{"name":"Policy Window — comparative-quality study (vs Wikipedia + published reviews)","url":"https://policywindow.org/wiki/comparative-quality","disclosure":"Self-administered adversarial assessment with controls (refute-by-default judges, 3 per subject, live-fetched rivals). NOT an independent third-party study. A dated snapshot, not live-recomputed.","ranAt":"2026-06-22","method":"self-administered adversarial head-to-head with controls: per subject, 3 independent judge agents fetched the LIVE Policy Window article, the LIVE Wikipedia article, and (best-effort) a published review/survey, then scored each 0-5 on 6 dimensions under a refute-by-default mandate to hunt Policy Window weaknesses. Means across judges. Two scopes of comparison are reported and are NOT the same: the per-subject outcome label ('PW strongest' / 'PW not strongest') is computed against the strongest of ALL sources (Wikipedia AND the published review), whereas the headline 'X/10' tally and mean-vs-Wikipedia figure are PW-vs-Wikipedia only. In this snapshot the two coincide (every win also beats the review; every loss is to Wikipedia), but a future snapshot where PW beats Wikipedia yet trails the review could diverge. NOT an independent third-party study: the judges are Policy Window-run LLM agents, so treat as a self-assessment with adversarial controls, not external validation.","dimensions":["coverage_breadth","citation_density","source_quality","currency","neutrality","claim_accuracy"],"subjects":[{"key":"biometric-id","sources":{"policy_window":{"overall":4.5,"n":3,"dims":{"coverage_breadth":3.67,"citation_density":4.67,"source_quality":4.67,"currency":5,"neutrality":4.33,"claim_accuracy":4.67}},"wikipedia":{"overall":3.22,"n":3,"dims":{"coverage_breadth":4,"citation_density":3.33,"source_quality":2.67,"currency":2.33,"neutrality":3.67,"claim_accuracy":3.33}},"review":{"overall":3.72,"n":3,"dims":{"coverage_breadth":3.33,"citation_density":4.67,"source_quality":5,"currency":2.67,"neutrality":2.67,"claim_accuracy":4}}},"outcome":"pw_wins","pwWeaknesses":["Authority deficit it self-discloses: the article is 'AI-authored & AI-reviewed - no human review,' whereas Wikipedia has multi-editor scrutiny and the review is formally peer-reviewed. On a legal-doctrine subject this is a genuine credibility gap regardless of the article's accuracy.","Contains a self-acknowledged unverified claim: the Council of Europe Framework Convention ratification/entry-into-force status is stated as reached 'late 2025 with EU ratification in 2026' but flagged '(not yet primary-source confirmed)' -- a gap the primary-law-anchored review would not leave open.","Does not catalog the US city/municipal ban landscape (San Francisco is cited but the broader wave of US city bans that Wikipedia notes is not enumerated).","Does not surface public-acceptance / cross-national survey evidence (e.g. acceptance vs privacy vs discrimination tradeoffs) that appears in the broader literature; PW's evidence base is bias/accuracy + efficacy-of-safeguards only.","Editorial 'consensus settled' framing sits awkwardly next to the article's own acknowledgment that the underlying policy debate is NOT closed - a minor internal-consistency weakness Wikipedia's plain 'Controversies' framing avoids.","Fully AI-authored with NO human review (labelled '✦ AI-authored & AI-reviewed -- no human review (§7.10)'); the rival review (Qandeel 2024) is peer-reviewed and Wikipedia is multi-editor-reviewed, so PW's prose has the weakest external review/accountability of the three despite its transparency.","Less granular on operational FRT deployments and contracts than Wikipedia (e.g. Wikipedia gives Greece's 1,000-device EUR4M Intracom contract, Netherlands' 2.2M-image database, UK Met's 16,000-suspect watchlist/360+ arrests) -- PW abstracts to legal modalities and omits this deployment-level detail.","Mild reformist framing on a contested point: PW asserts retrospective/post-hoc identification 'warrants comparable constraint' to real-time RBI, taking a side in a live scholarly debate rather than presenting it as fully open."]},{"key":"bletchley-declaration","sources":{"policy_window":{"overall":4.39,"n":3,"dims":{"coverage_breadth":4,"citation_density":4.67,"source_quality":5,"currency":5,"neutrality":4,"claim_accuracy":3.67}},"wikipedia":{"overall":3.39,"n":3,"dims":{"coverage_breadth":3,"citation_density":3.33,"source_quality":2.67,"currency":3.67,"neutrality":3.67,"claim_accuracy":4}},"review":{"overall":3.75,"n":2,"dims":{"coverage_breadth":1.5,"citation_density":4,"source_quality":5,"currency":3.5,"neutrality":4.5,"claim_accuracy":4}}},"outcome":"pw_wins","pwWeaknesses":["Claim accuracy slip: states the New Delhi Declaration on AI Impact (Feb 2026) had '89 signatories', but reported figures were 88 (as of 21 Feb 2026) rising to 91 — '89' matches no source found and is asserted with no inline citation.","Does NOT quote or paraphrase the Declaration's own headline affirmation (AI should be 'safe, human-centric, trustworthy and responsible') that both Wikipedia and the Lancet editorial foreground — PW abstracts the text into analysis and loses the instrument's actual operative language.","Does not mention the concurrent US developments Wikipedia foregrounds — President Biden's Executive Order requiring AI developers to share safety results and the announcement of the US AI Safety Institute at NIST as summit-adjacent outcomes.","Editorial register is mildly evaluative/advocacy-tinged in places ('diffusion and dilution', voluntary model 'increasingly seen as insufficient', Bletchley 'underdelivers') where Wikipedia stays more strictly descriptive.","FABRICATED STRUCTURE: PW repeatedly cites the Declaration by section number ('Declaration §1-2', '§3-5', '§6', '§8-10') yet its OWN text says the declaration 'consists of flowing prose, with no numbered sections'. The cited section anchors do not exist in the source instrument — self-contradictory fabricated precision Wikipedia never indulges.","Heavy reliance on preprints (49 of ~100 cited sources are arXiv/institutional) for some load-bearing 'does governance work' claims, lowering average source authority relative to its peer-reviewed-heavy self-presentation.","Lacks Wikipedia's clean forward-looking summit calendar (the series table through AI Summit Geneva 2027); PW's trajectory stops at New Delhi 2026.","MISSES the formal summit-series continuation Wikipedia lists through 2027 (AI Summit, Geneva 2027); PW's trajectory stops at New Delhi 2026."]},{"key":"ca-sb-1047","sources":{"policy_window":{"overall":3.75,"n":3,"dims":{"coverage_breadth":2.83,"citation_density":3.67,"source_quality":4.17,"currency":4.83,"neutrality":3.67,"claim_accuracy":3.33}},"wikipedia":{"overall":3.97,"n":3,"dims":{"coverage_breadth":4.5,"citation_density":4.67,"source_quality":3.17,"currency":2.83,"neutrality":4,"claim_accuracy":4.67}},"review":{"overall":3.03,"n":3,"dims":{"coverage_breadth":2.33,"citation_density":2.67,"source_quality":3.5,"currency":1.83,"neutrality":4,"claim_accuracy":3.83}}},"outcome":"pw_loses","pwWeaknesses":["Citation mix is skewed toward AI-safety/x-risk academic literature (RAND bioweapon red-team, extinction-risk surveys, sleeper-agents) used for the 'does it work' evidence section; this under-weights the legislative/political record that is the core of this subject and reads as somewhat advocacy-adjacent relative to Wikipedia's news-and-record-heavy sourcing of the actual events.","Does not capture the public-opinion / polling dimension (AIPI, YouGov, Chamber of Commerce, David Binder) that Wikipedia presents with figures and bias caveats.","Drops the named supporter/opponent coalition almost entirely. Wikipedia names ~25 supporters (Bengio, Hinton, Musk, Hendrycks, Stuart Russell, 113 AI-company employees, 120+ Hollywood signers) and ~20 opponents (Andrew Ng, Fei-Fei Li, Yann LeCun, 9 U.S. Congress members incl. Pelosi, a16z, Meta, OpenAI, CCIA, TechNet); the CACM piece also names both sides. PW reduces the whole 'created its own weather system' fight to 'Anthropic & Meta objected on different grounds' — a serious neutrality/breadth gap.","Far thinner support/opposition coverage: PW names only 'Anthropic + Meta', whereas Wikipedia names Hinton, Bengio, Musk, SAG-AFTRA and 120+ entertainment figures on one side and Andrew Ng, Fei-Fei Li, Pelosi, Lofgren, Khanna, Andreessen Horowitz, Y Combinator on the other — PW loses the political/coalition story.","Imprecise on the critical-harm definition: PW states critical harm = 'mass casualties or economic damage exceeding $500M' generally. Per the statute, the $500M floor attaches specifically to the cyberattack-on-critical-infrastructure and autonomous-crime categories; the CBRN-weapon category is mass-casualties with no dollar threshold. Wikipedia states all four categories precisely; PW flattens them.","Internally inconsistent status framing: PW labels the bill 'Adopted 2024-09-29; not yet in force' while also marking it 'Vetoed' — a vetoed bill was never adopted/enacted, so the 'adopted...not yet in force' label is misleading.","Low visible citation density in the body: despite claiming '102 sources reviewed,' PW's prose attaches inline citations only to its scholarly-critique paragraph (Pistillo/Villalobos, Ho, Hooker, Ord, Qi, RAND); core factual claims (vote, veto date, thresholds, successor) are largely unsourced inline, whereas Wikipedia footnotes nearly every sentence.","Mischaracterizes the bill's central mechanism: PW repeatedly calls SB-1047 'the first U.S. state-level model-testing mandate' requiring 'pre-deployment third-party testing for frontier AI models.' The actual bill (per Wikipedia + leginfo text) required developers to adopt safety-and-security protocols and submit a SELF-certified compliance statement before deployment; THIRD-PARTY AUDITS were to begin in 2026 — there was no pre-deployment third-party testing requirement. PW's framing overstates and distorts the core obligation."]},{"key":"catastrophic-risk","sources":{"policy_window":{"overall":4.19,"n":3,"dims":{"coverage_breadth":2.5,"citation_density":4,"source_quality":4.83,"currency":5,"neutrality":4,"claim_accuracy":4.83}},"wikipedia":{"overall":4.33,"n":3,"dims":{"coverage_breadth":5,"citation_density":4.5,"source_quality":3.5,"currency":4.5,"neutrality":4.5,"claim_accuracy":4}},"review":{"overall":3.89,"n":3,"dims":{"coverage_breadth":1.67,"citation_density":4.83,"source_quality":4.83,"currency":2.67,"neutrality":4.5,"claim_accuracy":4.83}}},"outcome":"pw_loses","pwWeaknesses":["BREADTH GAP (largest): PW omits the conceptual/technical core that Wikipedia treats at length — intelligence explosion, recursive self-improvement, instrumental convergence, the orthogonality thesis, corrigibility, and the alignment problem generally. It reads as a regulation tracker, not an encyclopedia entry on the subject.","Citation volume far lower: PW has ~22 distinct citations vs Wikipedia's ~190+; on raw breadth-of-evidence PW cannot match the encyclopedia.","Conversely, on neutrality PW is solid but its framing is implicitly governance-advocacy ('gap signal' for 20 silent regimes, 'no impact evidence that governance reduces risk') — it adjudicates a policy gap rather than neutrally surveying the underlying scientific debate the way the review does.","Far less current on the most recent EMPIRICAL x-risk evidence than Wikipedia: PW's empirics stop at 2024 bio-uplift null results (RAND/OpenAI), whereas Wikipedia covers the Dec-2024 Apollo Research o1 deception findings (sandbagging/self-exfiltration in 0.3-10% of scenarios) and the Anthropic/Claude 'alignment faking' result (12%->78%) — directly relevant deceptive-alignment evidence PW name-checks as a risk category but never substantiates with these studies.","Heavier reliance on arXiv preprints for several load-bearing claims (Shevlane, Anderljung, Schuett, Grace, Hubinger, Scholefield) — defensible and well-chosen, but the Hadshar review and Wikipedia both anchor more on the established Bostrom/Russell/Ord canon PW barely cites.","Lighter on the deep evidentiary nuance the Hadshar review provides: PW asserts the bio-uplift premise is 'contested' but does not distinguish specification gaming vs. goal misgeneralization vs. power-seeking, or note (as the review does) that there are still no public empirical examples of misaligned power-seeking — the strongest available argument that current x-risk claims remain speculative.","Narrower neutrality framing: PW's contestation is mostly 'does current-model uplift exist / is governance enforceable,' which understates the LIVE superintelligence-feasibility debate (can AGI even reach extinction-capable levels, takeoff speed, alignment tractability) that Wikipedia presents as the central neutrality axis.","Narrower taxonomy of risk types: PW lists only CBRN uplift, autonomous replication, and deceptive alignment, omitting value lock-in, totalitarian lock-in, and astronomical-suffering (s-risks) that Wikipedia enumerates."]},{"key":"coe-ai-convention","sources":{"policy_window":{"overall":4.5,"n":3,"dims":{"coverage_breadth":4.67,"citation_density":5,"source_quality":4.67,"currency":4.67,"neutrality":4,"claim_accuracy":4}},"wikipedia":{"overall":3.17,"n":3,"dims":{"coverage_breadth":3.33,"citation_density":2.33,"source_quality":2.67,"currency":3.33,"neutrality":4,"claim_accuracy":3.33}},"review":{"overall":4.39,"n":3,"dims":{"coverage_breadth":3.33,"citation_density":5,"source_quality":5,"currency":3.33,"neutrality":4.67,"claim_accuracy":5}}},"outcome":"pw_wins","pwWeaknesses":["Currency/accuracy on the headline fact: PW HOLDS status as 'adopted, not in force (pending primary-source confirmation)' even though the Convention DID enter into force 1 November 2025 — confirmed by the CoE itself (coe.int) and secondary trackers (eucrim), and EU ratification (15 May 2026) is likewise confirmed. PW's self-imposed editorial caution makes its top-line status LESS accurate and less current than even Wikipedia's formula and the live primary source.","Doctrinal depth gap vs the review: PW does not analyze the drafting-process dilution of concrete rights (Stoyanova shows early drafts contained access-to-records, human-review and AI-interaction-transparency rights that were stripped from the final text). PW asserts the weak private-sector opt-in but does not trace how the obligations were watered down across Zero Drafts.","Drafting-history depth: Wikipedia provides the CAHAI->CAI institutional drafting timeline (2019 CAHAI, 2020 initiation, 2022 CAI takeover) as discrete dated milestones; PW's narrative folds negotiation history into critique ('Zero Drafts', private-actor compromise) but is less crisp on the formal committee chronology a reader wants for orientation.","Entry-into-force accuracy: PW surfaces a 'reported 1 November 2025' entry-into-force date in the body that the strongest authoritative evidence contradicts — CoE/CADE sources indicate the Convention was still NOT in force as of June 2026 with ratifications still being encouraged. Wikipedia avoids this error entirely by simply not asserting a ratification count or in-force date. PW does hedge the claim ('not confirmed against primary source') and holds status as 'adopted, not in force,' but it still introduces a likely-wrong date that Wikipedia does not.","Explicit signatory roster: Wikipedia lists named signatories with individual signature DATES (Canada 11 Feb 2025, Liechtenstein 27 Feb 2025, Switzerland 26 Mar 2025, Ukraine 15 May 2025, Uruguay 17 Sep 2025); PW names major signatories (US, UK, EU, Israel) but does not give the dated per-country signature ledger, so on the narrow 'who signed and when' fact PW is less granular than Wikipedia.","Legal-doctrinal precision on private-actor mechanism: The peer-reviewed Stoyanova review articulates the private-sector point with more doctrinal exactness (indirect regulation via state-attribution/delegation/control, obligation-of-result grounded in the Explanatory Report) than PW's higher-level 'Art. 3(1)(b) opt-in' phrasing — i.e., on this specific legal mechanism the review is more rigorous than PW.","Narrative breadth gap vs Wikipedia: PW's matrix idiom under-covers the negotiation/drafting history (CAHAI 2019 -> CAI 2022 handover) and does not present a clean, dated signatory roster the way Wikipedia does (Canada/Liechtenstein Feb-2025, Switzerland Mar, Ukraine May, Uruguay Sep). A reader wanting 'who signed and when' is better served by Wikipedia.","Neutrality register: PW's framing is more evaluative/skeptical (critique sections, the 'two silences' effectiveness narrative) than the strictly descriptive encyclopedic tone of Wikipedia and the measured legal-scholarly tone of the Stoyanova review — a reader wanting a neutral statement of what the treaty is, before contestation, gets that more directly from the rivals."]},{"key":"deepfakes","sources":{"policy_window":{"overall":4.06,"n":3,"dims":{"coverage_breadth":3,"citation_density":4.33,"source_quality":5,"currency":4,"neutrality":4.33,"claim_accuracy":3.67}},"wikipedia":{"overall":4.11,"n":3,"dims":{"coverage_breadth":5,"citation_density":4.67,"source_quality":3,"currency":4,"neutrality":4,"claim_accuracy":4}},"review":{"overall":4.06,"n":3,"dims":{"coverage_breadth":3,"citation_density":4.67,"source_quality":4.67,"currency":3.33,"neutrality":4.33,"claim_accuracy":4.33}}},"outcome":"pw_loses","pwWeaknesses":["CONFIRMED ERROR/STALENESS: PW lists US Executive Order 14110 as 'governs/high (§4.5 content authentication, watermarking)' and never acknowledges it was RESCINDED by President Trump on 2025-01-20. Wikipedia maintains a current, dedicated EO 14110 article documenting the revocation. PW's stale status is especially damning because its own 'last verified' date is 2026-05-24 — 16 months after the rescission — and it does list the replacement EO 14179 (as merely 'silent') without ever stating the rescission relationship.","Citation-verifiability risk: PW leans on bleeding-edge preprints (e.g. Cuevas & Horta Ribeiro 2025 arXiv:2602.02754; Meding & Sorge 2024 arXiv:2412.09961) at least one of which could not be independently confirmed, whereas Wikipedia and the review cite established, checkable sources.","Detection technology is treated only in passing (a single fault-line sentence citing Harris 2024 on detector limits), whereas Wikipedia devotes a whole subsection to audio/video forensics, blinking artifacts, biometric and blockchain verification.","Far fewer citations in absolute terms: ~16 sources vs Wikipedia's ~240+ and the MethodsX review's ~140+. PW's are higher-quality (peer-reviewed/primary-law), but the sheer breadth of claim-by-claim backing is lower, and several whole subtopics (techniques, history) carry essentially no citations because they are barely covered.","Far narrower breadth than Wikipedia: PW gives little-to-no coverage of deepfake HISTORY/origins (Wikipedia traces 1997 Video Rewrite onward), the underlying TECHNIQUES (GANs, autoencoders, latent space — Wikipedia explains these), or benign APPLICATIONS (film VFX, art, satire). PW is a law map, not an encyclopedia of the phenomenon.","Far narrower subject coverage than Wikipedia: PW has no real History section (no Video Rewrite 1997, no GAN/autoencoder/diffusion technical explanation, no amateur/commercial development arc) — it is a governance evidence-map, so a reader wanting to understand what deepfakes ARE and how they are made gets much less than from Wikipedia.","Far narrower than Wikipedia: PW omits the entire technical/creation layer (autoencoders, GANs, face-swap, audio cloning, full-body synthesis) that Wikipedia covers in a dedicated Techniques section.","Less self-skeptical on its core framing than the Birrer & Just review: PW asserts harms/fault-lines, but the review's empirically-grounded thesis that deepfake concerns are frequently 'alarmism rather than well-founded evidence' is a meta-level critique PW does not foreground with equal rigor."]},{"key":"eu-ai-act","sources":{"policy_window":{"overall":4.78,"n":3,"dims":{"coverage_breadth":4.67,"citation_density":5,"source_quality":5,"currency":5,"neutrality":4.33,"claim_accuracy":4.67}},"wikipedia":{"overall":3.78,"n":3,"dims":{"coverage_breadth":4.33,"citation_density":3.67,"source_quality":3.67,"currency":2.33,"neutrality":4,"claim_accuracy":4.67}},"review":{"overall":3.44,"n":3,"dims":{"coverage_breadth":2,"citation_density":3.33,"source_quality":4,"currency":3.33,"neutrality":4,"claim_accuracy":4}}},"outcome":"pw_wins","pwWeaknesses":["Conflates 'adoption' with Official Journal publication: PW's At-a-Glance lists adoption date as 2024-07-12, but that is the OJ publication date. Wikipedia correctly separates the legislative milestones (Parliament vote 13 Mar 2024, Council approval 21 May 2024, OJ publication 12 Jul 2024, entry into force 1 Aug 2024). PW's single 'adopted 2024-07-12' is imprecise.","Datable error: PW states the provisional inter-institutional Digital-Omnibus agreement was 'reached on 7 May 2026'; independent legal trackers (Gibson Dunn, Pinsent Masons, DLA Piper, Travers Smith) record the political agreement on 6 May 2026 with Council/COREPER confirmation on 13 May 2026. Off by one day and conflates the agreement date.","Does not enumerate the full governance architecture as cleanly as Wikipedia: PW names the AI Office and market-surveillance authorities but does not lay out the European Artificial Intelligence Board, Advisory Forum, and Scientific Panel of Independent Experts as a structured set.","Does not foreground the Act's exemptions (military, national security, research, non-professional use) that Wikipedia states plainly — a scope-defining feature.","Does not narrate the legislative-procedure history as Wikipedia does: PW gives entry-into-force and application dates but omits the proposal date (21 Apr 2021), Parliament vote tally (523-46-49, 13 Mar 2024), Council approval (21 May 2024) and Official Journal publication (12 Jul 2024).","Less explicit on named civil-society opposition at the political-economy level: Wikipedia's 'Reactions' section names Amnesty International, La Quadrature du Net, EDRi and the 38-body creators' coalition with quoted critiques of loopholes/IP/transparency; PW's contestation is overwhelmingly academic-doctrinal and under-represents organized civil-society advocacy as a distinct voice.","Neutrality register: PW uses house editorial framing ('the second silence', confidence labels) that, while transparent, reads as more authored/voiced than Wikipedia's plainer encyclopedic tone — a marginal neutrality cost on an otherwise balanced article.","Omits legislative-procedure history that Wikipedia covers: no Commission proposal date (21 Apr 2021), no 2020 White Paper origin, no Parliament vote tally (523 for / 46 against / 49 abstaining), no Council-approval step. PW jumps straight to the in-force regime, so a reader gets no sense of how contested/how the text was negotiated."]},{"key":"foundation-models","sources":{"policy_window":{"overall":4.5,"n":3,"dims":{"coverage_breadth":3.67,"citation_density":5,"source_quality":4.33,"currency":5,"neutrality":4,"claim_accuracy":5}},"wikipedia":{"overall":4,"n":3,"dims":{"coverage_breadth":4.67,"citation_density":4,"source_quality":3.33,"currency":4,"neutrality":4,"claim_accuracy":4}},"review":{"overall":3.89,"n":3,"dims":{"coverage_breadth":2.33,"citation_density":4.67,"source_quality":4.33,"currency":3,"neutrality":4.33,"claim_accuracy":4.67}}},"outcome":"pw_wins","pwWeaknesses":["Does not develop the contestation it labels. PW names four 'fault lines' and links debate stubs, but the peer-reviewed Carey review actually argues the case (systemic risk as institutionalized uncertainty, private-actor delegation via the Code of Practice, the model-vs-system oversight blind spot). PW's treatment is a well-cited map, not the deep doctrinal analysis the review provides.","Less accessible framing of definitions than Wikipedia: Wikipedia juxtaposes the competing legal/institutional definitions (Stanford CRFM, EO 14110 'tens of billions of params', AI Foundation Model Transparency Act '1B+ params', EU Parliament, UK CMA) side by side, surfacing the parameter-threshold contrast explicitly; PW folds definition into the GPAI-threshold debate and does not lay out the parameter-count definitional variants as crisply.","Minor accuracy drift: PW dates the Digital Omnibus trilogue agreement to '7 May 2026', whereas reporting records a 6 May 2026 provisional agreement (Council-confirmed 13 May). Immaterial but a checkable one-day slip on a flagship date.","Minor currency nit on a fast-moving item: PW dates the EU Digital Omnibus provisional agreement to '7 May 2026' where most trackers report the provisional political agreement on 6 May 2026 (confirmed by Council 13 May); the substantive claims (Annex-III deferral to 2 Dec 2027, GPAI obligations kept) are accurate, but the headline date is slightly off the modal source.","Minor precision nuance: PW frames the EU 10^25 FLOP systemic-risk trigger as the Act 'presumes' systemic risk via Article 51, without foregrounding that it is a REBUTTABLE presumption (burden-shift) — substantively correct but slightly understated versus the statute's mechanism.","Misses regulatory definitions Wikipedia includes despite PW being the governance specialist: the US AI Foundation Model Transparency Act / Beyer-Eshoo parameter-based threshold (1,000,000,000 parameters) and the prior EO 14110 'tens of billions of parameters' definition are not engaged as definitional approaches.","Narrower named-model and ecosystem coverage: PW names a few developers (OpenAI, Anthropic, Google DeepMind, Meta/Llama, ByteDance/Doubao) but does not characterize specific production models the way Wikipedia does across its History/Examples content.","No history/timeline of the subject — Wikipedia chronicles GPT, BERT, the 2022 ChatGPT/Stable Diffusion breakthrough, and 2023 open models (LLaMA, Mistral); PW has no model chronology at all."]},{"key":"gdpr","sources":{"policy_window":{"overall":4.56,"n":3,"dims":{"coverage_breadth":3.5,"citation_density":4.67,"source_quality":4.83,"currency":5,"neutrality":4.5,"claim_accuracy":4.83}},"wikipedia":{"overall":3.94,"n":3,"dims":{"coverage_breadth":5,"citation_density":4.5,"source_quality":3.5,"currency":3.17,"neutrality":4,"claim_accuracy":3.5}},"review":{"overall":3.33,"n":3,"dims":{"coverage_breadth":2,"citation_density":3.67,"source_quality":3.83,"currency":2.67,"neutrality":4,"claim_accuracy":3.83}}},"outcome":"pw_wins","pwWeaknesses":["Breadth-of-aspects ceiling: PW intentionally covers only the AI-relevant slice (it flags '20 of 24 topics silent'), so for non-AI GDPR questions (e.g. household exemption, national-security carve-outs, processor contracts Art.28) it is silent where Wikipedia is not.","Breadth: PW openly covers only 4 'tracked topics' (biometric ID, transparency, redress, training-data rights) and marks 20 topics 'silent/pending', so it omits or barely touches core GDPR machinery Wikipedia covers in full — DPO appointment (Art.37), records of processing (Art.30), 72-hour breach notification (Art.33) as a general duty, data-protection-by-design/default (Art.25), and GDPR certification/seals (Art.42).","By-design AI-governance scoping means 20 of 24 tracked topics are flagged 'silent' for GDPR, and general-public GDPR concerns (cookie consent / dark patterns, consent withdrawal mechanics, adequacy decisions and SCCs as a transfer regime, blockchain incompatibility) are largely outside scope — Wikipedia treats all of these.","Consent: Wikipedia states the consent standard for a general reader ('specific, freely given, plainly worded, unambiguous affirmation'); PW references Art.6(1)(a) consent only as a lawful basis and does not lay out the Art.4(11)/Art.7 consent conditions.","Data-subject rights are enforcement/redress-skewed: PW foregrounds Arts 77/79/80/82/83 (complaint, judicial remedy, NGO representation, compensation, fines) but does not give the substantive everyday rights their own treatment — right of access (Art.15), rectification, right to erasure / right to be forgotten (Art.17), and data portability (Art.20) — which Wikipedia lists explicitly as named rights.","Evaluative rather than purely descriptive framing on effectiveness ('governance works? thin/absent', 'the second silence'). While this is a deliberate analytical strength, it is less encyclopedically neutral than Wikipedia's descriptive register on the same provisions.","General-GDPR mechanics are thin vs Wikipedia: PW lists data-subject rights by article number but does NOT explain the operational rights/obligations Wikipedia details — 72-hour breach notification (Art.33), DPO requirement (Art.37), data portability (Art.20), right-of-access mechanics (Art.15), data-protection-by-design (Art.25), records of processing (Art.30), and the certification scheme (Arts.42-43). A reader wanting a working GDPR reference is better served by Wikipedia.","Less detail on UK post-Brexit status: PW notes the Brussels Effect but does not give the UK-GDPR adequacy renewal specifics (extended to 27-Dec-2031) that Wikipedia carries."]},{"key":"nist-ai-rmf","sources":{"policy_window":{"overall":4.5,"n":3,"dims":{"coverage_breadth":4.33,"citation_density":5,"source_quality":4.33,"currency":5,"neutrality":4,"claim_accuracy":4.33}},"wikipedia":{"overall":2.56,"n":3,"dims":{"coverage_breadth":1,"citation_density":2,"source_quality":3.33,"currency":1.67,"neutrality":3.67,"claim_accuracy":3.67}},"review":{"overall":3.28,"n":3,"dims":{"coverage_breadth":2.33,"citation_density":3.33,"source_quality":3.67,"currency":2.67,"neutrality":4,"claim_accuracy":3.67}}},"outcome":"pw_wins","pwWeaknesses":["AI-drafted (labeled ✦ per its own charter §7.9), whereas the arXiv survey is a human-authored peer-review-track publication and Wikipedia is human-edited with community vetting — so on provenance/independent-vetting pedigree of authorship PW is weaker than both, despite its denser citations.","Descriptive/onboarding balance is thinner than the alternatives: the FLAIRS review cleanly enumerates NIST's seven trustworthiness characteristics as a usable taxonomy, and even Wikipedia's one sentence gives a crisp neutral definition; PW front-loads effectiveness-skeptic critique (voluntary=self-set standards, deepfake inefficacy, no impact evaluation) so a newcomer gets less neutral 'what it is / how to use it' grounding — a mild neutrality cost relative to the review.","Does not clearly flag the framework's live status/trajectory in a few respects the field was tracking by 2026 (e.g. the NIST critical-infrastructure profile concept note, AI RMF evolution) — its currency is otherwise excellent on the surrounding legal landscape, but the framework's own forward roadmap is underplayed.","Heavier editorial thesis than the neutral rivals: the recurring 'absence of evidence'/'the second silence'/'Key Fault Lines' framing is more argumentative than Wikipedia's flat one-sentence statement or the survey's descriptive subsection — i.e. on pure neutrality of voice PW is edged out by both rivals even though it is far richer.","Its headline coverage metric — 'governs 2 of 24 tracked topics, 16 silent' — is a Policy-Window editorial construct, not a property of the RMF, and risks understating/mischaracterizing the framework's actual breadth to a casual reader; neither Wikipedia nor the survey imposes such a frame, so on this point PW is arguably LESS neutral/representational than the rivals.","Omits ISO/IEC 23894 (the international AI risk-management standard the RMF is frequently mapped to/aligned with) — relevant interoperability context for a framework PW explicitly frames as 'an interoperability layer beneath harder law,' yet the most obvious standards counterpart is absent.","Omits the AI RMF Playbook entirely — a core, official companion artifact NIST publishes alongside the framework with actionable suggestions per subcategory; neither PW nor Wikipedia covers it, but PW positions itself as the comprehensive reference and still misses it.","Omits the framework's documented development provenance — the consensus, multistakeholder process (RFI, multiple public drafts, workshops, 18 months, 240+ contributing organizations) — which is well-sourced from NIST and lends legitimacy/context; this is institutional history PW's deep-dive format would normally include."]}],"summary":{"subjects":10,"pwWins":7,"ties":0,"pwLoses":3,"pwMeanOverall":4.37,"wikipediaMeanOverall":3.65},"followUpRerun":{"ranAt":"2026-06-22","phase":"post-fix re-run (iter-457) of the 3 baseline losses","method":"Same self-administered adversarial methodology as the baseline: 3 refute-by-default judge agents per subject fetched the LIVE post-fix Policy Window article and the LIVE Wikipedia article, scored each 0-5 on the 6 dimensions, and verified (verbatim) whether each recorded baseline weakness is now resolved. Not an independent study; absolute scores shift across judge panels, so the durable signals are the weakness resolution and the within-run direction.","subjects":[{"key":"deepfakes","priorPw":4.06,"priorWiki":4.11,"priorOutcome":"pw_loses","newPw":4.33,"newWiki":4.28,"newOutcome":"pw_wins","pwDims":{"coverage_breadth":3,"citation_density":3.67,"source_quality":5,"currency":5,"neutrality":4.67,"claim_accuracy":4.67},"wikiDims":{"coverage_breadth":5,"citation_density":5,"source_quality":3.33,"currency":4.67,"neutrality":3.67,"claim_accuracy":4},"weaknessesResolved":2,"weaknessesPartial":1,"weaknessCount":8,"note":"FLIPPED to a PW win. The confirmed EO-14110 staleness error is fixed (the cell + prose now state the 20 Jan 2025 rescission by EO 14148, EO 14179 silent) and the alarmism/evidence meta-critique is foregrounded; PW now leads on source quality, currency and neutrality. Residual losses are breadth/citation-volume (no history/techniques section) — inherent to the governance-map scope."},{"key":"catastrophic-risk","priorPw":4.19,"priorWiki":4.33,"priorOutcome":"pw_loses","newPw":4.28,"newWiki":4.5,"newOutcome":"pw_loses","pwDims":{"coverage_breadth":3,"citation_density":4,"source_quality":4.67,"currency":5,"neutrality":4,"claim_accuracy":5},"wikiDims":{"coverage_breadth":5,"citation_density":5,"source_quality":4.33,"currency":4.67,"neutrality":4,"claim_accuracy":4},"weaknessesResolved":2,"weaknessesPartial":2,"weaknessCount":8,"note":"Still loses, but the factual gap is closed: a 'mechanisms and evidence' section now grounds the debate (specification-gaming vs goal-misgeneralization vs power-seeking; Apollo scheming + Anthropic alignment-faking) with the honest no-public-power-seeking-example caveat — judged MORE precise than Wikipedia on the mechanism distinctions. PW leads claim-accuracy + currency; the loss is the conceptual-canon breadth (intelligence explosion, instrumental convergence, the Bostrom/Russell/Ord canon) PW does not aim to be."},{"key":"ca-sb-1047","priorPw":3.75,"priorWiki":3.97,"priorOutcome":"pw_loses","newPw":3.89,"newWiki":4.56,"newOutcome":"pw_loses","pwDims":{"coverage_breadth":3.67,"citation_density":3,"source_quality":4,"currency":4.67,"neutrality":3.67,"claim_accuracy":4.33},"wikiDims":{"coverage_breadth":5,"citation_density":5,"source_quality":4,"currency":4.33,"neutrality":4,"claim_accuracy":5},"weaknessesResolved":2,"weaknessesPartial":5,"weaknessCount":8,"note":"Still loses (the gap widened as the judge panel rated Wikipedia's record-completeness higher this run), but the two most damaging accuracy defects are fixed: the central-mechanism distortion (now correctly a safety-protocol + self-cert bill with 2026 third-party audits, not a pre-deployment testing mandate) and the critical-harm threshold precision, plus a real named coalition. Residual loss is encyclopedic record-completeness (vote counts, public-opinion polling, full coalition) — Wikipedia's strength, not a PW error."}],"summary":{"subjects":3,"flippedToWin":1,"stillLose":2,"factualWeaknessesResolved":"all 3 subjects: the specific factual/accuracy weaknesses recorded in the baseline are resolved (verbatim-verified); the residual losses are breadth/citation-volume, not error"}}}