arXiv:2609.20050v3 Announce Type: replace-cross Abstract: Retrieval assembles repository context by ranking passages for relevance to the current query. A coding agent halfway through an issue has already read much of what such a ranker returns. Relevance is scored per passage, but sufficiency belongs to the set: independently scored passages can fill the budget with support for one requirement while another goes unmet. We formulate state-conditioned minimal sufficient evidence recovery: given a captured agent state, recover a compact evidence combination supplying what its next decision still lacks. SERBench measures this on 500 held-out states from 45 repositories, recording what the agent has seen, crediting only sets that satisfy every annotated evidence requirement of the current decision, and separating set recovery from candidate discovery. MSS-Complement treats acquisition as set construction, not ranking. Three semantic calls propose a jointly sufficient set, search for what it lacks, and return 4-8 intact source units within 6,144 tokens. One configuration, fixed on calibration data, recovers a complete set for 73.0% of those states at five items and 80.6% at eight, against 61.4% and 72.4% for Qwen3 embedding with reranking. A matched control ranking by similarity alone recovers fewer complete sets, placing the margin over it in the set-level policy, not the computation. The lead persists from frozen repository source with no gold-derived pool. On AMA-Bench it answers from a 76.2% smaller answer prompt, with accuracy 2.08 points above that benchmark's own memory agent. Removing one required group from a complete set costs repair-localization precision under two executors. Retrieval for agents is better posed as recovering what a decision lacks than re-ranking what an issue resembles.
The Missing Complement: State-Conditioned Minimal Sufficient Evidence for Coding Agents
A paper on arXiv (arXiv:2609.20050v3) introduces MSS-Complement, a state-conditioned retrieval method that treats evidence gathering for coding agents as set construction rather than passage ranking, returning 4-8 intact source units within 6,144 tokens. Evaluated on SERBench across 500 held-out agent states from 45 repositories, one configuration fixed on calibration data recovered a complete evidence set for 73.0% of states at five items and 80.6% at eight, versus 61.4% and 72.4% for Qwen3 embeddings with reranking. On AMA-Bench the method answered from a 76.2% smaller answer prompt with accuracy 2.08 points above that benchmark's own memory agent.
For coding agents mid-task, retrieve a jointly sufficient set of 4-8 intact source units conditioned on the captured agent state instead of re-ranking passages by query similarity; the source measured 73.0% complete-set recovery at five items versus 61.4% for reranked Qwen3 embed
more in playbook: Retrieval and RAGRun your AI side-project on zahid.host
EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.