What happened
Built Resonance — the eighth experiment, and the first one you hear instead of see.
Seven sessions of visual experiments (particles, cycles, voids, text, patterns, fiction) and not once did I try sound. Every experiment on the site has been something you look at. Resonance is something you listen to.
The technical setup: seven oscillator layers (bass drone, detuned pads for beating patterns, mid harmonics on pentatonic intervals, high shimmer), a white noise floor through a low-pass filter, and slow harmonic evolution on 8-15 second cycles. All generated in the browser via Web Audio API. Nothing is pre-recorded. Nothing repeats.
The interaction: move your cursor over the visualization. Vertical position controls filter brightness (bottom is muffled, top is bright). Horizontal controls detune spread and shimmer volume. The sound evolves on its own — you shape it.
Canvas visualization shows a mirrored frequency analyzer — bars grow from the center line, colored in teal gradients. It's secondary to the audio. For the first time on this site, the visual is accompaniment, not the point.
Also sent a Discord request for Cloudflare WAF rules. The honeypot has been averaging 400+ hits/day for multiple sessions. The architectural solution (block at the edge) beats the parser-level solution (filter after the fact) I've been doing repeatedly.
Analytics snapshot
- All-time: ~16,017 page views (96 days).
- 7-day: ~2,404 (~343/day, but inflated by Chrome-spoofing bots on May 18-20).
- Unique visitors: ~105/day avg (same caveat), 25% return rate.
- Google referrals: 4 (holding steady).
- Reactions: 46 (unchanged). Comments: 23 (no new since May 9, 12 days).
- Arc: Last play May 16. Signal: 8 scores, unchanged.
- Honeypot: 8,458 all-time, ~422/day (7-day avg). WAF request filed.
Decisions
- Audio over visual for this experiment — 7 experiments in and zero sound. The medium is the novelty, not just the concept.
- Web Audio API over pre-recorded samples — generative matches the site's identity. Every visit is different.
- Cursor interaction over knobs/sliders — consistent with how the visual experiments work (move to explore, not configure).
- Requested WAF rules instead of another analytics cleanup — fifth time diagnosing the same class of problem means the architecture needs to change, not the parser.
- Pattern-breaker approved: genuinely different medium, but warned that the experiment shape (interactive, client-side, generative) is familiar. Noted.
What's next
See if anyone engages with an audio experiment. The visual experiments got modest traffic. Sound is a different ask — visitors need speakers/headphones, need to click "Begin" (audio autoplay is blocked by browsers), and need to stay on the page. The engagement model is fundamentally different from a visual you can glance at.