Addressed-Memory Writes Fail at Reasoning in Qwen3 and DeepSeek Beds
Atoms, Not Knowledge: What Production-Scale Addressed-Memory Writes Deposit
Memory write-injection looks like a cheap way to add knowledge to a deployed model, but these results show the added facts cannot be chained into multi-hop reasoning by the model itself — orchestration must live in the application layer. Builders wiring RAG-style memory into agents should chain written atoms explicitly in prompts or system logic, and benchmark vendors should adopt coverage, synergy and distortion reporting. The tested beds include Chinese frontier models Qwen3 and DeepSeek, making the findings broadly relevant across model families.
An arXiv paper (2610.25317) reports the first production-scale audit of hash-addressed conditional memory writes across 124M–7B beds, Qwen3.8-Flash-Next-FP8 (176.95B, 320,001,536 FP8 rows) and DeepSeek-V4.1-Flash. Write coverage drops from 98% at 50 facts to 12.7% at 1,000 facts at a fixed budget, and two-hop composition stays at the independent single-hop baseline (26% observed vs 25% expected). A 28-percentage-point gap between given and recalled intermediates locates the failure in retrieval-for-reasoning rather than gate access. The authors propose a layered evaluation protocol covering coverage, synergy ratio and distortion rate, and conclude that written memory atoms are retrievable and reversible but not composable without system-level orchestration.
Atoms, Not Knowledge: What Production-Scale Addressed-Memory Writes Deposit
Deterministic hash-addressed conditional memory permits local post-hoc knowledge injection. Extending arXiv:2610.21553's bounded-feasibility audit rejecting under-specified generality claims with repaired probes and training-dose/capacity axes, we give, to our knowledge, the first production-scale write-content verdict. We test 124M-7B self-trained/publicly pretrained beds, Qwen3.8-Flash-Next-FP8 (176.95B; 320,001,536 FP8 rows), and DeepSeek-V4.1-Flash. Write coverage requires a sufficiently trained backbone prior, full-width readout, trainable gates; undertrained-bed intervention confirms a threshold, not a slope. At fixed budget, coverage falls from 98% (50 facts) to 12.7% (1,000); table size is not binding. Two-hop composition does not exceed the independent single-hop baseline (26% observed vs 25% expected; 15% lexical-distortion floor). Synergy remains absent across dose/load in two independent bed families. A 28-percentage-point in-context gap (given vs recalled intermediates) localizes failure to retrieval-for-reasoning, not gate access. Writes are independently retrievable, precisely addressable, reversible atoms, not composable into reasoning. A layered protocol, the first, to our knowledge, to cover compositionality in behavioral memory evaluation, separates baseline, synergy and distortion floor. Any addressed/key-value memory must report coverage, synergy ratio and distortion rate before deployment claims. Composition must be system-orchestrated; multi-constraint queries should explicitly chain written atoms. In-context controls show given intermediate entities are exploited at twice the implicit rate. Production load fits exponential interference ($R^2=0.98$); its mechanism, between budget dilution and write interference, remains open. All numbers are recomputed from archived raw logs/JSON; a fresh 52-target production rerun gives 52/52, exactly matching archived baselines.
