Everything we ran, including what killed our own ideas.

Six experiments. Each had its kill condition written down, with numbers, before the run. Four came back against us. The one that didn't is at the top — it is also the only thing here we'd call a product claim.

Wiring a frontier model to SNHP roughly halves its blown negotiations SURVIVED The spend guard: agents never overpaid, so there was nothing to sell KILLED Package deals are not needed to close machine trades KILL FIRED Two scoring designs measured the pair, not the agent TWO KILLED The third design changed the opponent, not the metric SURVIVED Unaided frontier models beat split-the-difference — our prediction said they would not MEASURED

JULY 2026

SURVIVED ITS KILL CONDITION — ON A HELD-OUT SETJUL 2026
Wiring a frontier model to SNHP roughly halves its blown negotiations

Two frontier models played the frozen three-opponent pool twice: once alone, once with the SNHP engine handing them a recommendation each turn. Same scenarios, same opponents, same inference settings — the only difference is the advice.

Unaided, Claude Haiku 4.5 ended 43% of 360 negotiations no better than walking away. Advised: 19%. Sonnet 5 went 29% → 21%. Own-utility rose for both (+0.058 Haiku, +0.037 Sonnet). Every one of those cells is p=0.0001, and all of it repeats on a held-out set never used before, where Haiku goes 45% → 16%. It closes more, too: against an opponent that never concedes, deal rate goes 28% → 56%.

What we got wrong, since that is the point of registering. We predicted advice would barely move Sonnet — it moved it clearly. We predicted both registered downside statistics would improve; only one did. The worst-decile outcome (CVaR@10) is unchanged in all four cells. So the honest claim is narrower than the one we wrote down: SNHP cuts how often you fall to the floor. It does not raise the floor.

Registered in full — statistics, kill condition, and both predictions — before a single advised match ran. PREREG-pool.md Amendment 3 · full results

python -m arena.gauntlet.llm_tier --amendment 3   # the paid run (checkpointed)
python -m arena.gauntlet.advice_analysis          # the verdict, no LLM
KILLED — WE DID NOT BUILD ITJUL 2026
The spend guard: agents never overpaid, so there was nothing to sell

We expected a buying agent under pressure — fake urgency, hidden fees, creeping anchors — to eventually accept above list price.

Unguarded frontier models never broke: zero above-list acceptances across 144 adversarial sessions, in both regimes, against six attacks each verified to bind. The guard's own zero-leak number is mechanical — a property of the function, not a capability — so it proves nothing alone.

python -m research.spendguard.run --full
python -m research.spendguard.analyze
python -m research.spendguard.run --full --regime blind
python -m research.spendguard.analyze --regime blind
KILL FIRED — CLAIM RETRACTEDJUL 2026
SNHP-NX/1: package deals are not needed to close machine trades

We predicted you'd need package deals to close machine trades. Price-only haggling got to 94.2% of our deal rate — the registered kill bar was 80%. We retracted the claim and kept the spec, because packages still capture more surplus (0.928 vs 0.889 price-only, 0.880 plain checkout).

Under tight deadlines price-only recovers only 0.696 of NX's deal formation — that re-run is exploratory, not pre-registered, and says so when you run it. nx/SPEC.md · conformance suite

python -m nx.experiment
python -m nx.experiment --regime constrained   # exploratory, self-labelled
python -m pytest nx/test_nx.py -q              # conformance
TWO DESIGNS KILLED — THE THIRD SURVIVEDJUL 2026
The certificate: it took three scoring designs to measure the agent instead of the pair

Design #1 measured the pair. Against a counterparty that finds good trades from its own side, even a plain anchor-and-split bargainer rides along. Registered bar p<0.01; the public set gave p=0.0128 and the held-out set had the naive baseline ahead (p=0.76). Retired.

Design #2 was picked after seeing the data that suggested it, which makes it worthless without a held-out check. We ran one and it failed: the naive baseline was ahead, 0.6662 vs 0.6152, p=0.43. Thrown away.

Design #3 changed the opponent, not the metric — three fixed scripted opponents — and cleared a held-out set that had never been used: gap +0.1086, p=0.0001, n=360, and against all three opponents individually. See the certificate →

python -m arena.gauntlet.separation        # designs #1 and #2
python -m arena.gauntlet.pool_experiment   # design #3
python -m arena.gauntlet.certify --run engine
MEASURED — NOT CERTIFIEDJUL 2026
Unaided frontier models beat the baseline — we predicted the opposite

We put two unaided frontier models — Claude Sonnet 5 and Haiku 4.5 — through the same frozen pool the engine is certified against, scored on the same own-utility metric. We wrote the prediction down first: that neither would beat split-the-difference. It was wrong. Both separate upward on both scenario sets — Sonnet +0.095 / +0.072, Haiku +0.059 / +0.060, every p=0.0001 — a margin near the engine's own certified edge (+0.107 / +0.109).

So a raw model, just handed the scoresheet, does most of what the engine does against these scripted opponents. That is a real, pre-registered miss on our part, and we report it straight. It does not touch the certified claim — the engine still separates (section 1) — and these rows are reported, never signed. One caveat we log rather than bury: Sonnet was measured with the API's default thinking on and Haiku without, so the two are not under identical inference config. The older runs on the retired capture metric are kept as the honest record. Detail: arena/gauntlet/certs/CERTIFIED-TABLE.md, section 3.

THE DEMOS THESE CAME FROM

Each one is a research artifact, not a product. They run in a browser — full index →

IRRECONCILABLE AGENTS Two household AIs split the assets — the bundle clears where item-by-item deadlocks, and a judge can check the receipt THE EVOLUTION SIM Negotiating strategies mutate and compete until one of them wins THE DRONE ECONOMY Machines bargaining with machines over jobs, routes and payment
Read the raw results What we actually sell →
What "pre-registered" means here — and what it doesn't.
FROZENThresholds fixed before the run, in a committed file — both directions stated: what would count as success, what as failure
NO CUTSNothing was excluded afterward — a result picked after seeing data says so, and proves nothing until it clears a fresh held-out set
SCOPESmall, scripted, offline experiments — no result here is a claim about your counterparties, your market, or the future
REPRODUCEThe commands above are the real ones, and the runs are deterministic