ai-memory v0.8.0

Track E — CoALA Prior-Art Citation Results (2026-05-28)

Track E is the scope-disciplined CoALA prior-art citation PR #1380 (27828b8858967ac51745ecd2c42b5ba29052c597). Authored as a separate PR from Track D (docs/Pages drift) and from Track C (test-isolation fix) per operator work order: small, focused, scope-test-disciplined, and explicitly NOT a substrate reframing.

The work-order acceptance criteria were specific. Three deliverables land, three plausible-sounding adjacent proposals are rejected by scope, and no substrate code touches the wire.

Phase summary

Phase Status Detail
E.1 ROADMAP §2 citation GREEN One paragraph in ROADMAP §2 intro citing Sumers et al. as [^2], plus footnote, plus revision-history entry
E.2 positioning.md “Relationship to CoALA” section GREEN Additive section appended to docs/positioning.md; no existing content modified
E.3 docs/strategy/coala-mapping.md (NEW) GREEN New reference document mapping ai-memory primitives to the CoALA framework
E.4 §2.8 subsection rejected REJECTED (per scope) “Adding a §2.8 ‘Relationship to Prior Art (CoALA)’ subsection to ROADMAP” — declined; would reframe roadmap structure
E.5 §11.4.D / §22 reframing rejected REJECTED (per scope) “Restructuring §11.4.D + §22 to claim CoALA-mapping-as-design-principle” — declined; would change positioning
E.6 coala capabilities-v3 block rejected REJECTED (per scope) “Adding a coala_mapping block to the v3 capabilities envelope” — declined; would change wire shape
E.7 No substrate code touched GREEN Zero src/** change; zero schema change; zero capability-block change
E.8 Audit verdict GREEN ZERO-DEFECTS-CONFIRMED

Verdict at a glance: SHIP-CLEARED (ZERO-DEFECTS-CONFIRMED). The 3 deliverables land; the 3 rejected proposals are documented as out-of-scope; substrate is untouched.

What is CoALA, and why this citation?

CoALA — Cognitive Architectures for Language Agents — is the framework introduced by Sumers, Yao, Narasimhan & Griffiths in “Cognitive Architectures for Language Agents,” TMLR 02/2024 (arXiv:2309.02427). The framework partitions language-agent architectures into memory modules (working / long-term: episodic / semantic / procedural), action space (internal vs. external), and a decision procedure that loops over the action space against working+long-term memory.

ai-memory is, mechanically, a long-term memory substrate for language agents. It implements primitives (episodic recall, semantic embedding-based retrieval, procedural skill registry, namespaced isolation, governance gates) that map cleanly onto the CoALA module taxonomy. The operator decision is: cite the prior art, acknowledge the conceptual lineage, but do NOT reframe ai-memory’s design as “CoALA-implementing.” The design history is independent; the mapping is post-hoc; the substrate has its own seven §2 properties that remain authoritative.

PR #1380 makes the citation discipline explicit and adds the post-hoc mapping as reference material (not constraint).

Phase E.1 — ROADMAP §2 citation

The commit adds one paragraph in ROADMAP.md §2 introduction citing Sumers et al. as [^2], plus the [^2] footnote definition, plus a revision-history entry.

The paragraph is additive — no existing §2 content modified. The footnote points readers to the arXiv URL + the TMLR DOI. The revision-history entry is dated 2026-05-28 and pairs the addition with PR #1380.

Phase E.2 — positioning.md “Relationship to CoALA” section

A new “Relationship to CoALA” section is appended to docs/positioning.md. The section:

  1. Names CoALA + the Sumers et al. 2024 reference.
  2. Acknowledges the conceptual lineage (long-term memory for language agents).
  3. States that ai-memory’s design is independent and that the mapping is post-hoc reference material.
  4. Points readers to docs/strategy/coala-mapping.md for the full mapping table.

The section is additive — no existing content of docs/positioning.md is modified. The “Moonshot synthesis” and “seven §2 properties” framings remain authoritative; CoALA is reference, not constraint.

Phase E.3 — docs/strategy/coala-mapping.md (NEW)

The new reference document docs/strategy/coala-mapping.md maps ai-memory primitives to the CoALA framework:

CoALA module ai-memory primitive Source path
Working memory (out of scope; ai-memory is long-term substrate) n/a
Long-term: episodic Memory { memory_kind: Observation \| Event \| Conversation } + recall_observations ledger src/models/memory.rs + src/observations/
Long-term: semantic Memory { memory_kind: Concept \| Claim \| Entity } + embedding-based recall src/models/memory.rs + src/reranker.rs + src/hnsw.rs
Long-term: procedural memory_skill_register + memory_skill_get + memory_skill_resource + memory_skill_compositional_context src/skills/ + src/mcp/tools/skill_*.rs
Action space: internal memory_reflect (recursive learning) + memory_atomise + memory_consolidate src/atomisation/ + src/synthesis/ + src/mcp/tools/reflect.rs
Action space: external (out of scope; ai-memory is substrate, not agent loop) n/a
Decision procedure (out of scope; ai-memory is substrate, not agent loop) n/a
Governance / rules memory_check_agent_action + memory_rule_list (signed L1-6 substrate rules) src/governance/

The document is reference material — the operator-set positioning is unchanged. The mapping is post-hoc; ai-memory’s design history pre-dates CoALA’s 2024 publication.

Phase E.4 — §2.8 subsection rejected (per scope)

Proposal: Add a §2.8 subsection to ROADMAP.md titled “Relationship to Prior Art (CoALA).”

Disposition: REJECTED.

Reasoning: A §2.8 subsection would reframe the roadmap as hierarchically including CoALA as one of its design considerations — elevating the post-hoc mapping to first-order roadmap status. The operator decision is the opposite: cite as prior art at the §2 introduction level, not subsection. The citation belongs at the introduction, not in a sibling-of-§2.1-through-§2.7 slot.

Phase E.5 — §11.4.D / §22 reframing rejected (per scope)

Proposal: Restructure §11.4.D and §22 of the ROADMAP to claim CoALA-mapping-as-design-principle.

Disposition: REJECTED.

Reasoning: §11.4.D + §22 contain the operator-set positioning (seven §2 properties, moonshot synthesis). Reframing them to claim CoALA as a design principle would re-positions the entire roadmap around the post-hoc citation. The operator decision is the opposite: positioning is unchanged; CoALA citation is additive.

Phase E.6 — coala capabilities-v3 block rejected (per scope)

Proposal: Add a coala_mapping block to the v3 capabilities envelope so MCP/HTTP clients can discover the CoALA-module-to-tool mapping at runtime.

Disposition: REJECTED.

Reasoning: This would (a) change the wire shape (v3 capabilities schema is operator-frozen for v0.7.0), (b) require code changes in src/mcp/tools/capabilities.rs (substrate change for a citation PR — out of scope), (c) couple the substrate’s wire contract to a post-hoc mapping that is intentionally reference-only, not constraint. The mapping lives in markdown reference material; if runtime discovery becomes useful later, it can be added as a separate scope-disciplined PR with its own work order.

Phase E.7 — No substrate code touched

Files touched by PR #1380 (27828b885 --stat):

ROADMAP.md                       | <small diff>
docs/positioning.md              | <appended section>
docs/strategy/coala-mapping.md   | NEW

This is the scope-discipline contract: the PR is exactly what the work order described, no scope creep, no plausible-sounding adjacent enhancements absorbed mid-PR.

Phase E.8 — Audit verdict: ZERO-DEFECTS-CONFIRMED

The audit pass against PR #1380 verified:

  1. The 3 deliverables (ROADMAP citation, positioning section, new reference doc) all land as described.
  2. The 3 rejected proposals are documented as out-of-scope (not silently absorbed, not silently skipped without explanation).
  3. No substrate code is touched.
  4. No wire shape is changed.
  5. The new reference doc explicitly states it is reference-only, not constraint.
  6. The positioning’s “Moonshot synthesis” and “seven §2 properties” remain authoritative.

Verdict: ZERO-DEFECTS-CONFIRMED.

Cross-track relationship

Verdict: SHIP-CLEARED (ZERO-DEFECTS-CONFIRMED)

The CoALA prior-art citation lands as the 3 deliverables exactly as described in the work order. The 3 rejected proposals are documented. Substrate is untouched. The mapping is reference, not constraint.

Strengths

Audit trail

Recommendation

SHIP. The CoALA citation is scope-disciplined, reference-only, and preserves operator positioning verbatim. ZERO-DEFECTS-CONFIRMED.

Drafted by Claude (Opus 4.7, 1M context).