Re-derivability Decides What a Staged Agent Pipeline Recovers After an Upstream Fault
Re-derivability, not extra stages, decides how staged LLM agents recover after an upstream fault.
The study argues that re-derivability determines how much accuracy a staged language-model agent pipeline recovers after an upstream fault. One deterministic fault is injected into the first stage, and the original problem is re-exposed to zero through three downstream stages on 120 gsm_hard items at temperature zero. Accuracy under fault rises on all four open-weight backbones, from +0.233 to +0.392. On Qwen3-14B the first re-grounded stage buys +0.394 matched retention for 59.8 extra tokens, while later stages add nothing, and with no fault the pipeline loses to one direct call on three backbones.
- Re-derivability predicts recovery after one upstream pipeline fault
- Faulted accuracy rises +0.233 to +0.392 on four backbones
- First re-grounded Qwen3-14B stage adds +0.394 retention
- Later stages add no further matched retention
- Unfaulted pipeline loses to one direct call on three backbones
Full article310 words · extracted from arxiv.org · click to collapse
One variable sets what an upstream fault costs a staged pipeline of language-model agents: re-derivability, how much of what a stage needs it can rebuild from the original problem. Grounding an inspector agent in that problem is worth +0.608 [+0.517, +0.700] to +0.358 over a blind one on four open-weight backbones served with thinking disabled, and on the two Qwen backbones the blind inspector changes no item at all. That head-to-head is exploratory. One deterministic fault enters the first stage, and we re-expose the original problem to $k = 0,\dots,3$ of the downstream stages with agents, items, fault and topology held fixed, on 120 gsm_hard items per arm at temperature zero. Accuracy under fault rises on four of four backbones, from +0.233 to +0.392, the largest Holm-adjusted $p$ being $2.1\times10^{-6}$. A registered kill test rules out tokens. Blanking every word holds the word slots fixed, and retention tracks the visible fraction on four of four, climbing from 0.221 to 0.692 on the primary. Those two families are confirmatory and everything else here is exploratory. The interaction excludes zero on two of four backbones under the registered pipeline, four of four under a three-stage pipeline, and three of four under full message history, the primary at +0.317. On Llama-3.1-8B the fault carries no detectable cost at any dose, so the other three carry every claim about what a fault costs. Re-derivability also sets what the architecture costs, and no decomposition we measured reliably beats one direct call. With no fault injected the registered pipeline loses to that call by -0.267, -0.125 and -0.317, and on Phi-4 reads +0.058 at $p = 0.118$, which the test fails to separate from zero. The repair that works is cheap and front-loaded: the first re-grounded stage buys +0.394 of matched retention for +59.8 tokens per item on Qwen3-14B, and the stages after it buy nothing.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.32802