Lifecycle Analysis

Summary

ARIS Research Wiki is valuable because it treats research memory as a lifecycle, not just a filing scheme. Commonplace has good artifact destinations but weak transitions. We know where snapshots, ingest reports, notes, reviews, and instructions belong, but the middle states are under-specified: candidate, tested, failed, stale, superseded, reactivated, and promoted.

That is the lifecycle weakness this workshop should address.

Lifecycle ARIS Enforces

ARIS Research Wiki defines this practical loop:

  1. init creates the wiki structure, index, gap map, query pack, graph, and mutation log.
  2. ingest creates paper pages, deduplicates arXiv IDs/slugs, rebuilds generated views, and appends log receipts.
  3. Agents add explicit edges in graph/edges.jsonl, making relationships queryable and auditable.
  4. query_pack.md is loaded before ideation, so future work starts from compressed project memory.
  5. New ideas are written as first-class artifacts, with failed ideas preserved as a high-value banlist.
  6. Experiments update ideas and claims through tested_by, supports, and invalidates relationships.
  7. lint checks orphan pages, stale claims, contradictions, missing connections, dead ideas, and sparse pages.
  8. Re-ideation is triggered by events such as enough new papers, enough failed or partial ideas, a new contradiction, or a new gap with no idea.

The important mechanism is not the exact schema. It is the closed loop from source intake to ideation to experiment to claim update to reactivation.

Implementation Reality

The ARIS helper code is strongest for paper ingest, deduplication, index rebuilds, dynamic orientation generation, stats, and mutation logging. The idea/experiment/claim lifecycle is specified mostly in skill prose. That matters for borrowing: we should not copy the repo expecting a complete lifecycle engine.

Use ARIS as a design pattern, not a dependency.

Commonplace Lifecycle Gap

Commonplace has stable endpoints:

  • kb/sources/ preserves and analyzes external material.
  • kb/notes/ holds durable transferable claims and theory.
  • kb/reference/ documents the shipped system.
  • kb/instructions/ holds procedures.
  • kb/work/ holds temporal working artifacts.

The missing part is explicit state movement inside kb/work/ and between kb/work/ and the library. Current status: seedling/current/outdated is too coarse for active investigations. It says how mature a library artifact is, but not what should happen next to a candidate idea, failed hypothesis, stale claim, unresolved contradiction, or experiment result.

The lifecycle-management workshop already names promotion, maturation, retirement, and partial extraction as open problems. ARIS adds a concrete pattern for the active middle: keep an investigation map that agents must consult and lint.

Commonplace-Shaped Lifecycle

For an ARIS-inspired workshop subsystem, use these states:

Entity Candidate states Promotion target
Source candidate seen -> snapshotted -> ingested -> filed/promoted/ignored kb/sources/ snapshot and ingest report
Idea proposed -> selected -> tested -> supported/invalidated/parked -> promoted/retired note, instruction change, ADR, or retained failure record until workshop closes
Experiment/probe planned -> running -> concluded -> interpreted review result, validation change, note evidence, instruction update
Claim reported -> under-test -> supported/invalidated/qualified -> promoted/retired kb/notes/ note or structured-claim
Gap open -> addressed-by idea -> resolved/superseded note gap section, index entry, or future workshop
Edge proposed -> active -> promoted/dropped markdown link in durable artifact if promoted

What Should Be Enforced

The enforcement should be light but real:

  • Every mutation appends to a workshop log.md.
  • Every active claim must have a status and either evidence, an experiment, or a stale marker.
  • Failed ideas stay visible until the workshop closes or they are explicitly retired.
  • Query packs have hard budgets, so active context does not grow without bound.
  • Lint reports missing edges, dead ideas, stale claims, unconnected source cards, and open gaps with no idea.
  • Promotion requires a target: note, structured claim, instruction, reference, ADR, source report, or explicit no-promote decision.

Main Lesson

ARIS does not make our sources system obsolete. It shows that source intake is only one phase of a larger research lifecycle. Commonplace should keep its source/library boundary, then add an optional lifecycle map inside workshops for investigations that need to remember what has been tried, what failed, and what should be activated next.