Analysis
AI Analysis Tool
A reader workbench for checking Ukraine-war claims, scanning public feeds by region and applying transparent mathematical heuristics before a claim is shared or cited.
Prototype page: pasted text is analysed locally in the browser. The web radar fetches public feeds through this site API; it is not an exhaustive intelligence system and does not store submitted material.
Sections
Analysis
What the tool is for
The tool is a disciplined pause before sharing war claims. It asks whether a sentence names a source, gives a date, defines a unit, identifies a place and preserves the uncertainty carried by the evidence.
The scanner is intentionally transparent. It does not present an oracle verdict. It shows risk flags, source diagnostics, named places, repeated phrases, likely consequence categories and the verification steps that should come before publication.
Key points
- Treat every output as a draft for human review.
- Prefer original sources over aggregator headlines.
- Separate facts, estimates, scenarios, opinion and propaganda claims.
- Never turn a party claim or a viral number into a verified total without method and date.
Analysis
Mathematical methods used
The workbench follows the problem-grouping style visible in the extracted Dolinsky algorithm text: each hard problem is broken into explicit subproblems that can be tested. Stringology supplies multi-pattern matching over claim language. Combinatorial optimisation selects a compact verification set. Computational geometry turns named places into text-geography signals. Graph theory links signals, places, flags and verification actions. Game theory gives a posture label for incentives suggested by the text.
These methods make the analysis inspectable, but they do not prove battlefield truth. For direction of war, the output is only a text-derived signal and must be checked against dated maps, official updates and independent reporting.
Key points
- Pattern matching finds signal language and repeated phrases.
- Greedy set cover chooses the smallest useful verification checklist.
- Geometry summarises named places without pretending to be a live map.
- The C++ kernel under native/war-analysis mirrors the heavy-analysis path for offline experiments.
Features
- Scan public RSS and Atom feeds for UK, EC/EU, USA, Russia, Ukraine and rest-of-world coverage.
- Flag unsupported numbers, missing dates, missing geography, casualty claims without attribution and loaded narrative language.
- Use multi-pattern string matching, greedy set-cover verification, graph modelling, text geography and game-theory posture scoring.
- Keep source diagnostics visible when official feeds block, time out or return no matching items.
Experimental tool: results are automated indicators for review. Verify sources before relying on any claim, score or generated interpretation.
After completing the Claim scanner, press Math engine to analyse the results.
Words
0
False-claim risk
25%
Source completeness
0%
Claim flags
No major claim flags detected.
War signals
No strong war-signal pattern detected.
Direction and geography
No geographic direction can be inferred from the text.
Cautious rewrite
Add a claim to generate a cautious source-bound rewrite.