Methods & Data
These analyses explain architectures at frozen source boundaries, not the latest release of every project. Start with a system's overview, then its boundary and synthesis. Follow the numbered records in its lens reports for the evidence behind a conclusion.
How the research is produced
The analysis method coordinates source inspection, separate runtime, memory, knowledge and workflow accounts, reconciliation and verification. The analysis-set contract explains the accepted members and manifest. The Systems Directory is generated from current accepted sets at each website build; their contents are not rewritten for the site.
Interpreting evidence
Documentation can claim a mechanism. Inspected code can show it is wired; an exposed capability may only be afforded to a caller. Observed execution, measured outcomes and evidence of causal contribution answer different questions. A route through code alone does not establish that a deployed agent used it or benefited from it.
Read the record contract for evidence classifications and the memory profile contract for scoped comparison findings and coverage. Uninspected operation is not a claim that a system has never been tested. Partial coverage is not absence. The retained verification reports expose disagreements and remaining limits.
Memory datasets and synthesis
- Comparison table — inspect mechanisms and follow rows to their accepted analyses.
- CSV download — machine-readable findings, evidence bases, scope, profile revisions and artifact hashes.
- Agent-memory landscape —
a frozen synthesis of 191 analyses at inputs commit
5f92f75e5, dated 2026-10-22. Its counts describe that selected corpus, not the field. - Current memory accounts and profiles — build-time listing; it can include analyses newer than the table or synthesis.
- Population selection, interpretation and regeneration — the single operational reference for these datasets.
- Landscape synthesis procedure — how a separately commissioned synthesis binds its population to a commit.
- Taxonomy maintenance — maintenance of shared comparison terms.
- Layout migration evidence — historical path/hash mapping and acceptance checks.
The table and CSV are committed snapshots, not live queries. Their dataset landing identifies their generation boundary. Profile revisions are not silently interchangeable. Neither the selected population nor its counts are a representative sample of agentic systems. There are no equivalent structured datasets for the other three topics yet.
Research scope
The primary research program charts the design space and seeks solutions useful to Commonplace; this public resource makes the resulting evidence reusable. Source-system findings stay separate from Commonplace adoption judgments, as the research design brief requires. The historical Agent Memory Systems collection remains available; its reviews and statistics are not the current population.