{ "schema_version": 1, "builder_plan": { "scenario": "A support or data assistant repeats work across related customer requests.", "change": "Try the paper's memory change in a replay of past tasks, with separate memory per customer.", "baseline": "Your current memory pipeline plus a no-memory control using the same agent.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Task success, stale memories, retrieval misses, response latency, and all memory maintenance costs.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2608.31057", "title": "Measure Before You Manage: Evaluating Agent Working Memory in Coding Agents", "summary": "The method replaces uniform, token-budget-only context management with semantically informed policies, namely object-aware compression and retrieval over typed memory objects (instructions, artifacts, tool outputs, agent state). It costs added pipeline complexity and management work, and equal nominal token budgets do not guarantee equal delivered context or serving cost, as the real-system replay exposes serving limits beyond budget arithmetic. Calibration gains observed on the studied trajectories may not transfer to held-out tasks, so a policy tuned offline can underperform or fail silently in production.", "rationale": "The paper offers an evaluation framework and cautionary evidence rather than a validated production mechanism, so the reader should pilot semantic memory policies on their own workloads instead of adopting them directly.", "technique": "Semantic working-memory management for coding agents", "family": "memoria_e_contexto", "recommendation": "testar", "published": "2026-08-31", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2608.31057", "independent_implementations": 0, "total_implementations": 1, "stars": 2, "citations": 0, "claims": [ { "claim_id": "claim-01", "statement": "The method replaces uniform, token-budget-only context management with semantically informed policies, namely object-aware compression and retrieval over typed memory objects (instructions, artifacts, tool outputs, agent state). It costs added pipeline complexity and management work, and equal nominal token budgets do not guarantee equal delivered context or serving cost, as the real-system replay exposes serving limits beyond budget arithmetic. Calibration gains observed on the studied trajectories may not transfer to held-out tasks, so a policy tuned offline can underperform or fail silently in production.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.31057", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "No quantified improvement is claimed; the paper reports that semantically distinct memory objects show different retention and compression behavior across 55 archived coding-agent trajectories, and that calibration gains may not transfer to held-out tasks.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2608.31057", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [ { "anchor": "S4.E2", "label": "(2)", "section": "4 Managing Heterogeneous Working Memory", "role": "baseline", "latex": "S_{o}^{\\mathrm{GA}}=0.5\\,\\widetilde{0.99^{\\,s-\\ell_{o}}}+3.0\\,\\widetilde{\\cos(e_{o},e_{q})}+2.0\\,\\widetilde{I_{o}},", "mathml": "SoGA=0.5​0.99s−ℓo~+3.0​cos⁡(eo,eq)~+2.0​Io~,", "context": "The later study adapts Generative Agents’ recency/relevance/importance components (Park et al., 2023), not its reflection, planning, or social environment. The score is", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "A3.EGx1", "label": "(C.1)", "section": "Appendix C Calibration-Informed Object-Aware Policy", "role": "proposed_method", "latex": "u_{o} =\\max\\!\\left(10^{-6},b_{t}e^{-d_{t}a_{o}}m_{o}[1+0.35\\min(k_{o},5)](1-p_{t}z_{o})q_{o}\\right),", "mathml": "uo=max⁡(10−6,bt​e−dt​ao​mo​[1+0.35​min⁡(ko,5)]​(1−pt​zo)​qo),", "context": "The policy is a hand-specified heuristic, not a learned estimator of future utility. For type t, subtype multiplier mo, recorded access age ao, access counter ko, and recorded stale indicator zo, the implementation uses", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" }, { "anchor": "A3.EGx1-2", "label": "(C.2)", "section": "Appendix C Calibration-Informed Object-Aware Policy", "role": "proposed_method", "latex": "S_{o} =u_{o}/\\max(s_{o}^{\\mathrm{rendered}},1)^{0.5}.", "mathml": "So=uo/max⁡(sorendered,1)0.5.", "context": "The policy is a hand-specified heuristic, not a learned estimator of future utility. For type t, subtype multiplier mo, recorded access age ao, access counter ko, and recorded stale indicator zo, the implementation uses", "headline": "", "explanation": "", "objective": "", "symbols": {}, "evidence": "", "evidence_url": "", "source_url": "" } ], "equations_status": "selected", "core_kind": "formula", "equations_fetched_at": "2026-09-23", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }