Evidence Explorer (34 categories)

One row per outcome category — the bird's-eye view. Use this to compare topics against each other: which have the most agreement, the strongest evidence, the most recent activity. To read the individual claims behind a category instead, use the Evidence page (or click any row here to drop into them).

Pick a lens to re-sort the board; click any column header to sort by it; click a row to browse its claims.

What do these lenses and columns mean?

Lenses

Consensus: How much the papers in a category agree on the direction of the effect.
Strength: How strong the underlying evidence is, by study design and grade.
Momentum: How much new research is arriving in the category.
Contested: How much the findings conflict with each other.
Coverage: How much evidence exists in the category at all.

Columns

Papers / Claims: Distinct papers, and total extracted claims, in the category.
Pos / Neg: Papers whose claims lean positive or negative once grouped by paper.
Direction: The balance bar. The badge reads "Positive" or "Negative" when at least 60% of directional papers agree, otherwise "Split".
Depth: A Solid / Moderate / Thin roll-up of evidence strength. Hover a cell for the exact formula.
Causal %: Share of papers using an RCT or quasi-experimental design.
High-grade %: Share of claims the model graded "high".
Gen.: Distinct units of analysis the evidence spans. A rough breadth proxy, not a formal external-validity test.
Effect: The most common effect-size phrase in the category. Only about 11% of claims carry one.
30d / 90d: Papers added in the last 30 and 90 days. The bar is scaled to the busiest category.
Tensions: Pairs of claims that point opposite ways (one positive, one negative) across different papers. Machine-detected candidates, not confirmed contradictions: the rule does not check whether the two claims describe the same population, period, or outcome.
Synthesis: A written "state of the evidence" page, where one has been published.

Every label here is AI-generated and unverified. See About & Methodology for how it is built and where it falls short.

Primary: Secondary:

Category	Papers	Pos	Neg	Direction	Depth	Causal %	High-grade %	Gen.	Effect	30d	90d	Tensions	Claims	Synthesis
Governance & Regulation 📄	684	381	154	Positive	Moderate	11.0%	5.3%	25	statistically significant	145	437	⚡ 141664	1630	View
Organizational Efficiency 📄	531	336	90	Positive	Moderate	13.9%	4.2%	14	+8.49%	130	363	⚡ 75628	1243	View
Technology Adoption Rate 📄	562	324	122	Positive	Solid	15.1%	7.7%	23	over 80%	142	378	⚡ 76888	1178	View
Research Productivity 📄	383	151	45	Positive	Moderate	7.3%	11.2%	14	18-25% increase	66	160	⚡ 21018	1023	View
Output Quality 📄	315	191	77	Positive	Solid	23.8%	8.2%	14	+29.2% signal-to-noise ratio	86	227	⚡ 55650	807	View
Decision Quality	285	145	86	Positive	Solid	21.4%	9.0%	8	≈20%	65	192	⚡ 32001	668	No synthesis yet
AI Safety & Ethics	270	80	144	Negative	Moderate	8.1%	2.7%	11	100% elimination of security risks	46	139	⚡ 23872	630	No synthesis yet
Firm Productivity 📄	283	198	29	Positive	Moderate	14.5%	2.4%	6	≈8–12% (larger gains in high-income countries)	43	136	⚡ 11993	627	View
Market Structure	255	80	90	Split	Moderate	5.5%	1.4%	9	−0.8 percentage points per 1pp robot density	37	100	⚡ 7910	507	No synthesis yet
Task Allocation 📄	221	114	32	Positive	Solid	17.2%	5.5%	12	ĥχ ≈ +1.0 × 10^{-3}	61	172	⚡ 7304	401	View
Skill Acquisition 📄	196	108	33	Positive	Moderate	12.2%	1.6%	13	g = 0.14, 95% CI: [-0.18, 0.47]	35	108	⚡ 4281	317	View
Innovation Output	174	122	15	Positive	Moderate	14.4%	3.0%	10	up to 30%	32	116	⚡ 2959	303	No synthesis yet
Employment Level 📄	154	46	26	Positive	Moderate	11.7%	3.2%	11	≈80% employment rate (prime-age 25–54)	21	67	⚡ 2870	284	View
Fiscal & Macroeconomic	108	47	36	Split	Moderate	12.0%	0.0%	6	β = –0.455, p < 0.001	15	48	⚡ 3464	284	No synthesis yet
Consumer Welfare	125	60	34	Positive	Solid	16.0%	2.0%	7	well below the price of a STEM textbook per student per semester (under worst-case usage ceiling)	24	68	⚡ 3803	244	No synthesis yet
Firm Revenue	149	96	29	Positive	Moderate	14.1%	4.5%	5	+0.13% UCTR, +0.25% UCTCVR	24	66	⚡ 2616	242	No synthesis yet
Task Completion Time	171	126	26	Positive	Solid	25.7%	9.2%	9	≈10x reduction in expensive evaluations	54	128	⚡ 3902	239	No synthesis yet
Inequality Measures 📄	153	23	91	Negative	Moderate	11.8%	3.1%	10	50–70%	26	77	⚡ 2067	227	View
Worker Satisfaction	101	49	33	Split	Solid	31.7%	4.0%	6	strong association between career optimism and perceptions of AI-related competencies	23	70	⚡ 3390	202	No synthesis yet
Error Rate	114	40	66	Negative	Solid	20.2%	6.9%	8	statement/item identity explains more label variance than annotator-level characteristics	23	71	⚡ 3520	189	No synthesis yet
Regulatory Compliance	99	43	41	Split	Solid	15.2%	2.8%	9	zero feeder-limit violations	10	29	⚡ 1812	178	No synthesis yet
Automation Exposure 📄	100	39	38	Split	Moderate	7.0%	6.7%	9	75-80%	27	78	⚡ 1540	163	View
Training Effectiveness	96	65	13	Positive	Moderate	11.5%	2.6%	8	slower than personalized instruction by a factor linear in the number of learner types	17	56	⚡ 742	153	No synthesis yet
Wages & Compensation 📄	84	39	26	Positive	Solid	20.2%	5.5%	6	β = 0.61, p < 0.001	13	36	⚡ 1484	146	View
Developer Productivity	83	59	10	Positive	Solid	21.7%	2.8%	8	β = 0.18, p < 0.01	26	71	⚡ 988	144	No synthesis yet
Team Performance	70	48	7	Positive	Solid	24.3%	2.1%	4	significant partner-type × service-empathy interaction	13	41	⚡ 550	143	No synthesis yet
Job Displacement	96	10	65	Negative	Moderate	7.3%	0.9%	9	substantially larger (for higher income households vs lower income families)	15	62	⚡ 454	117	No synthesis yet
Hiring & Recruitment	40	28	4	Positive	Moderate	10.0%	11.3%	6	traditional IT service hiring displaced by product-focused roles and GCC expansion	4	18	⚡ 224	71	No synthesis yet
Social Protection	40	23	12	Positive	Moderate	12.5%	4.5%	6	public cultural services can act as productive social infrastructure advancing SDG 8 (decent work) given adequate digital capacity	6	23	⚡ 149	66	No synthesis yet
Creative Output	28	15	7	Positive	Solid	21.4%	0.0%	4	5.5 to 10.2 percentage point higher likelihood	8	21	⚡ 380	64	No synthesis yet
Skill Obsolescence	53	0	46	Negative	Moderate	11.3%	0.0%	5	up to 53% reduction	13	38	⚡ 132	61	No synthesis yet
Labor Share of Income	31	8	12	Negative	Solid	19.4%	3.8%	5	tends to zero	5	19	⚡ 199	53	No synthesis yet
Worker Turnover	19	9	10	Split	Moderate	5.3%	0.0%	4	β = 0.348, p < 0.001	2	3	⚡ 87	32	No synthesis yet
Industry	1	0	0	—	Moderate	0.0%	0.0%	1	—	—	1	—	1	No synthesis yet

⚡ A tension count is how many claim pairs in a category point opposite ways (one positive, one negative) across different papers. These are machine-detected candidates, not confirmed contradictions: the rule does not check whether the two claims describe the same population, period, or outcome. Read both papers before treating any pair as a real disagreement.