Single canonical source for "where is Policy Window today?" — two sections: (1) Three Rs reliability dashboard (reproducible / replicable / robust / honest disclosure), and (2) 10 measurable KPIs with on-track / behind / at-risk / not-started status. Both computed deterministically from live catalog state — no surveys, no fitted models. Snapshot . The roadmap (quarterly milestones) lives separately at /wiki/roadmap.
Same analysis on the same data should produce the same result.
Catalog → HTML render: deterministic by construction
(This is software determinism, notresearch-claim reproducibility. See "What this is NOT" below.)
Articles render from typed catalog constants. Same catalog input → same article output, every time. No random sampling, no fitted model. Prose may be AI-assisted under charter §7.9 (named-editor reviewed, provenance-labelled, every claim primary-source cited).
What this is NOT. This is not a claim that the research papers cited inthe catalog are themselves reproducible — that's a separate measurement that requires human replication review (see /wiki/reproducibility-policy). We do not currently issue reproducibility verdicts on any specific paper. The Nature April 2026 social-science collection's “Reproducible” criterion measures the latter; this dashboard reports only the former.
Verify yourself: download /wiki/catalog/json or /wiki/catalog/csv — the entire catalog state, machine-readable, CORS-open for any secondary analysis.
2 · Replicable
An independent expert reading the same primary sources should reach the same coverage classification.
Editorial agreement rate (Q2 2026, 12-cell stratified sample):75% on type · 100% on confidence (calibration cells only). Full event record: /wiki/coverage-games/2026-q2.
Honest limitation:the Q2 event used 1 human (the founding editor) + 1 LLM classifier (Claude Sonnet 4.5). This is materially weaker than the protocol's target of 3-5 independent expert classifiers and is not a peer-review-grade reliability estimate. The event was a process shake-out establishing the methodology before editorial-board recruitment.
The protocol (docs/coverage-games-process.md) is rendered at /wiki/coverage-games. Q3 2026 event targets ≥50 cells with ≥3 named human classifiers (gated by board recruitment below).
Editorial board status: 1 founding editor in place; 5 subject-editor slots open for recruitment. See /wiki/editorial-board.
3 · Robust
Alternative analytical assumptions should not flip the conclusion. Per-cell confidence tier surfaces where a stricter rubric (e.g., operative-article-only) would plausibly produce a different label.
26%
of cells classified with explicit confidence tier (270/1035). Editorial backfill in progress — first quarterly Coverage Games will prioritise the remaining 765 cells.
134high confidence
93medium
43low (contested)
4 · Honest disclosure
Catalog freshness + the gap between “structurally committed” and “operationally implemented.”
Known not-yet-implemented commitments (per methodology page honest-disclosure pattern):
DOIs via Zenodo — current persistent identifier is the committed-stable wiki URL. See methodology §7.
Editorial-board subject-editor slots (5 of 6) — being recruited. See /wiki/editorial-board.
Recently shipped commitments:
?asOf=<ISO> URL pinning — articles now render from the historical ArticleRevision snapshot at-or-before the requested date. See methodology §6.
5 · Coverage of selected documents
171
governs
163
implicit
1
conflicts
190
recorded as not addressing the topic
510
no classification recorded
These counts describe selected document–topic pairings, not all applicable law. A missing record does not establish whether a document has been assessed. The other labels are the catalog’s recorded classification categories, not an assurance of current legal effect.
6 · Goals + measurable KPIs
What Policy Window is trying to achieve, how we'll know if we did it, and where we are today. Every goal has a measurable metric and a public status. Honest disclosure of behind-target state is the point: a goals page that always says "on track" is theatre.
0 on track · 5 behind target · 0 at risk · 5 not yet started
G1-COVERAGE
Catalog coverage (v1: frontier AI regulation)
Behind target
Cover every operative provision across ≥3 jurisdictions for the v1 vertical's foundation_models, catastrophic_risk, agentic_systems_governance, compute_reporting topics
How we'll measure
Coverage cells with confidence=high citing operative-text provisions
Where we are today
42 of 494 cells tagged with confidence (8.5%). v1-vertical topics have higher density but no per-cell threshold met.
12-month target
200 of 624 cells confidence-tagged (32%); 100% of v1-vertical topics have ≥3 high-confidence jurisdictions
36-month target
500 of 624 cells confidence-tagged (80%); full coverage matrix peer-reviewable
G2-EDITORIAL
Editorial board recruitment
Behind target
Constitute a working editorial board with disclosed subject expertise and COI
How we'll measure
Filled subject-editor slots on /wiki/editorial-board with COI + recusal docs
Where we are today
1 of 6 slots filled (founder). 5 slots placeholder; recruitment opened iter-311.
12-month target
≥3 subject editors, each with disclosed COI + Q3 Coverage Games participation
36-month target
6 of 6 slots filled; board acts as gate for high-stakes decisions (firehose contracts, reproducibility verdicts)
G3-REACH-RESEARCHERS
Reach: academic researchers
Not yet started
Become a regular citation in AI-governance research papers
Q3-Q4 events at 50+ cells with ≥3 human classifiers; documented disagreement matrix per event
36-month target
Quarterly events at 100+ cells; ≥85% agreement reproducible across rotating classifier sets
G10-FIELD-INFRASTRUCTURE
Field infrastructure contribution
Behind target
Contribute to the AI-governance evidence commons (open data, citable infrastructure)
How we'll measure
Public API uptake (catalog JSON downloads, RSS subscribers, embed widget impressions) + open-data citations
Where we are today
/wiki/catalog/json + /wiki/catalog/csv live (CORS-open). RSS at /wiki/changelog/feed. Embed widgets at /embed/wiki/[slug]. No usage analytics yet.
12-month target
Public usage dashboard live; ≥100 unique catalog-JSON consumers/month; ≥10 publications referencing the open data
36-month target
Catalog used as base data for ≥3 downstream research projects; cited as the comparative-evidence layer in ≥1 evaluation framework
Status reviewed quarterly. The next review is due Q3 2026 (September). Revised targets preserve the original commitment in the git history of this file: git log src/app/wiki/meta/page.tsx.