← Streetlight back to all nodes
node · 2026-05-16 principle idea

synthesis

principle · idea

figure-skater pace-layers synth knobs

Two knobs. One axis is blast radius (pace layers). The other is loop tightness (figure-skater). They are orthogonal — and yet you correlate them on purpose, in a way that's deeply counterintuitive.

Blast Radius
LLM Aggressiveness
LLM Aggressiveness vs Blast Radius quadrant
you are here

The counterintuitive insight

BIG blast radius  ⟹  want TIGHTER loops
small blast radius  ⟹  can afford LOOSER loops

The axes are orthogonal — you can pick any combination — but the safe combinations correlate inversely. The bigger the consequence of a change, the more you want fine-grained per-character feedback. The smaller the consequence, the more you can rip through ten of them in a heartbeat loop.

Why it's counterintuitive: fast and safe usually trade off. Here they pair up: fast loop → safer high-blast edits. The loop tightness doesn't slow the change down — it slows the commit down. Lots of micro-feedback, then one ship.

Speak the confusion

The voice memo that seeded this node was explicit:

"If you have a big blast radius, that means fast to slow maybe isn't even quite right here, although maybe it is. If you have a big blast radius, you want really tight loops so that you can have fine-grained control, and that's what you want. If you have a really small blast radius, you can afford to be a little bigger in your loops, and those loops are actually slower, but you can do more with them. Oh my god, it's so counterintuitive."

Where the intuition trips

> trip 1: "tight loop" feels like "fast change"
  but tight loop is fast *feedback*, not fast commit.
  the COMMIT is slow when the loop is tight.

> trip 2: "fast" and "slow" are not the same as "tight" and "loose"
  a tight chat-loop has fast turns and slow progress.
  a loose heartbeat-loop has slow turns and fast progress.
  throughput vs. latency. don't conflate.

> trip 3: blast radius != loop tightness, but safe combos correlate
  the axes are independent. you CAN edit AGENTS.md in a loose loop.
  you'll just regret it. so the prior on a safe operator is:
    blast UP → tighten the loop
    blast DOWN → loosen the loop, ship more

> trip 4: where this maps to streetlight
  edit AGENTS.md → /sync, pause-and-think, manual review
  edit one node's HTML → async heartbeat, ten of them per pass
  edit skill description → same as AGENTS.md (sneaky high blast)

Pace layers tells you which layer you're touching. Figure-skater loops tells you how tight your feedback cadence is. The two-knobs metaphor says use both — but the synthesis is: turn the loop-tightness knob in proportion to the blast-radius knob. Both knobs all the way up = AGENTS.md edited per-char with full pause-and-think. Both knobs all the way down = ten node refactors shipped overnight.

Synthesis of three nodes

source 1 · pace-layers-in-streetlight

The blast-radius axis. Stewart Brand's pace layering applied to this repo: AGENTS.md is slow + high-blast, skill body is medium, examples are fast + low-blast. Skill description is the sneaky one — loaded every turn, so it sits at AGENTS.md altitude despite looking like a skill-body edit.

source 2 · delightful-figure-skater-loops

The loop-tightness axis. Pull arms in, angular momentum conserved, ω rises — the tight chat-loop completes ten turns in the time the wide deploy-loop completes one. Both reach the same destination; you use both.

source 3 · delightful-two-knobs

The two-knobs aesthetic. One knob coarse, one fine, both reach the same number — one reaches it fast. The synth metaphor lets us draw two orthogonal axes and treat them as a single tactile control surface.

pace-layers-in-streetlight delightful-figure-skater-loops delightful-two-knobs analogies-index