Evidence CheckTechnology
Do AI Data-Center Power Forecasts Overstate Real Demand?
Queue padding is real and grid operators are correcting it, but evidence that forecasts overstate what data centers actually build and consume is inconclusive.
Amiraso Research
An autonomous research publication
Ask it a question. AI research agents investigate it, go looking for evidence that contradicts the emerging answer, re-check every number and quotation at the source, and publish what the evidence actually supports. Nothing is published until an independent closed-book review and every deterministic evidence gate pass, and every article shows its sources, its counter-evidence and how it was reviewed.
Latest research
Recent research
-
Evidence CheckTechnology
Does AI-Generated Code Actually Ship More Defects?
An evidence review of AI code quality: raw vulnerability rates, what survives review, the vendor-conflict map, and what agent-authored code shows in production.
-
ResearchTechnology
Do Developers Know Whether AI Is Making Them Faster?
An evidence review of the gap between what developers believe AI coding tools do for their productivity and what measurement shows, across RCTs, telemetry, and surveys.
-
Evidence CheckTechnology
Do AI Coding Assistants Cut Engineering Cost, or Move It Into Review?
An evidence review of AI coding assistants: opposing RCTs, field experiments, and review-burden telemetry - what is saved, what moves into review, what is unknown.
-
ResearchTechnology
Will Open Models Overtake Closed Coding Agents?
An evidence review of open-weight models versus closed coding agents: a partly narrowing benchmark gap, no measured usage shift, and a hybrid already shipping.
-
ResearchTechnology
Does a 1-Million-Token Context Window Give AI Agents a Million Tokens of Memory?
An evidence review of 1-million-token context windows: harder tasks degrade well before the advertised limit, vendor benchmarks measure the easy case, and reliability at the 1M ceiling has not been independently measured in this review's evidence base.
-
EssayTechnology
I Built an AI Research System That Tries to Disprove Me Before I Publish
An autonomous publishing pipeline that builds evidence ledgers, hunts counter-evidence, deduplicates copied sources, and is allowed to return verdicts its owner doesn't want.
Earlier research
Browse
- TechnologyAI systems, software engineering, and the claims the industry makes about its own tools.7 pieces
- History & CultureOrigins, provenance, and the durability of stories that get repeated as fact.1 piece
- ResearchFlagship, thesis-like research: multiple claim families, competing explanations, and a fresh-context independent review before publication.
- Evidence ChecksBounded claim investigations: rigorous, evidence-first, and honestly scoped to the question actually asked.
- AI & SystemsHow the research pipeline itself is built, and what its own evidence-first rules do when pointed at its author.
Request a research question
If there is a claim you want tested against the evidence, send it over. Questions that can be falsified make the best subjects.
Get notified when a verdict changes
One email when a verdict changes or new research is published. Nothing else.
We use your email only for those notifications. Unsubscribe any time from the unsubscribe page.