Per-lens directories for analysis summaries and tables

Posed by the operator on 2026-10-22, while retiring Agent Memory Systems from the site's main menu.

Goal

Give each analysis lens its own directory for cross-system summaries and tables. Each directory links into the per-system members already published under kb/agentic-system-analyses/retained/<system>/ (for the memory lens, memory.md and memory-profile.md); it does not copy, move or re-derive them. Readers then navigate by lens rather than through the frozen Agent Memory Systems collection, which stays buildable but leaves the main menu.

Operator decisions

  • Per-lens directories hold summaries and tables. Per-system analysis parts stay where publication puts them; the lens directories link to them. This is not a change to the analysis set layout, ARTIFACT.yaml or integrate.
  • Agent Memory Systems leaves the main menu but its collection stays. Its links keep resolving.
  • The menu change belongs here, because the memory lens directory is its replacement.

Starting evidence (2026-10-22)

  • The lenses are the analyst members of a set: runtime, memory, epistemic, workflow. Only memory has structured comparison fields (memory-profile.md, under the memory profile contract). The other lenses have prose reports only.
  • The memory lens already has table machinery in kb/agentic-systems/comparisons/: scripts/build_systems_matrix.py, render_systems_table.py and analyze_matrix.py read the retained sets directly. The landscape synthesis skill writes the cross-system summary. The comparison population there is currently empty until regeneration.
  • Coverage: 152 of the 159 historical Agent Memory Systems reviews have a retained analysis, matched by repository name; all 191 retained sets carry memory.md and memory-profile.md. Not covered: ace, sift-kg, clawvault, operational-ontology-framework, agentic-harness-engineering (unresolved in agentic-system-completion), Awesome-Agent-Memory (out of scope) and scroll; five of eight lightweight reviews also lack one.
  • The site menu is built by on_config in src/commonplace/docs/properdocs_hooks.py from every kb/<dir>/README.md. Its UNPUBLISHED_COLLECTIONS exclusion also suppresses generated directory listings, so hiding a collection from the menu alone needs a menu-only exclusion. kb/index.md line 34 also links Agent Memory Systems from Home.

Decisions (operator, 2026-10-22)

  • Placement: kb/agentic-systems/comparisons/<lens>/. Both collection contracts already route comparative work there: "Comparative writing belongs in kb/agentic-systems/" and "comparisons/ holds cross-system analyses and generated tables". Per-system members stay in kb/agentic-system-analyses/. The memory table scripts' outputs move from comparisons/ to comparisons/memory/.
  • Per-system lists are generated at build time by the docs hook, as current-analyses.md is, from current_analyses(root), linking each system's <lens>.md (and memory-profile.md for memory). Nothing is committed when an analysis lands. Each lens directory keeps an authored README.md explaining the lens and linking its list, tables and summaries.
  • Lenses without structured fields (runtime, epistemic, workflow) start with the lens README and the generated list only. A table waits for comparison fields that emerge from prose across runs; a summary waits for a commission.
  • Staleness by kind. Generated lists cannot go stale. Memory tables stay command-generated and committed; their rows already carry artifact hashes, and the lens page states the selected population and generation date or commit. A build-time population check is deferred until a stale table causes harm. Landscape syntheses are commit-bound snapshots; the lens README lists each with its date and system count.
  • Order, revised. The generated list is lens-generic, so all four lens directories landed together (2026-10-22): each has an authored README and a build-time systems.md; memory also links profiles and holds the relocated table outputs.

  • Menu: one "Agentic Systems" dropdown replaces the separate "Agentic System Analyses" and "Agentic Systems" entries: Analyses (the generated current-analyses list), one entry per lens, and Authored accounts (the kb/agentic-systems/ landing). A lens joins the dropdown only when its directory lands; memory comes first. "Agent Memory Systems" leaves the menu.

Open questions

  • Comparison fields for runtime, epistemic and workflow. Their members carry structured body content — shared records of six kinds, and the epistemic object inventory and authority-route ledger with controlled route-function and architectural-status values — but no frontmatter comparison block like memory-comparison. A table needs either an extractor over those records or a profile step per lens; either is a method change.

Boundaries

  • Do not edit retained analysis sets or the publication method; per-lens comparison fields for runtime, epistemic or workflow would be a separate method change.
  • A landscape synthesis run needs its own commission.
  • Historical Agent Memory Systems content stays as it is.

What closes this workshop

The memory lens directory exists and replaces Agent Memory Systems in the main menu; the placement and generation choices are recorded where the collection contracts or an ADR will find them; each other lens either has a directory or a recorded reason it waits. Then remove this directory and its Active Workshops entry.