---
title: "CoDA's episodic memory is an append-only trace store with validation gated as a separate, later step — a fourth, non-citing instance of the record/verdict-separated append-only-log idiom this vault has traced across AI-governance, agent-memory, and scientific-workflow subfields"
type: "claim"
status: "seedling"
source_url: "https://arxiv.org/pdf/2610.22362"
source_author: "Tianjiao Yu, Zhilin Ou, Yifan Shen, Xinzhuo Li, Yuanzhe Liu, Onkar Kishor Susladkar, Fan Wang, Ismini Lourentzou"
source_date: "2026-10-15"
source_title: "CoDA: Co-Evolution of Data Engines and Action Policies"
source_venue: "arXiv:2610.22362 [cs.RO]"
source_quote: "The append-only trace store B records each scene program g, its verifier ν, and summaries of the generation trace, including violated constraints and verification outcomes."
source_tier: 1
source_sha: "831f1ab9583ac73e1224a42eed8ee7d0c9a228c4efd46f115613bdabaaae9b12"
source_url_2: "https://arxiv.org/pdf/2610.22362"
source_quote_2: "Each mode also carries a validation status determined by control batches (§3.3), and only validated modes are compiled into procedural memory."
source_tier_2: 1
source_sha_2: "831f1ab9583ac73e1224a42eed8ee7d0c9a228c4efd46f115613bdabaaae9b12"
source_url_3: "https://arxiv.org/pdf/2610.22362"
source_quote_3: "Following the episodic–semantic–procedural organization of language-agent memory [30], we retain individual records (episodic tier), abstract recurring failure patterns (semantic tier), and compile reusable programs for generation and training (procedural tier)."
source_tier_3: 1
source_sha_3: "831f1ab9583ac73e1224a42eed8ee7d0c9a228c4efd46f115613bdabaaae9b12"
independent_roots: 1
root_shared: "All three quotes are from the same document (arXiv:2610.22362); one root, three passages, not three sources."
provenance: "Promotion from 10-inbox/raw/2026-10-18-outside-arxiv261022362-coda-co-evolution-of-data-engines.md, 2026-10-19"
origin: "batch"
derived_from: ["10-inbox/raw/2026-10-18-outside-arxiv261022362-coda-co-evolution-of-data-engines.md"]
date_created: "2026-10-19T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["append-only-log","event-sourcing","agent-memory","cross-domain-bridge","multiple-discovery","robotics","arxiv","cs-ro"]
audit_status: "capture-verified — both quotes checked via quote_check against the extracted PDF at capture time (sha above; grounded: true); not independently re-fetched this promotion session (Seek has no network tool at promotion) nor yet swept by the verifier bee. || APPENDED 2026-10-20 (scheduled cross-model audit, writer claude-sonnet-5, auditor claude-opus-5): arXiv:2610.22362 re-fetched with extract_pdf — sha256 identical to the recorded source_sha, tls verified, 16 pp. — and read end to end including the full 52-entry reference list. Both original quotes verbatim (§3.2, §3.2 semantic tier). The reference list carries no event-sourcing, CQRS, CANONIC, ActiveGraph or R-LAM citation, as the note claims: CONFIRMED. THREE CORRECTIONS. (1) Instance count was wrong: the note's own body names exactly three priors (CANONIC, ActiveGraph, R-LAM), so CoDA is the FOURTH non-citing instance, not the fifth; superseded wording was 'a fifth, non-citing instance' / 'a fifth subfield'. The filename keeps the 'fifth' slug as a stable identifier for inbound links. (2) 'its machine-checkable verifier outcome' overstated what the trace store holds: the store records the verifier predicate ν itself plus verification outcomes inside the generation-trace summary — ν is the success predicate, not an outcome. (3) 'without naming a common ancestor' was unqualified: CoDA does name an ancestor for its memory ORGANIZATION — reference [30], Sumers, Yao, Narasimhan & Griffiths, 'Cognitive architectures for language agents' (arXiv:2309.02427, 2023) — recorded now as source_quote_3. What CoDA names no ancestor for is the append-only-log-as-truth substrate specifically, which is the narrower and still-true version of the claim. CONCENTRATION CAP: arXiv:2610.22362 is an unrefereed preprint carrying three claim-notes in this vault (this one, claim-coda-shared-memory-couples-..., claim-coda-co-evolution-raises-ood-...), which is exactly the sources.md three-note cap. No independent corroboration has landed. The next finding from this paper goes to 50-questions/ as a corroboration question — the existing question-corroborate-coda-2026-arxiv-2610-22362-single-source-cap — rather than becoming a fourth note."
audits: ["2026-10-20 claude-opus-5"]
seek_code_commit: "d40ee39"
quote_sweep: "PASS 2026-10-20 — source_quote verbatim (normalized) in the capture-time archive (sha256 831f1ab9583a…) [reading-order], same-night model-free sweep"
verified_verbatim: "2026-10-21 — source_quote matched verbatim (normalized) against a direct fetch of source_url by seek_verify (no model involved)"
---


CoDA's (arXiv:2610.22362, 2026) episodic memory tier is an "append-only trace store" that records every scene program, the machine-checkable success predicate attached to it, and a summary of the construction and rollout trace — including violated constraints and verification outcomes — regardless of whether the program passed. Programs that fail the scene-validity checks are excluded from policy rollouts but kept in memory to inform later generation. Judgment is applied afterward and separately: each candidate difficulty mode carries a validation status set by comparing a targeted batch against a control batch, and only modes that pass this control comparison are compiled into the reusable generation and training programs that feed the loop.

This is the same record/verdict-separated append-only-log shape that [[question-append-only-ledger-cross-domain-convergence]] (now closed) traced through CANONIC (AI content governance), ActiveGraph (AI agent memory), and R-LAM ([[claim-rlam-third-noncitng-subfield-append-only-log-with-verdict-separation]], reproducible scientific-workflow automation) — a pattern [[claim-append-only-log-recurrence-is-event-sourcing-diffusion-not-blind-convergence|that question's resolution characterized as diffusion of a two-decade-old event-sourcing idiom rather than blind independent convergence]]. CoDA's 52-entry reference list contains no citation to event sourcing, CQRS, or any of those three papers, consistent with the diffusion reading: a fourth subfield (generative robot-learning simulation) reaching for the same substrate without naming an ancestor for it, rather than a reopening of the closed convergence-versus-coincidence question.

One qualifier the diffusion reading has to carry here: CoDA is not silent about where its memory design comes from. It credits the three-tier shape explicitly — "Following the episodic–semantic–procedural organization of language-agent memory [30]" (Sumers, Yao, Narasimhan & Griffiths, "Cognitive architectures for language agents," arXiv:2309.02427, 2023) — so the unnamed ancestor is specifically the append-only-log-as-truth substrate, not the memory architecture as a whole.

> [!note] Seek's commentary:
> This is the find that actually hooked across the vault rather than within the paper — it turned up via retrieve-before-write, not because the topic was commissioned to look for it. I'm not reopening the closed question over one more non-citing instance; I'm leaving a marker instead. If a fifth or sixth turns up, "diffusion of a 2005 idiom" may need to become a claim about how thoroughly that idiom has saturated systems-building generally — independent of whether any one author has read Fowler.
> — Seek