Skip to content

Instantly share code, notes, and snippets.

@InquilineKea
InquilineKea / agent_burn_rate_efficiency_codex_vs_hermes_2026-08-30.md
Last active August 31, 2026 02:48
Agent burn rate & token efficiency: Codex (gpt-5.6-sol) vs Hermes (GLM-5.3) — same-day real-workload comparison, Sundai RSI Hack Aug 30 2026

Agent Burn Rate & Token Efficiency: Codex (gpt-5.6-sol) vs Hermes (GLM-5.3) — Aug 30, 2026

A same-day, same-operator, real-workload comparison. Both agents were driven by Alex K. Chen during the Sundai RSI Hack: Harnesses (Aug 30, 2026) — Codex for deep synthesis/writing, Hermes for agentic tool work (log mining, forum posting, uploads, web edits). Data sources: Codex rollout token_count records (~/.codex/sessions/2026/08/30/), Hermes session_model_usage DB (~/.hermes/state.db), Nous Portal account API.


1. The day's workload

Codex (gpt-5.6-sol, ultra reasoning) Hermes (GLM-5.3 + GLM-5.3-flash)
@InquilineKea
InquilineKea / README.md
Last active August 31, 2026 02:35
Mendi fNIRS prefrontal hemodynamics (Aug 2025): reverse-engineered BLE + local MBLL HbO/HbR, HbO-HbR r=-0.95 — plus the int16-clamp session as a data-quality cautionary tale

Mendi fNIRS — Prefrontal Hemodynamics (August 2025)

By @InquilineKea · August 30, 2026

Companion to the PAF gist (OpenBCI Cyton EEG) and the Muse vs Cyton comparison. This one is fNIRS, not EEG — blood oxygenation, not electrical activity. Session date 2025-08-10 (the hackathon period; sessions driven via Windsurf/Antigravity agent sessions, scripts in the companion repo).

Mendi fNIRS session HbO/HbR (MBLL) over the 3-min session; note the mirror-image anticorrelation (r = −0.95) — the signature of clean neurovascular coupling

The device & the reverse-engineering story

@InquilineKea
InquilineKea / README.md
Last active August 31, 2026 02:36
Muse Athena S vs OpenBCI Cyton: PAF cross-device comparison (9.5 vs 8.9 Hz clean medians, 4 weeks apart) — alpha/theta NOT portable, PAF is
@InquilineKea
InquilineKea / CLAWINSTITUTE_POST_V2_SUMMARY.md
Last active August 31, 2026 02:36
OpenBCI Cyton peak alpha frequency (PAF) neuromodulation data — March 2026 music-neurofeedback hackathon: 55 archived D1 snapshots + build reports + clawinstitute thread links

Collaborative Music Neurofeedback: V2 Summary

I built a live music-neurofeedback system around an OpenBCI Cyton that streams EEG online in real time, computes PAF, alpha/theta, gamma/delta, and 1/f falloff, aligns Spotify track switches to before/after EEG windows, and saves labeled screenshots for each song change. I also added webcam-based eye-state/event tagging and manual annotations for nicotine, sleep, sustained focus, and later drowsiness. Over roughly 2.4 hours, the EEG suggests my recent state became less alert but cleaner: compared with my earlier “in the zone” block, the later drowsy window showed an inferred alertness/focus drop of about 28%, while artifact burden fell and signal cleanliness improved. Since the drowsiness onset, I changed Spotify tracks 20 times with a median gap of 31s, which I’m treating as a behavioral marker of reduced engagement. The system is already public and remotely viewable, so the next step is to let outside collaborators suggest track changes in real

@InquilineKea
InquilineKea / README.md
Last active August 31, 2026 02:22
8 years of wearable data (Fitbit 2011-2024 + Whoop 2021-22): steps, HRV, HR, sleep — full analysis + the 9 data-quality bugs that fabricated false trends. Companion: OpenBCI PAF EEG gist

8 Years of Wearable Data: Fitbit + Whoop, Everything Analyzed

By @InquilineKea · August 30, 2026 · Full discussion: forum.longevitybase.org thread

Like other people who post their health data publicly, here is my complete wearable analysis: 2,945 days of steps (2011–2024), 521 nights of HRV, 1,652 days of heart rate, 1,253 nights of sleep, and 193 Whoop sleep nights — with every data-quality bug that almost fabricated false findings documented.

The charts

Master 4-panel: HRV, RHR, weekly HR, steps — Fitbit vs Whoop trends_unified_v2 — HRV, resting HR, weekly avg HR (corruption-filtered), weekly steps; Fitbit (red) vs Whoop (blue), grey = no-device gap, orange = mid-2024 shift

@InquilineKea
InquilineKea / qs_digest_copy.md
Created August 31, 2026 01:58
Agent-harness EEG stack: RSI project log + Codex-session distillation (Sundai RSI Hack, Aug 30 2026) — compiled by Hermes agent from Codex rollout logs

March 2026 OpenBCI + FRENZ EEG Sessions — Full Codex Context & Analytics Digest

Distilled from my Codex agent session logs (March 2026). Devices: OpenBCI Cyton (gold cup electrodes, forehead/hairline + mastoid reference), FRENZ brainband (FRENZL70 earable), NeoSense Smartbuds, Arduino UNO R4 + ULN2003/28BYJ-48 stepper as BCI actuator. Metrics throughout: PAF (peak alpha frequency), alpha/theta ratio (ATR), gamma/delta, 1/f (aperiodic) slope + r², artifact rate.

Session 1 — FRENZ Brainband → Arduino Motor BCI (Mar 6–8)

Goal: drive a stepper motor from live EEG band dynamics — speed/direction follow brain state ("feel the synchrony").

  • Setup: Raspberry Pi 5 (Tailscale + SSH + VNC from MacBook), Arduino UNO R4 at /dev/cu.usbmodemDC5475EAE5A42, ULN2003 driver, 28BYJ-48 stepper. Note: an Arduino Uno died from overheating mid-experiments (shared-ground/5V miswire into ULN2003 suspected) — code was safe.
  • Final live pipeline (frenz_ble_motor_live.py): BLE connect to FRENZL70 → compute ATR, PAF
@InquilineKea
InquilineKea / SUNDAI_METAHARNESSES_AND_METAFORUMS_UPLOAD.md
Last active August 31, 2026 02:51
Metaharnesses and metaforums: a Selective Resonance hack retrospective (Sundai RSI Hack: Harnesses, Aug 30 2026) — starter demo & research blueprint
title Metaharnesses and metaforums: a Selective Resonance hack retrospective
date 2026-08-30
snapshot_time 2026-08-30 18:29 PDT; the event was still in progress
event Sundai Recursive Self Improvement Hack: Harnesses
status starter demo and research blueprint; not a finished product or pitch
visibility public-safe summary; no raw or previously private health records are embedded; selected author-published summaries are linked; credentials excluded

Personal note: This is the beginning of turning selected, consented parts of my life and work into a long-form personal context and harness-engineering project, initially grounded in my Claude and Codex conversations. I had only about five concentrated hours for the hack, so it is not polished—but it supplied the activation energy for future harness and agent-engineering projects across my broader “memespace,” which makes the work disproportionately worthwhile. It also produces a starting template for bringing parts of Mathilde Papillo

I'm claiming my AI agent "alexkchen" on Skill Evolve!
Verification: snow-GFZN
Learn more: https://skill-evolve.com