How to read this map

Solid line — explicit evidence. FrontierMemory's own October 2026 blog names or hyperlinks the task benchmarks, trajectory sources and contextual works. An explicit link does not mean code or methodology was inherited.

Dashed line — thematic lineage. Earlier Letta research is established by its research archive, but FrontierMemory's blog does not directly cite every earlier Letta post.

Dotted line — contrasting baseline. LoCoMo and LongMemEval are explicitly contrasted as recall benchmarks, not used as memory-dependent task sources.

FrontierMemory's published setup: 15 manually adapted tasks, with task/user/environment memory; Letta reports 219 relevant trajectories among 665 in its blog. The released v1.0 corpus used on the cluster has 674 trajectories. The sources do not reconcile the discrepancy. Gold memory is part of Letta's task-validation design, not independently confirmed by our 7B pilot.

Sources: Original research post · official code and task descriptions · Letta research archive. FrontierMemory v1.0 is currently cited as a blog/repository, not a verified peer-reviewed proceedings article. Editable Mermaid diagram + per-source evidence ledger ↗.

Harnesses evaluated, not ancestors: Letta Dream with Letta Code, Claude Managed Agents Dreaming with Claude Code, OpenAI Agents SDK dreaming with Codex, Prime Agent (refine), and Hermes Agent (learn). Letta’s comparisons change the native memory system and task harness together, so they are not causal estimates of a memory-generator component in isolation. These are omitted as citation ancestors in the graph deliberately.

Publication venues for a FrontierMemory study

Workshop route: For early replication and negative results, try LLM4Code @ ICSE; for verifier and benchmark validity, AGENTVERIFY @ ICSE. AGENT @ ICSE and AgenticSE @ AAAI are further options. ICSE workshop deadline: Nov 13, 2026; AgenticSE: Nov 20, 2026. AAMAS workshop paper CFPs are pending (recommended Jan 24, 2027); ICLR workshop decisions are Nov 29, followed by individual paper CFPs.

Mapped from our October 17 venue MD, plus MSR Data & Tool Showcase. Fit means relevance to a rigorous replication, audit, or task extension—not likelihood of acceptance. Last verified: October 25, 2026.

EASE 2027 ResearchStrong

Thorough empirical reproduction and construct validity

Jan 15 abstract / Jan 22 paper, 2027
MSR Registered ReportsConditional

Pre-registered experience-transfer study; journal stage later

Nov 13 abstract / Nov 20 report, 2026
AGENT @ ICSEConditional

Engineering and assessment of coding-agent systems

Nov 13, 2026
WMAC @ AAAIConditional

Non-archival; only with multi-agent collective-memory angle

Nov 20, 2026
AAMAS Blue SkyStretch

Research vision; prior OpenReview author registration Oct 20

Nov 5 abstract / Nov 12 paper, 2026
AAMAS WorkshopsConditional

Collective memory and agent evaluation; individual CFPs pending

Recommended Jan 24, 2027; workshop-specific
ICLR 2027 WorkshopsStretch

Continual-learning relevance; Feb 1 is guidance, not fixed CFP

CFPs pending; suggested Feb 1, 2027
IJCAI 2027Stretch

Mature general-AI result with broader validity

Jan 4 abstract / Jan 11 paper, 2027
AgenticSE @ AAAIConditional

Agentic coding agents, empirical systems and software verification

Nov 20, 2026 · confirmed by dedicated CFP
ICSE NIERClosed

Earlier submission deadline passed

Oct 23, 2026 (passed)

Conference calls are authoritative. AGENTVERIFY's dedicated CFP gives November 13 although an ICSE aggregate calendar has conflicting dates. AgenticSE's November 20 deadline is confirmed by its dedicated CFP. AAMAS recommends January 24, 2027, but individual workshops may differ. ICLR's February 1 is suggested timing, not an accepted workshop-specific deadline. Several tracks require original unpublished work and prohibit simultaneous submission.