This site is a working record of the framework being tested. The framework makes specific claims; research agents run a daily discovery pass and a weekly curation pass to track evidence either supporting the framework or pushing against it. Each item lives on the relevant page with status and trajectory labels that update as the picture changes.
The pipeline uses practices the framework describes: delegation of tasks onto a kanban, with agents owning discovery and curation; a human applying discernment on what to publish; and root-cause learning loops improving the research pipeline over time.
Where new material comes from. Pre-curated channels sit alongside scheduled searches, and raw discovery feeds.
Agents read the sources, use SKILL files to filter and tag for orgchart.md relevance.
Agents evaluate the daily discoveries. Applies source-quality hierarchy, the exclusions, and the per-layer caps, making editorial recommendations, and updating claims and trajectories.
Read the weekly curation. Promote, drop, or set aside. Counter-evidence carries equal weight to confirmation.
Promoted items appear on the relevant page in three display tiers. Trajectory updates at quarterly review.
Every kept item carries three tags applied during the daily discovery pass.
Type is either illustration or evidence. Illustrations show the pattern in practice. Evidence tests the framework's predictive claim. The strong-form prediction (firms treating AI as a tooling-and-talent exercise plateau within 18 months; firms solving for capital-C Capability compound) is testable only against evidence.
Power takes one of four values. Strengthening corroborates a layer's claim with a new mechanism or fresh empirical data; the confirmation check is the discipline (would the item rate equally if it argued the opposite?). Weakening contradicts or scope-limits a claim, and sits inline with strengthening items on each layer page rather than in a footer. The site is structurally a falsification record. New is a first encounter with a pattern, or a live signal worth tracking. Holding confirms what the framework already says without adding mechanism, sharpening, or empirical anchor, and is used sparingly.
Trajectory (↑ rising, ↔ holding, ↓ falling) updates over time as the evidence base grows. Reassessed quarterly.
The daily pass casts a wide net. The weekly pass applies the bar.
Source quality runs in four tiers. Primary sources (studies, interviews, cases) from frontier firms carry the most weight. Established practitioner voices (articles, interviews) on platforms with editorial quality come next. Sharp academic papers with a specific empirical finding or formal result follow. Institutional research is useful when the analysis adds depth beyond the bare data point.
Four exclusions discipline the picks. Polemic: practitioner reputation doesn't clear the bar; mechanism or empirical anchor does. Single-incident: an isolated event is anecdote until causal evidence links it to a named claim and multiple corroborating instances exist. Tier-4 falsification is a flag, not a promotion: when a news data point pressures an existing on-site claim, the move is to mark the existing item contested and note the new data inline. Empty slots stay empty: a layer with no refutation-shaped weakening evidence shows the gap; sharpening dressed as a refutation makes it less visible without resolving it.
Caps: maximum six promotions per week across all layers, maximum two per layer, default zero. A high-bar empty week is a healthy week.
On the layer pages, items appear in three display tiers. The three newest get full descriptions. The next two get a single sentence. The rest are title-only hyperlinks. Older items carry less surface area but remain visible.