water never auto-clears the alert; a dry check and manual acknowledgement are required.
Store and forwardSTORE-AND-FORWARD
0packets waiting
QUEUE EMPTY
During radio shadow, a bounded queue retains packets and releases them only to a trusted mobile hub. Physical transport and authentication are not implemented.
Fire adaptationSMOKE + HEAT · ADVISORY
DRY · NO SMOKE EVIDENCE
One smoke tile creates a local warning; two independent smoke tiles or smoke-plus-heat cross-check creates a network warning. This does not replace a certified fire detector.
Deployment and durability guidePROVISIONAL HARDWARE
ANTI-ROLLUse a trapezoid or triangular prism, bottom ballast, anchor wire, and protective cage. A loose thrown object is not a depth reference.
ELECTRODESUse an independent electrode pair at each height and redundant top A/B. Drain grooves, hydrophobic surrounds, and polarity reversal reduce mud bridging and corrosion.
PROTECTED BRAINKeep power, logic, and antenna in the protected overhead hub. An independent brick needs potting, an antenna mast, tether, and residual-power tests.
LEGO-style porting contractPURE FUNCTIONS · NO DOM
01 BRICK COREevaluateBrickNode → state, evidence, local action
02 CLUSTER COREevaluateBrickCluster → quorum and network action
03 FRAME COREencode/decodeFloodFrame → 16-byte frame
04 REPLAY GUARDacceptMeshEnvelope → identity, authentication, sequence
05 STORE FORWARDboundedQueuePush / flushToTrustedHub → bounded queue
06 FIRE ADAPTERevaluateFireTileCluster → smoke/heat cross-check
The cores depend on no DOM, storage, or radio API. Transport adapters and key management remain separate implementation work.
Idea Amplifier design basis6-STAGE
5 assumptions → constraints → 5 analogies → 4 filters → five-minute action
Official design basis. A Clark County Regional Flood Control District workbook describes a channel going from dry to about seven feet deep in eight minutes. NWS warns that about 15 cm of fast-moving water can knock an adult down. Depth, rise rate, velocity, and escape time therefore need joint validation.
Displayed values are used by the decision path. The risk estimate is not a calibrated field probability.
Evidence Ledger (log-odds)WEIGHTED LOG-ODDS · deciban
Units are decibans. Weights are human-authored design values, not Bayesian likelihoods learned from field data.
CounterfactualSINGLE-FACTOR SENSITIVITY
Similar Casesk-NN CASE BASE
RejectionINTERFERENCE MATRIX
The table states what is not treated as a threat. Human voice and human movement have no threat row.
Learning never changes the table automatically. Outputs are proposals and require human review.
Online Regression WeightsRLS λ=0.98
Confirmed and false-alarm labels estimate feature contribution. Deferred labels do not train the model; all weights remain visible.
Duty-Cycle PolicyTHOMPSON SAMPLING
Rewards include confirmation outcome and battery-limit violations. With few samples, policy selection is experimental.
Confidence CalibrationECE / BRIER
Surprise & ArousalKL DIVERGENCE
Surprise rises when observations depart from the reference distribution. It requests review; it does not change thresholds automatically.
Calibration SuggestionOPERATOR ACTION REQUIRED
Scenario Table SwapTABLE SWAP
State TableVG_STATES
State identifiers and flags are fixed contracts. Do not change thresholds or durations without field validation and a change record.
Node Type TableVG_NODE_TYPES
Threat BestiaryVG_BESTIARY
Flag Bit Dictionary32-BIT
Channel REJECT MatrixPLATFORM × CHANNEL
Node LayoutRELATIVE BEARING
Select a node to simulate relative bearing and arrival order. No satellite or personal location is collected.
Trust & QuorumTRUST LV · QUORUM
Silence Pattern InferenceOBITUARY · SILENT WITNESS
Silence is fault evidence, not disaster confirmation. Interpret it only with independent physical evidence such as water or smoke.
Observation LineageLINEAGE
Observations derived from this node's own lineage do not count. Physical transport still needs cryptographic identity and replay protection.
Self-Source AuditZERO-EGRESS SOURCE SCAN
A heuristic count of known network primitives and external-resource strings. Zero matches is not a formal proof or independent audit.
Simulated TestsTESTS · ISOLATED STATE
Runtime ProfileOBSERVED · NOT ENERGY PROOF
Observed processing time and Battery API values. Unsupported values remain explicit; these are not proof of power draw or endurance.
Structural Privacy CheckSTRUCTURAL
Stated LimitationsMANDATORY
ExportLOCAL ONLY
Export occurs only after a user action. No automatic transmission path exists.
This tab reads the source of the block that is running right now and shows it on screen. No separate download and no repository visit is needed. What you read here is what executes.
Source fingerprintSHA-256 · VANILLA
The hash is computed by a plain JavaScript SHA-256 implementation contained in this block, with no external call. It yields the same value even from a file:// copy where no secure context exists. Because an identical file gives an identical fingerprint on any device, a reviewer can compare the copy they received with the copy being demonstrated.
Mobile editorNO COMPUTER NEEDED
Function assemblerDETERMINISTIC · NOT A MODEL
Licence note for self-relianceAPACHE-2.0 ALREADY ALLOWS THIS
Why each algorithm is hereINVENTORY
TABLE-DERIVED · $0.30 MCU
The header is not transcribed by hand. It is derived from the tables that are running right now, so editing a table on screen changes the header too and the reference implementation cannot drift from the firmware. Documentation drifting away from code is the most common failure in this field, so that path has been removed.
Outbound call countCOUNTED AT RUNTIME
The table above is the block counting forbidden patterns in its own source string. It is not a pre-recorded figure; it is computed at the moment the tab is opened.
This mode measures no height. Tape a discarded handset to a brick, a piece of timber or a bundle of rubbish, put it on the floor, and the installation is finished.
The decision rests not on depth but on how quickly peers go quiet and what they said on the way out.
This is a pending-verification simulation. It drives no radio, no authentication, no speaker and no public warning system. The point table and the density table are design assumptions that must be replaced with loss statistics from a real deployment.
Attrition decisionNO HEIGHT · NO RULER
Peers heardROSTER · TAP TO LOSE
Tap a cell and that peer leaves a wet last will and goes quiet. Tap again and it comes back. The h value is a hop distance, not a coordinate.
ScenariosSHOW THE FAILURES TOO
A demonstration that only shows the success path proves nothing. A flat battery, a kicked brick and a broken radio are each an event this method must ignore, and pressing them shows the score refusing to move.
Evidence ledgerINTEGER POINTS
Every score is an integer. That removes floating point from a thirty-cent part, and it is why the JavaScript reference and the C port reach identical decisions.
Density threshold tableMORE NODES · LOWER BAR
The more peers, the smaller the loss fraction required. Seven simultaneous losses out of fifty do not happen by accident; two out of three happen all the time. This table is the exact point at which a low unit price turns into accuracy.
Height learned from orderORDINAL · NO MEASUREMENT
Watching who goes under first, several times over, reveals who sits lower. An ordering appears although nobody used a ruler, and it sharpens with every rainfall.
How to use itFOR THE PERSON HOLDING IT
Every earlier screen answered what is happening. This one answers whether the person is awake and which way is out.
For the same reason the silence of a node is treated as evidence, the silence of a person is treated as evidence too.
This is a pending-verification simulation. It drives no speaker, no siren and no rescue dispatch. The wake stages and the time-band bias are design assumptions that must be replaced by field trials.
Wake ladderESCALATE UNTIL ANSWERED
Escape guidanceLEARNED GRADIENT · NO MAP
Roll callTHREE COUNTERS · NO IDENTITY
Loss signatureFLOOD vs REMOVAL
Swarm memorySURVIVES A BATTERY SWAP
A swarm that forgets its ordering on every reboot never becomes any wiser. What is stored is a peer label, a mean position and an episode count. No coordinates, no wall-clock times, no audio, nothing that describes a person.
A browser slows the code down when the screen goes dark. For a locust survey that is an inconvenience; for somebody asleep in a storm drain it is fatal, because the value of this tool is concentrated in the hours when nobody is looking at the screen. This screen measures the problem instead of hiding it, and generates the Android shell that removes it.
Host capabilityFEATURE DETECTION
Throttle watchdogMEASURED, NOT CLAIMED
Android shell sourceGENERATED · NO INTERNET PERMISSION
Dual-channel decisionRADIO + ULTRASOUND
Peer statesR=radio U=ultrasound
ScenariosSHOW THE CONFOUNDS TOO
Cross-channel matrixFOUR CELLS
Roster and blind spots10 WITNESSES · COMPUTED
Evidence ledgerINTEGER POINTS
Managing settingsNUMBERS ONLY
Fire indicatorsINDIRECT ONLY
What a phone can seeHONEST INVENTORY
Escape routeCONSECUTIVE FLOORS
Zone statusDANGER OVERRIDES SAFE
Relay transmissionRADIO 8B · ULTRASOUND 2B
Locust bearing decisionSOUND + DISPLACEMENT
Acquainted pebblesU=4m R=35m
ScenariosSHOW THE CONFOUNDS TOO
The baton2 BYTES
Automatic link checkRULE OUT OUR OWN RADIO FIRST
Idle modeLAST WILL STAYS ARMED
ScenariosINCLUDING OUR OWN FAILURE
Result per peerRADIO → RETRY → SOUND
Swarm verdictWEIGHTED BY CREDIBILITY
Deployment plan checkGEOMETRY, NOT PROTOCOL
ScenariosKICKED vs DESTROYED
Company statusLEVEL vs RATE
Self-retractionSURVIVAL DISCONFIRMS
What is trusted less nowFLOOR · ASYMMETRIC · GUARDED
ScenariosINCLUDING THE ONES THAT MUST NOT RETRACT
Zone verdictCOMPUTED, NOT VOTED
ScenariosINCLUDING A LYING COORDINATOR