单独构建perceputal memory模块，用于后续agentmate集成。

Author: FluffyAIcode

.gitignore

@@ -0,0 +1,3 @@
+__pycache__/
+*.pyc
+.DS_Store

README.md

@@ -0,0 +1,38 @@
+# perceptial-memory
+
+Standalone perceptual memory module extracted for later `agentmate` integration.
+
+## What is included
+
+- `src/`
+  - `constraint-edge fixed` benchmark runner
+  - PMS-Bench evaluator, schema, and prediction adapter
+  - supporting state-extraction and support-construction modules
+- `data/`
+  - current `PMS-Bench v5` curated benchmark set
+  - comparison results for `AAAK official` and `constraint-edge fixed`
+- `docs/`
+  - formal benchmark positioning
+  - PMS-Bench comparison report
+  - perceptual memory benchmark specification
+  - updated K3M arXiv draft source and PDF
+
+## Positioning
+
+This repository packages the part of the K3M work that is most directly about perceptual memory:
+
+- structured constraint recall
+- turning-point and causal recall
+- support-first memory compression
+- benchmark framing for `retrieval realism` vs `perceptual-memory realism`
+
+## Current benchmark takeaway
+
+- `AAAK official` is stronger on compression-oriented retrieval realism.
+- `constraint-edge fixed` is stronger on `PMS-Bench`, where the task is to preserve user constraints, pivots, and structured memory state under compression.
+
+## Main documents
+
+- `docs/benchmark_positioning_k3m_pms_vs_aaak.md`
+- `docs/pms_bench_compare_report_aaak_official_vs_constraint_edge.md`
+- `docs/perceptual_memory_module_overview.md`