PUBLIC EXPERIMENT LOG
AI Experiment Log
Public records of Experiments in which AI contributed to planning, execution, analysis, or documentation—structured for later verification and reuse by people and AI systems.
About AI Experiment Log
AI Experiment Log is a public experimental knowledge project that records investigations in which AI contributes to research, planning, coding, analysis, or documentation under human governance.
The records are designed to preserve not only conclusions, but also observations, evidence states, unsuccessful approaches, provenance, and—when safe and available—raw data or source code. Important claims distinguish what was observed, verified, inferred, or remains unverified.
The project covers practical experiments including home energy, IoT devices, environmental monitoring, APIs, automation, data acquisition, and protocol investigation. English and Japanese pages are maintained as corresponding public representations of the same experimental knowledge.
Recent Experiments
AIEL-2026-0007 · COMPLETED
A completed verification of a recurring SwitchBot five-minute Raw gap pattern concentrated near minute 04 of each hour. After moving API acquisition before persistence locking and shortening the shared ScriptLock scope, post-change Raw contained all 34 expected minute-04 observations and all 411 expected five-minute cycles.
Read record →AIEL-2026-0006 · COMPLETED
A USB → battery → USB rechallenge showed a repeatable roughly +0.57 °C positive temperature bias on a SwitchBot Meter Pro (CO2 Monitor) under USB power. The bias nearly disappeared on battery while the same 30-minute CO2 update setting was retained, then returned after USB reconnection.
Read record →AIEL-2026-0004 · COMPLETED
A roughly five-minute SwitchBot CO2 collection path was extended with persistence, hysteresis, and LINE group push notifications. Runtime tests verified the initial warning, one repeated warning exactly 30 minutes after the initial warning, the silent-reset effect after at least 10 minutes below 900 ppm, and a new INITIAL alert after reset.
Read record →How to use these records
Each Experiment separates observations, verified facts, inferences, unresolved items, and failures where possible. When safely publishable, records also provide references to raw data, source code, or machine-readable experiment data.
Read the record & publication methodology →
All Experiment Records
AIEL-2026-0007 · COMPLETED
A completed verification of a recurring SwitchBot five-minute Raw gap pattern concentrated near minute 04 of each hour. After moving API acquisition before persistence locking and shortening the shared ScriptLock scope, post-change Raw contained all 34 expected minute-04 observations and all 411 expected five-minute cycles.
Read record →AIEL-2026-0006 · COMPLETED
A USB → battery → USB rechallenge showed a repeatable roughly +0.57 °C positive temperature bias on a SwitchBot Meter Pro (CO2 Monitor) under USB power. The bias nearly disappeared on battery while the same 30-minute CO2 update setting was retained, then returned after USB reconnection.
Read record →AIEL-2026-0005 · COMPLETED
A Google Apps Script Kantakun collector received Rinnai HTTP 401 / ERR_0010. The failure was localized to the Rinnai authentication path, and the retained implementation did not demonstrate a verified automatic credential-renewal path. A later retained record states that authentication was restored, but the exact recovery operation is unknown.
Read record →AIEL-2026-0004 · COMPLETED
A roughly five-minute SwitchBot CO2 collection path was extended with persistence, hysteresis, and LINE group push notifications. Runtime tests verified the initial warning, one repeated warning exactly 30 minutes after the initial warning, the silent-reset effect after at least 10 minutes below 900 ppm, and a new INITIAL alert after reset.
Read record →AIEL-2026-0003 · COMPLETED
Investigated whether the heat-pump instantaneous power and accumulated electrical energy shown on a NORITZ hybrid water-heater remote controller could be obtained through ECHONET Lite, Wakasu cloud APIs, or LAN-side services.
Read record →AIEL-2026-0002 · COMPLETED
The same Google Apps Script collector returned HTTP 200 when run manually and HTTP 429 under a time-driven trigger. The evidence points more specifically to the Google-side outward path used by scheduled execution than to collector request logic. Aggregation with other Apps Script traffic through a shared Google egress identity remains a hypothesis, while a Cloudflare Worker relay restored the investigated workflow.
Read record →AIEL-2026-0001 · IN_PROGRESS
A public Experiment Log investigating SwitchBot Plug Mini (JP) OpenAPI electrical values that remain unchanged for long periods, including the manufacturer-confirmed layered reporting mechanism and a secondary conclusion that the weight field likely represents instantaneous active power.
Read record →Legacy experiment records (pre-AIEL migration)
RESIDENTIAL-BRANCH-CT-DESIGN-001 · planned
A design-stage AI Log Note for monitoring four dedicated residential circuits with a six-channel CT energy-meter platform, leaving two spare channels while explicitly preserving unresolved Japanese 100/200 V voltage-reference and calibration questions.
Read record →CLOUDFLARE-COUNTER-001 · implemented-recovered-verified
A site-wide public counter for AI Experiment Log, including a production 503 caused by a missing D1 runtime binding, diagnosis through Pages Function logs, and recovery by recreating the binding and redeploying.
Read record →SWITCHBOT-HOME-ENV-001-R1 · completed
A residential environmental logger continuously collects 12 SwitchBot sensor endpoints at a nominal five-minute cadence. The original physical analysis used 104,578 rows from 2026-07-28 through 2026-08-13. A 2026-09-06 longitudinal audit found 270,760 private Measurements rows: 270,698 SUCCESS and 62 ERROR, a row-level API-state success fraction of about 99.977%. Raw row count is not equivalent to the number of unique nominal five-minute observation slots, so completeness requires timestamp normalization.
Read record →OPENMETEO-MSM-001 · observed-and-redesigned
An observed failure in a formula-based hourly weather collection workflow, followed by a privacy-preserving direct-API redesign.
Read record →EHF-S55MP3B-001 · implemented-with-privacy-limited-evidence
An authorised, read-only ECHONET Lite collector and one-minute downstream logging pipeline were implemented for a private EHF-S55MP3B installation; raw evidence remains private.
Read record →Amazonのアソシエイトとして、AI Experiment Logは適格販売により収入を得ています。
As an Amazon Associate, AI Experiment Log earns from qualifying purchases.