AI contradiction finder for conflicting document evidence
Find where documents disagree about the same fact, event, entity, date, number, obligation or conclusion. Continua compares claims across the complete file set, separates true contradictions from differences in timing, scope, definition or later updates, and cites the evidence on every side.
No credit card required · Uploaded files deleted after the task
What is an AI contradiction finder?
An AI contradiction finder compares claims about the same underlying subject and identifies where the evidence does not agree.
Continua goes beyond keyword matching or document diffing. It aligns related claims across documents and spreadsheets, checks whether timing, scope, definitions or later updates explain the difference, and preserves unresolved conflicts instead of silently choosing one answer.
The goal is not to find the largest possible number of differences. It is to answer:
Which sources genuinely disagree, why do they disagree, and could the conflict change the decision?
See both sides before deciding what is true
Illustrative finding: customer concentration
| Source | Claim | Context |
|---|---|---|
| Borrower Application, p. 6 | Largest customer represents 18% of receivables | Management-provided application |
| AR Aging, Current Balance column | Customer A represents 31% of current receivables | Same uploaded underwriting package |
| Management Note, p. 2 | Concentration calculation excludes disputed invoices | No reconciled concentration figure is provided |
Classification: Unresolved definition-driven inconsistency.
Why it matters: The application and AR aging appear to use different definitions of receivables. The uploaded record does not establish a reconciled concentration figure, so the 18% headline should not be treated as directly supported by the aging report.
What would resolve it: A defined concentration calculation using the same reporting date, receivable population and treatment of disputed invoices.
This is an illustrative example. Continua keeps every material claim, its context and the analysis connected to its source.
Different does not always mean contradictory
Two documents can report different values without either one being wrong. Continua tests the context before classifying the difference.
| Difference type | What it means | Example |
|---|---|---|
| Timing | Both claims may be accurate at different dates | Headcount was 120 in March and 134 in June |
| Scope | The sources measure different populations or boundaries | Global revenue vs. North America revenue |
| Definition | The same label is calculated differently | “Active customer” includes trials in one report but not another |
| Later update | A newer source may supersede an earlier position | An amendment changes a 30-day deadline to 15 days |
| Source hierarchy | One document may control another for a specific question | Signed amendment vs. superseded draft |
| Unresolved contradiction | The claims remain incompatible after context is aligned | Two reports give different totals for the same period, scope and definition |
A useful contradiction analysis does not flatten all differences into “conflicts.” It explains what can be reconciled, what appears superseded and what remains genuinely unresolved.
Six layers of contradiction analysis
Claim matching
Connect statements concerning the same underlying fact, number, event, entity or obligation even when the documents use different wording, aliases, layouts or file types.
Competing positions
Present each materially different position with its source and relevant context instead of collapsing several claims into one synthesized answer.
Context alignment
Check whether the claims refer to the same period, entity, scope, definition, unit, version or contractual position before calling them contradictory.
Conflict classification
Classify the difference as timing, scope, definition, update, supersession, incomplete evidence or an unresolved contradiction.
Materiality
Prioritize conflicts that could affect a decision, valuation, obligation, chronology, compliance conclusion, risk assessment or professional judgment.
Resolution path
State what evidence, definition, calculation or clarification would be needed to reconcile an unresolved conflict.
When the evidence disagrees, keep the disagreement visible
A generic AI answer may encounter two different figures and return one of them. A useful evidence review should not.
- Preserve materially different claims
- Show the evidence behind every side
- Distinguish established facts from inference
- Identify when a later source appears to supersede an earlier one
- Explain when timing, scope or definitions may account for the difference
- Keep the issue unresolved when the uploaded record cannot establish a winner
A plausible explanation is not the same thing as a proven reconciliation.
Ask the complete document set where the evidence disagrees
Analyze all uploaded files for material contradictions and inconsistencies. Match claims about the same facts, events, entities, dates, numbers, obligations and conclusions even when terminology differs. Show each competing position with its context and citation. Determine whether timing, scope, definitions, source hierarchy or later updates reconcile the difference. Do not silently choose a winner. Label inference, preserve unresolved conflicts and state what evidence would be needed to resolve them. Prioritize conflicts that could materially change the decision.
Built for evidence-heavy professional review
Commercial credit and underwriting
Compare borrower applications, credit memos, financial statements, aging reports, covenant certificates, debt schedules and agreements. Surface inconsistent financial figures, unsupported borrower claims and conflicting covenant conclusions. See AI finance underwriting and AI credit review.
Due diligence
Compare the deal narrative with financial statements, contracts, disclosures, schedules and supporting exports to find where the investment story stops matching the underlying evidence. See AI due diligence.
Contract review
Find inconsistent obligations, stale schedules, conflicting definitions and clauses changed by later amendments, and distinguish a genuine conflict from a valid superseding term. See AI contract review.
Procurement and RFP review
Identify requirements, deadlines or instructions that conflict across the base RFP, attachments, addenda and Q&A. See AI RFP analysis.
Medical and claims review
Surface different onset dates, histories, diagnoses, restrictions, treatment accounts or event descriptions across providers and statements. See AI medical chronology.
Audit, compliance and investigations
Compare policies, certifications, approvals, reports, correspondence and operational records when they describe the same requirement or event differently. See AI audit evidence review.
Common contradictions Continua can find
Numbers
Revenue, EBITDA, debt, headcount, prices, volumes, balances, percentages and other figures that disagree across narrative and structured files.
Dates and timelines
Different reported event dates, deadlines, effective dates, delivery dates, treatment dates or milestone dates.
Contractual obligations
Clauses, schedules, amendments and side letters that describe different obligations, thresholds, notice requirements or rights.
Entity facts
Different names, ownership information, customer classifications, counterparties, locations or relationships attributed to the same entity.
Status and conclusions
One document reports an item as complete, compliant or approved while another shows an open condition, exception or missing evidence.
Narrative claims vs. source evidence
A memo, presentation, application or report makes a statement that is not supported — or is contradicted — by the underlying spreadsheet, contract or source document.
From separate claims to a cited conflict report
- 01Upload the complete evidence set — the documents, spreadsheets and supporting files that may contain competing accounts, calculations, definitions or obligations.
- 02Define the disputed question: which facts, periods, entities, figures, obligations or conclusions matter to the review.
- 03Continua aligns related claims, connecting statements about the same underlying subject even when terminology, format or document structure differs.
- 04Classify the difference: timing, scope, definition, later updates, supersession, incomplete support or a genuine unresolved contradiction.
- 05Review both sides at the original source location and export a cited conflict report for professional review.
Not every contradiction is a document change
AI document comparison is most useful when the question is: what changed between these documents or versions?
Contradiction Finder is most useful when the question is: which claims about the same subject cannot be reconciled across the evidence set?
A contract amendment changing a threshold may be a valid update. A compliance certificate using the old threshold may create a contradiction. The two workflows are related, but they answer different questions.
AI contradiction finder vs. basic search
| Keyword search, diff or basic AI summary | Continua Contradiction Finder |
|---|---|
| Finds repeated words or changed text | Matches claims about the same underlying subject |
| Reviews passages independently | Compares competing positions in shared context |
| May treat every different number as a conflict | Tests timing, scope, definitions and updates first |
| May silently choose one answer | Preserves materially different positions |
| Can explain away a conflict without proof | Labels unproven explanations as inference |
| Often leaves the reviewer to assemble sources | Cites the evidence on every side |
| Focuses on difference | Adds classification, materiality and resolution path |
Questions contradiction analysis can answer
- Which sources disagree about the same fact, figure or event?
- Are conflicting numbers based on the same period, scope and definition?
- Does a borrower application match the supporting financial record?
- Did a later amendment supersede an earlier contractual term?
- Does a compliance conclusion rely on a stale threshold?
- Are two accounts irreconcilable, or are they describing different things?
- Which contradiction could materially change the decision?
- What explanation is supported by the uploaded evidence?
- What remains inference?
- What additional evidence would resolve the issue?
A contradiction is evidence. A risk is a decision consequence.
Not every contradiction is material, and not every material risk begins with a contradiction. Contradiction analysis establishes what the evidence does and does not support; risk analysis determines which issues could change the outcome.
See AI risk and red flag analysis to turn evidence-backed conflicts into prioritized decision risks.
Supported files, security and professional judgment
Continua analyzes PDF, DOCX, XLSX and CSV files together, including narrative claims that must be checked against structured data.
Uploaded files are deleted after the task and are not used to train AI models.
Material findings can remain linked to the supporting document, page, sheet, row or available source location so reviewers can inspect the evidence directly.
The absence of a reconciliation in the uploaded documents does not prove that none exists elsewhere. Continua analyzes the evidence provided; qualified reviewers remain responsible for obtaining missing sources and making final determinations. See Trust & Security.
Related Continua workflows
- AI risk and red flag analysis — which evidence-backed issues could materially affect the decision
- AI document comparison — material changes between documents or versions
- AI executive summary — findings, risks and open questions in a decision-ready brief
- AI timeline — chronology when sources report dates or sequences differently
- AI contract review — agreement families, amendments and current effective terms
- AI due diligence — financial, contractual and operational evidence together
Frequently asked questions
Can Continua find contradictions across multiple documents?
Yes. Continua compares claims across the uploaded file set and can identify conflicts involving facts, dates, amounts, entities, obligations and conclusions.
How does it distinguish a contradiction from a normal difference?
It checks whether the claims concern the same subject, time period, scope, definition and version. Differences caused by timing, scope, terminology or later updates are classified separately from unresolved contradictions.
Will Continua decide which source is correct?
Only when the uploaded evidence establishes a clear basis, such as a later controlling amendment or another source with explicit authority for the question. Otherwise, competing positions remain visible.
Can it explain why two documents disagree?
Yes. Continua can identify evidence-supported explanations. Plausible explanations that are not established by the uploaded record remain labeled as inference.
Does each contradiction include citations?
Material conflicts can cite the evidence on every side, including the relevant document and available source location, so the reviewer can inspect the underlying record.
Can it compare a spreadsheet with a PDF?
Yes. Continua can analyze PDF, DOCX, XLSX and CSV files together, including narrative claims that need to be checked against structured data.
Is this the same as comparing two document versions?
No. Version comparison focuses on what changed. Contradiction analysis focuses on incompatible or apparently incompatible claims across the broader evidence set. See AI document comparison.
Can it find contradictions in commercial credit or underwriting files?
Yes. Examples include inconsistent borrower figures, different concentration calculations, stale covenant thresholds, credit-memo claims that do not match source financials, or application statements that conflict with supporting schedules.
Find the conflict before it becomes the conclusion
Upload the complete evidence set and get a cited analysis of what disagrees, what can be reconciled and what remains unresolved.
No credit card required · Zero document retention · Cited, reviewable outputs
Start freeRun Your First Investigation Free
Attach the file pile, ask the hard question, get back a cited report with every conflict flagged. No credit card required.