a11oy /Restraint /Governed + Measured Frugality Gate Console Agent Loop Agent Model Energy Restraint

a11oy Restraint Doctrine v11

Before the a11oy Code agent emits a diff, it descends a 6-rung ladder and stops at the first rung that holds — (1) YAGNI, (2) stdlib, (3) native platform feature, (4) already-installed dependency, (5) one line, then (6) minimum viable code. Deliberate simplifications are marked with restraint: ceiling comments that name the upgrade path. Every decision becomes a signed DSSE receipt and an advisory Λ trust score; the benchmark reproduces code-reduction / cost / speed numbers on our stack, labelled MEASURED only when actually run, else SAMPLE/ROADMAP.

Provenance — honest  The ladder and the lite / full / ultra intensity levels are adopted from the open-source Ponytail coding-agent skill (github.com/DietrichGebert/ponytail, MIT, © 2026 DietrichGebert) — adopted + governed, not invented here. Our additions are the signed receipts, the Λ score, the measured-on-our-stack benchmark, and the energy tie-in. Ponytail's published numbers (80–94% less code, 47–77% cheaper, 3–6× faster, median across Haiku/Sonnet/Opus) are cited as Ponytail's, never claimed as ours. Λ uniqueness is Conjecture 1 (OPEN), advisory floor < 1.0 — not a proven theorem.
Ladder rungs
6
stop at first that holds
Intensity levels
3
lite · full · ultra
Per-decision receipt
DSSE
in-image ECDSA-P256
Energy thesis
J/tok
less code · fewer joules

Evaluate a task POST /api/a11oy/v1/restraint/evaluate

Type what the code agent is about to build. Restraint runs the pre-write ladder, names the ceiling, scores Λ, estimates lines/tokens/joules saved, and signs the decision.

Ladder decision

Run the ladder to see which rung stopped it and why.

Signed receipt in-image ECDSA-P256 / DSSE

The governed differentiator: every restraint decision is a tamper-evident DSSE receipt. Verify in-browser against /cosign.pub.

status — run the ladder —

Energy tie-in

Less code → fewer tokens → fewer joules on our GPU. Estimates are MODELED; the joules label defers to the on-box NVML meter (reads sample unless a fresh exporter reading is present).

tokens saved

Benchmark — baseline vs a11oy-restraint GET /api/a11oy/v1/restraint/bench

Ported from Ponytail's promptfoo methodology (MIT): the same five everyday tasks, two arms, median reported. Numbers are SAMPLE/ROADMAP on this Space (no model run wired here) and flip to MEASURED only when a model run is supplied. We measure our own; we never reprint Ponytail's numbers as ours.

TaskRungBaseline LOCRestraint LOCLOC −%Cost-proxy −%Latency −%Label
Run benchmark to populate.
🌐Spaces