From 4d46e9ba091dee156284abe009502381e1a29140 Mon Sep 17 00:00:00 2001 From: Thamilvendhan Munirathinam <37844963+mthamil107@users.noreply.github.com> Date: Thu, 13 Aug 2026 11:45:15 +0530 Subject: [PATCH] docs: remove Co-memorize attribution from kickoff notes The kickoff planning docs still equated Taheri's Governed Memory (arXiv:2603.17787) with a "Co-memorize HITL" pattern it does not contain (it governs autonomous agents automatically, with no human gate). Reframe as a contrast and drop the fabricated term, matching the paper/README/spec corrections. No Co-memorize reference now ties to Taheri anywhere in the repo. --- docs/kickoff/AMP-SPEC-v0.md | 2 +- docs/kickoff/FINDINGS-CONTEXT.md | 6 +++--- 2 files changed, 4 insertions(+), 4 deletions(-) diff --git a/docs/kickoff/AMP-SPEC-v0.md b/docs/kickoff/AMP-SPEC-v0.md index 82d7613..fc568d5 100644 --- a/docs/kickoff/AMP-SPEC-v0.md +++ b/docs/kickoff/AMP-SPEC-v0.md @@ -418,6 +418,6 @@ Memory research informing the design: - LoCoMo benchmark - BEAM benchmark - "Remember Me, Refine Me" — procedural memory -- "Governed Memory" arXiv 2603.17787 — Co-memorize HITL +- "Governed Memory" arXiv 2603.17787 — automated agent-memory governance (human-gated contrast) - Mem0 architecture writeup - Letta (MemGPT) hierarchical memory paper diff --git a/docs/kickoff/FINDINGS-CONTEXT.md b/docs/kickoff/FINDINGS-CONTEXT.md index 2d1fccd..474d199 100644 --- a/docs/kickoff/FINDINGS-CONTEXT.md +++ b/docs/kickoff/FINDINGS-CONTEXT.md @@ -84,7 +84,7 @@ The specific evidence from the architecture/memory research agent: > > Memory benchmarks (LongMemEval, LoCoMo, BEAM) are fragmented with no canonical leaderboard, suggesting the standardisation race is still open. -Module 14 Co-memorize HITL (diff-and-approve UI for memory writes) composes naturally — no product ships this today; "Governed Memory" arXiv 2603.17787 raises the problem; Oracle AI Agent Memory acknowledges it; nobody owns the diff-and-approve flow. +Module 14 human-in-the-loop diff-and-approve for memory writes composes naturally — no product ships this today; "Governed Memory" arXiv 2603.17787 governs autonomous-agent memory but automates it (no human gate); Oracle AI Agent Memory acknowledges it; nobody owns the diff-and-approve flow. --- @@ -109,7 +109,7 @@ Three independent signals converged: 1. **The architecture/memory agent flagged three specific virgin pieces** — FSM procedural memory, standalone STM↔LTM primitive, cross-vendor protocol. Each is verifiable: search for any of them on GitHub and you'll find research papers but no products. -2. **The eval/HITL agent independently flagged Co-memorize as the least-covered HITL mode** — "no product ships a human-review UI for agent long-term memory writes/edits/deletes." This composes directly with the memory protocol. +2. **The eval/HITL agent independently flagged human-review-of-writes as the least-covered HITL mode** — "no product ships a human-review UI for agent long-term memory writes/edits/deletes." This composes directly with the memory protocol. 3. **The benchmarks are fragmented** — LongMemEval, LoCoMo, BEAM, MemoryBench, all exist but none is canonical. Fragmentation across benchmarks is a leading indicator of an open standardization race. @@ -122,7 +122,7 @@ The compound product (protocol + governance UI) is also defensible against the o All cited research files live in the source repo at: - [`output/agentic-ai-modules-scan/research_architecture_memory.md`](../output/agentic-ai-modules-scan/research_architecture_memory.md) — full Module 5 + 8 evidence with URLs -- [`output/agentic-ai-modules-scan/research_eval_hitl.md`](../output/agentic-ai-modules-scan/research_eval_hitl.md) — Co-memorize gap evidence +- [`output/agentic-ai-modules-scan/research_eval_hitl.md`](../output/agentic-ai-modules-scan/research_eval_hitl.md) — diff-and-approve HITL gap evidence - [`output/agentic-ai-modules-scan/research_rai_governance.md`](../output/agentic-ai-modules-scan/research_rai_governance.md) — Module 9-13 saturation evidence - [`output/agentic-ai-modules-scan/research_vertical_agents.md`](../output/agentic-ai-modules-scan/research_vertical_agents.md) — Module 1, 3, 4, 6 saturation evidence - [`output/agentic-ai-modules-scan/BRIEF.md`](../output/agentic-ai-modules-scan/BRIEF.md) — full synthesis brief with kill scenarios