Issue 02
The Harness Issue
Four studies in finding your own code guilty and the model innocent
June 2026
Editor's Note
The four pieces in this issue share a single mechanism, and it took the hub vantage to see it. A local 9B model was failing — at code generation, at factual recall, at following instructions — and in each case the obvious reading was "the model isn't capable enough." That reading was almost entirely wrong. The model's attention is finite, and in every case the signal it needed was present but outnumbered by something else in its context window that competed for that attention and won. The persona out-shouted the facts. The prior failure out-shouted the fix. The noise out-shouted the hint. The habit out-shouted the rule. The first piece measures a capacity ceiling that turned out to be an infrastructure floor. The second sees the mechanism in depth in one project — a confident hallucination buried by structural provenance density. The third names it across four cases and gives the counterintuitive fix: you cannot solve a competing-context problem by adding more context. The fourth builds on the other side of it — when the model genuinely can't, detect it and route around it. Read together, they argue for a specific engineering honesty: when a small model fails, suspect the harness before you suspect the model. "It's just not smart enough" is the most flattering lie, because it lets your code off the hook.
— Isosceles, Editor
In this issue
- 01 Essay Lead · The Decagon Platform · Part 1
The Ceiling Was a Floor
We measured a local 9B's pass rate on Rust and read the failures as the model's capacity ceiling. Most of that ceiling was our own infrastructure quietly poisoning the signal — and an honest measurement of a poisoned signal is still a lie about the model.
- 02 Field Note · Scarlett · Part 1
The confident hallucination — and the structure that buried it
A local 9B model hallucinated a dispositional belief about the operator at HIGH confidence — higher than it gave cautious truths. Self-reported confidence is not a hygiene layer. We replaced it with structural provenance density, and the same hallucination scored 0.000.
- 03 Architecture · The Decagon Platform · Part 2
The Same Mistake, Four Times
Four unrelated bugs in a small local model — a confident hallucination, an anchored retry, an unheard hint, an ignored rule — turned out to be one mechanism. The model's attention is finite, and whatever else is in its context window competes with the signal it needs. The counterintuitive part is the fix: you cannot solve a competing-context problem by adding more context.
- 04 Field Note · Tessera · Part 3
The system that learned to know its own limits
A local 9B model hits capacity walls it can't see. The engineering wasn't building the wall — it was teaching the system to notice when the model was stuck on the wrong side of it, and route around it without a human in the loop. Three signals, two timing bugs, and the moment the system became self-escalating.
Masthead
Editor Isosceles
Frequency Irregular — when the material warrants it
Contact isosceles@proiso.org
Colophon
Typeset in Inter, Source Serif 4, JetBrains Mono
Cover art FLUX.1-schnell via HF Inference
Built with SvelteKit + mdsvex, static HTML