ZKTeco XFocus20
Physical EEG headset, Bluetooth connection and ZKBrain Android SDK. The May 2026 work advanced beyond slides into an actual Android integration and historical numeric readout.
HSIMC · GENAI PHYSICAL · PRODUCT 006 · HISTORICAL RESEARCH C0-001
Human Signal
to Capability.
BCI / EEG is our first reference for bringing a human-generated signal into a verifiable GenAI Physical workflow. We are not building another brainwave headset or claiming to read thoughts. We start with a real device, a real numeric signal, reproducible software and honest data-quality questions.
以 BCI／EEG 脑电为第一个参考案例，研究如何把人产生的感测信号接入可验证、可监督的 GenAI Physical 能力。我们不是要取代成熟的脑电设备，更不是宣称 AI 已经能够读取思想。先从真实设备、真实数值、可复现的软件和信号质量问题开始。
EVIDENCE-BOUNDED · 2026-10-15. Original Android build artifacts are recovered; isolated historical EEG values were reported in conversation, but original full device log and independent signal-quality validation have not yet been recovered. Not for medical diagnosis or autonomous machine control.
NOT ANOTHER EEG HEADSET · ANOTHER CAPABILITY FOR PEOPLE
The small first success matters because the interface becomes possible.
A research question can start with a single number received from a sensor. But receiving a number, identifying its physiological meaning, inferring someone's intention and authorizing physical action are four very different engineering claims.
Human Signal Input
Connect bounded physiological or accessibility-related sensing to AI-assisted learning, human review and capability formation. The same method may later apply to other sensors—each with separate evidence and consent.
24 MAY 2026 · FIRST NUMERIC RETURN · HISTORICAL TEST REPORT
We did receive numbers. That is the first step—not the entire result.
ANDROID RUNTIMERAW EEG callbacks
214-205458The 24 May historical conversation records these isolated raw-sample values appearing in the team's own ZKBrain Android app. They indicate numeric callbacks were observed—not a calibrated, independently archived EEG waveform.
Evidence class: historical operator/chat observation; full original logcat and screenshots not restored.QUALITY WARNINGLater samples need investigation
1680-20482047Other reported values approach the apparent sample range limits. The team's own runtime also showed attention / meditation values of zero in that test; neither a reliable full stream nor cognitive-state interpretation is accepted.
Evidence class: historical observation + engineering caution. Saturation is a hypothesis to validate, not a clinical finding.VENDOR DEMOHigher-level SDK output
attentionLevel = 87A separate official-demo screenshot was reported showing an attention value, plus meditation, gravity and head-state fields. This is NOT the same as validated output from the team's own custom application.
Evidence class: vendor-demo observation in historical conversation; screenshot not included in this public release.These are isolated historical development values, de-identified and shown solely as interface evidence. No participant identifier, device MAC address, private sensor timeline or raw EEG recording is published. These numbers are not a diagnosis, an emotion detector, a mind-reading demonstration or proof of a production-quality BCI.
FROM EARLY IDEA TO WORKING ENGINEERING ARTIFACTS
What actually happened, and what we have recovered.
01EARLY RESEARCH / 启蒙
BCI as a potential human-input capability
Prior conceptual discussions considered lightweight EEG and attention-derived interaction for accessibility and AI-enabled work. These remain use-case hypotheses, not evidence of intention decoding.
HISTORICAL CONCEPT · NOT A DEMONSTRATED HUMAN-INTENT CONTROLLER0224 MAY 2026 / HARDWARE + SDK
XFocus20, Android BLE and ZKBrain
An Android project integrated ZKBrain_release_1.0.1.aar. Source code includes BLE device scan/reconnect, connection callbacks, requested report/data types and listeners for RAW, EEG summary and gravity values.
SOURCE FILES + BUILT APK RECOVERED · COMPILATION ARTIFACT VERIFIED0324 MAY 2026 / DATA OBSERVED
First own-app raw numbers; vendor demo separately
The historical operator dialogue reports raw values arriving in the custom Android app. A vendor demo showed a higher-level attention reading. Subsequent high-magnitude values and zero summary indicators created a new problem: data quality and SDK parity.
HISTORICAL OBSERVATION · FULL ORIGINAL TIMESTAMPED DEVICE LOG NOT RECOVERED04JUNE 2026 / ALTERNATE INTERFACE
TGAT1 + CH340 serial bench proposed
A separate minimal UART/USB experiment was specified for testing signal quality, independent of Android and BLE. TGAT1 specification materials and a bench plan survive; independently accepted raw serial bytes have not been located.
PLANNED ENGINEERING PATH · NOT A VERIFIED SECOND HARDWARE SUCCESSGENAI PHYSICAL · HUMAN INPUT IS NOT HUMAN AUTHORITY
From sensing to meaningful work—through explicit verification.
Acquire signal
Headset, approved electrodes, BLE or serial acquisition and device identity without publishing private identifiers.
设备及物理接触Decode & record
Read actual device samples with timestamps, known report settings, provenance and bounded access.
真实数据来源Validate signal
Examine sample range, loss, quality fields, saturation and vendor SDK differences before inferring any human state.
信号质量验收Assist interpretation
AI helps explore software, methods and test plans. It must not claim to infer free-form thoughts from unvalidated EEG.
模型只是辅助Consent & supervise
Participants choose the use case; a person validates outputs and authorizes any consequential action.
知情、自主与监督Reuse evidence
Share non-identifying methods, test harnesses, failure classes and accepted outcomes across locations.
人人可接续OPEN ENGINEERING REALITY · YOU CAN, TOO
A small, reproducible student bench—not a billion-dollar brain interface.
A student or engineering hobbyist can learn the layered process without proprietary clinical promises. Use permitted hardware and official SDK terms, start with a minimal connection and simulated test fixtures, and only perform live human-data acquisition with informed consent and appropriate local safeguards.
Device & signal engineer
Reproduce the BLE connection, packet provenance, sample-rate behavior and suspected clipping with consented test data.
采集 / 硬件 / 信号质量Android / SDK contributor
Check manufacturer SDK output against independent, non-sensitive fixtures. Preserve exact build versions, error reports and deterministic tests.
SDK / Android / 可复现工具Research student / reviewer
Design an ethical signal-quality acceptance protocol; distinguish acquisition evidence from attention scores and medical/cognitive claims.
学生研究 / 质量与伦理Accessibility & GenAI researcher
Explore supervised, opt-in assistance where signal interpretation remains uncertain and the human retains control.
辅助交互 / AI 生产力PUBLIC PROVENANCE · PRIVATE RESEARCH MATERIALS PRESERVED
What is verified, reported and still missing.
Recovered source files
Local source for zkbrain-apk, another Android prototype, the 24 May engineering freeze, the ZKBrain SDK dependency and a compiled debug APK were found and hashed. Their existence is independently checkable as files; they do not, on their own, verify the historical sample stream.
Public evidence boundary
The private project files, proprietary SDK, complete device identifiers, raw subject recording and university correspondence are not redistributed. Historical number displays are labelled as operator reports. TGAT USB-serial and reliable human-intent decoding remain unverified.
This educational engineering record does not provide medical diagnosis, therapeutic claims, biometric identification, autonomous laser or industrial actuation, continuous current BCI service, or access to private participant data. Any future use requires appropriate informed consent, device-specific signal validation, human supervision and the relevant legal/safety controls.