How do you use Super Intelligence for research? Use SI to widen search, organise sources, extract comparable evidence, preserve provenance, identify disagreement, generate hypotheses and prepare synthesis—while keeping source quality, methodology, uncertainty and final interpretation visible to the human researcher.
This article is part of the eduKateSG Workplace Super Intelligence Hub. It follows the Professional Writing article. Professional writing turns evidence into documents. This page owns the earlier stage: research as a source-and-evidence workflow.
In this eduKateSG series, Super Intelligence is the practical machine-intelligence layer commonly described as artificial intelligence, generative AI, assistants, copilots, agents and connected automation. Strong SI research is not “ask the model for the answer.” It is “use machine intelligence to make evidence easier to discover, compare and interrogate without allowing the synthesis layer to replace the evidence layer.”
Research Is an Evidence Pipeline
A useful research workflow is: Question → Scope → Search → Source Selection → Extraction → Comparison → Synthesis → Challenge → Conclusion → Decision or Publication. Each state creates different risks.
Super Intelligence is particularly useful in search expansion, extraction, comparison and synthesis. Humans remain important for source quality, methodology, interpretation, relevance and consequential conclusions.
The Research Question Comes First
A vague research request produces vague evidence. Before using SI, define the decision or knowledge gap the research must resolve.
- Question: what exactly are we trying to know?
- Population or object: who or what does the question concern?
- Time period: which dates matter?
- Geography: which jurisdiction or market?
- Outcome: what measure or state matters?
- Decision: what will change if the evidence supports one answer rather than another?
- Evidence standard: what sources are acceptable?
A well-formed question reduces the temptation to accept whatever evidence is easiest to find.
The Research Brief
A research brief turns the question into an operating contract. It defines scope, source preferences, exclusions, terminology and output format.
- Primary question
- Secondary questions
- Required source types
- Excluded source types
- Date range
- Geographic scope
- Key definitions
- Required fields for extraction
- Known uncertainties
- Deadline
- Reader or decision-maker
Search Expansion
SI can generate synonyms, related concepts, technical terms, historical names, product names, organisational acronyms and alternative phrasings that broaden search coverage.
This is useful because researchers often search using the terminology they already know. Search expansion can expose adjacent vocabulary without claiming that every new term is relevant.
Search Narrowing
SI can also refine broad search into specific sub-questions. A general request such as “AI productivity” may need to split into task type, worker group, measurement method, time horizon and industry.
Narrowing improves evidence comparability.
Source Discovery
Use SI to identify likely source classes: primary research, official statistics, standards, company filings, legislation, technical documentation, academic papers, reputable journalism or domain reports.
The system should help the researcher find where evidence might live. It should not fabricate a source when none has been retrieved.
Source Quality Is a Human Judgment
A model can summarise a weak source as fluently as a strong one. Researchers should evaluate authorship, method, date, population, incentives, transparency and whether the source directly supports the claim.
Source quality should be explicit in the workflow rather than inferred from confident prose.
The Source Hierarchy
- Primary evidence: original data, law, study, filing, official record or direct observation.
- Authoritative synthesis: standards body, regulator, systematic review or respected institutional analysis.
- Secondary analysis: expert interpretation, journalism or industry commentary.
- Community evidence: practitioner experience, forums or social discussion.
- Model synthesis: navigation and interpretation layer, never the original source.
Different questions justify different evidence. Community experience may be valuable for usability research while official data may be stronger for prevalence or legal status.
Provenance Is Non-Negotiable
Every material extracted claim should retain a path back to its source. Provenance includes source title, date, author or institution, URL or identifier, and the passage, table or section supporting the field.
SI can automate much of this bookkeeping. That is a high-leverage use because provenance work is repetitive but critical.
The Evidence Table
For multiple sources, create a structured evidence table so claims can be compared rather than summarised separately.
- Source
- Date
- Population or object
- Method
- Sample or scope
- Main finding
- Effect or value
- Uncertainty
- Limitations
- Relevant quote or passage
- Researcher note
- Confidence or source-quality note
The exact fields depend on the domain. The important point is comparability.
Structured Extraction
SI can extract the same fields from many papers, reports or documents. This reduces manual reading burden but should be validated, especially where tables, footnotes or nuanced definitions matter.
Missing information should remain missing rather than being inferred.
The No-Guess Extraction Rule
If a source does not state a value, the system should return “not reported” rather than estimate one from context unless estimation is explicitly part of the method.
This rule protects the evidence table from becoming a mixture of source data and model inference.
Comparing Sources
Once evidence is structured, SI can identify agreement, disagreement, gaps, different definitions and different populations.
The comparison should preserve why sources differ rather than averaging incompatible findings into one smooth conclusion.
Definition Mismatch
Two sources may use the same word differently. “Productivity”, “adoption”, “success” or “risk” can have incompatible definitions.
SI can surface definition mismatches before the researcher compares values.
Population Mismatch
A finding from one group may not generalise to another. Research should record population, industry, age, geography and other relevant scope.
Time-Period Mismatch
Evidence from different years may reflect different technology, regulation or market conditions. Current questions should not silently rely on historical evidence as if it were present state.
Method Mismatch
Survey, experiment, observational study, case study and anecdote answer different questions. SI can classify methods and help explain what each design can and cannot establish.
Metric Mismatch
Two sources may measure related but different outcomes. The evidence table should preserve original metrics before any synthesis.
Contradiction Is Useful
When sources disagree, do not force consensus. Ask whether the disagreement comes from method, scope, data, definition or genuine uncertainty.
The research output should preserve unresolved disagreement where the evidence does not support closure.
The Contradiction Map
- Claim
- Supporting sources
- Contradicting sources
- Difference in method
- Difference in population
- Difference in date
- Difference in definition
- Possible explanation
- What evidence would resolve the disagreement
Gap Detection
SI can identify which important questions remain unsupported after the evidence table is built. This is more useful than simply expanding the document with more prose.
A good research workflow knows what it does not know.
The Research Gap Register
- Missing population
- Missing time period
- Missing source type
- Missing comparison
- Conflicting evidence
- Unreported variable
- Unclear definition
- Method limitation
- Unavailable current data
Hypothesis Generation
SI can generate plausible explanations for observed patterns. Treat these as hypotheses, not findings.
A useful hypothesis should identify what evidence would support or disconfirm it.
Counter-Hypothesis Generation
Ask SI for alternative explanations that fit the same evidence. This can reduce anchoring and confirmation bias.
The researcher should then test which explanation has stronger evidence.
The Disconfirmation Pass
Before settling on a conclusion, search deliberately for evidence that would weaken it. SI can propose search terms, opposing claims and missing populations.
Research becomes stronger when the system helps challenge the preferred answer rather than only support it.
Synthesis
Synthesis should answer the research question by combining evidence according to strength and relevance. It is not a long sequence of source summaries.
SI can draft a synthesis from the evidence table while citing each material claim. The researcher verifies that the weighting of sources is justified.
The Synthesis Structure
- Research question
- Short answer
- Strongest evidence
- Important disagreement
- Scope and limitations
- Implication
- Unresolved questions
- Decision relevance
Fact vs Interpretation
Research writing should separate what sources report from what the researcher concludes. SI can tag statements as sourced fact, interpretation, assumption or recommendation for review.
Confidence Language
Use confidence language that matches the evidence. “Shows” is stronger than “suggests”. “Associated with” is not the same as “causes”. “In this sample” is narrower than “in general”.
SI can help identify overconfident wording, but the researcher sets the evidentiary standard.
The Citation Check
For each citation, verify that the source actually supports the nearby claim. A correct source can still be misused if the text overstates it.
The Quote Check
Quoted material should preserve meaning and context. Keep quotation short and use paraphrase for synthesis.
The Number Check
Verify percentages, denominators, units, currencies and time periods against the source. Exact arithmetic belongs in deterministic tools.
The Date Check
Record publication date and, where relevant, the date the underlying data was collected. A 2026 article may report 2023 data.
The Geography Check
A national statistic should not be presented as global evidence. A Singapore regulation should not be treated as applying elsewhere. SI can help tag jurisdictions and scope.
The Causal-Inference Check
Ask what study design supports the causal claim. Observational associations, self-reports and before-after comparisons have different limitations.
SI can explain those distinctions but should not upgrade weak design through persuasive language.
The Researcher-in-the-Loop Model
The researcher should enter at the points where quality cannot be reduced to routine extraction: source selection, methodology, interpretation, uncertainty and final conclusion.
Machine intelligence carries more of the repeated information processing so human attention can focus on evidence judgment.
Research Pattern: Market Research
SI can collect competitor information, structure product features, compare positioning and summarise public evidence. Researchers should distinguish company claims, observed behaviour and independent market data.
Research Pattern: Policy Research
Use official legislation, regulator guidance, government statistics and current policy documents as primary sources. SI can compare versions and summarise implications, but jurisdiction and date must remain explicit.
Research Pattern: Academic Literature
SI can help generate search terms, extract study characteristics and cluster findings. Methodological quality and citation verification remain essential.
Research Pattern: Technical Research
Documentation, standards, repositories, benchmarks and release notes may matter more than general web summaries. SI can compare versions and explain architecture while exact specifications remain source-grounded.
Research Pattern: Customer Research
SI can cluster interview transcripts, support tickets and survey comments into themes. Human researchers should preserve representative examples, minority views and context so frequency does not become false importance.
Research Pattern: Competitor Research
Separate competitor claims, pricing, product documentation, reviews and independent evidence. SI can maintain a comparison table with dates so changes are visible.
Research Pattern: Legal Research
SI can assist with source discovery, chronology and comparison. Qualified legal professionals remain responsible for authority, jurisdiction, currency and legal interpretation.
Research Pattern: Financial Research
Use filings, audited statements, official market data and other authoritative sources. SI can extract and explain, while calculations and investment judgments remain separately verified.
Research Pattern: Education Research
SI can compare studies, curricula, assessment data and learning evidence. Researchers should distinguish classroom context, age groups, intervention type and outcome measurement.
Research Pattern: Operational Research
SI can analyse incident histories, process documentation and performance records to identify repeated causes or bottlenecks. Causal conclusions should remain evidence-based.
The Research Notebook
Maintain question, search terms, sources reviewed, evidence table, gaps, hypotheses and decisions in one visible research workspace.
The notebook prevents SI conversations from becoming disconnected fragments.
The Source Register
Keep a register of accepted sources and why they matter. This is particularly useful for recurring research topics where the same authoritative sources should be checked repeatedly.
The Research Change Log
Record material changes to the evidence base: new report, revised statistic, retracted paper, updated regulation or product change.
This makes recurring research maintainable rather than disposable.
The Living Research Brief
For fast-moving topics, the research output can become a living brief with update triggers. SI can monitor approved sources and flag material changes.
Human review decides whether the conclusion needs revision.
The Research Monitoring Rule
Monitor specific unresolved questions or authoritative sources rather than “all news about the topic”. Tight monitoring produces more actionable signal.
The Research Scope-Creep Trap
SI can make it easy to explore endless adjacent questions. Return to the decision or research question and record tangents separately.
The Citation-Hallucination Trap
Never trust a citation merely because it looks plausible. Retrieve the source and confirm the claim.
The Summary-Without-Reading Trap
SI summaries can reduce reading load but may omit methodological details. Read the primary source where the conclusion depends on those details.
The Source-Volume Trap
More sources do not automatically improve research. Ten low-quality sources can add noise. Prioritise relevance and quality.
The Consensus Trap
A model may smooth disagreement into consensus. Preserve outliers and contradictions when they are supported.
The Recentness Trap
Newer is not always stronger, but for fast-changing technology or policy the date can materially affect relevance. Record both publication and evidence dates.
The Popularity Trap
Search ranking and social discussion reflect attention, not necessarily truth. Community evidence can be useful for experience, but separate it from authoritative factual claims.
The Authority Trap
An authoritative institution can still publish a source outside the exact population or time frame of the question. Authority does not remove scope limits.
The Model-Knowledge Trap
A model may know background information but lack current updates. For time-sensitive research, retrieve current evidence rather than relying on remembered model knowledge.
The Research Automation Ladder
- Search-term generation
- Source discovery
- Metadata extraction
- Evidence-table population
- Source comparison
- Gap detection
- Draft synthesis
- Counterargument generation
- Human-reviewed living brief
- Condition-based monitoring of approved sources
Every level should preserve provenance.
The Research Metrics
- Time to relevant source
- Number of high-quality sources
- Extraction correction rate
- Citation support rate
- Unresolved contradiction count
- Researcher review time
- Time to decision-ready synthesis
- Rate of important new evidence found
- Update latency for living briefs
The Research Pilot
Choose one repeated research question or evidence-gathering task. Baseline search and extraction time. Use SI to generate search terms, populate an evidence table and draft synthesis from retrieved sources.
Keep final source-quality and interpretation judgments human-controlled. Measure whether evidence coverage improves without unacceptable correction burden.
The Research Readiness Test
- Question is well defined.
- Source types are known.
- Provenance can be preserved.
- Extraction fields are explicit.
- Researcher can judge source quality.
- Important claims can be traced.
- Currentness requirements are known.
- Output is tied to a reader or decision.
The Research Deletion Test
Ask whether the research is needed. Cheap generation can encourage unnecessary reports. If no decision, learning or maintained knowledge will use the output, do not research merely because SI makes it easy.
The Research Simplification Rule
Start with the minimum evidence required to answer the question. Expand only when uncertainty or consequence justifies deeper work.
The Research Source Rule
The model is an interface to evidence, not the evidence itself.
The Research Uncertainty Rule
When evidence does not support closure, preserve uncertainty and specify what would resolve it.
What This Article Owns
This page owns research as a workplace Super Intelligence workflow: question design, source discovery, extraction, comparison, provenance, synthesis, challenge and maintenance.
Professional Writing owns document production after the evidence is assembled. The next article will apply SI to spreadsheets and data analysis.
Frequently Asked Questions
Can SI do research for me?
It can accelerate discovery, extraction, comparison and synthesis. Important research still requires source verification, methodology judgment and human interpretation.
Can SI find academic papers?
It can help generate search terms and identify likely sources, but the actual paper should be retrieved and verified before citation.
Can SI summarise research papers?
Yes. For consequential conclusions, inspect the methods, population, limitations and original results rather than relying only on the summary.
Can SI compare multiple sources?
Yes. Structured evidence tables are especially useful because they preserve differences in date, population, method and definition.
How do I prevent fake citations?
Require retrieval and provenance. Do not accept citation text that has not been opened or otherwise verified against the real source.
Can SI monitor research over time?
Yes. Use a bounded set of authoritative sources or unresolved questions and have material changes flagged for human review.
What comes next?
Continue to How to Use Super Intelligence with Spreadsheets and Data Analysis, which covers structured data, formulas, data cleaning, analysis, interpretation and verification.
The Core Research Rule
Use Super Intelligence to lower the cost of finding and organising evidence while making the evidence trail stronger, not weaker.
The best SI research leaves the human with more sources, clearer comparisons, visible uncertainty and a conclusion whose path back to evidence remains intact.
The Source-Quality Rubric
Source quality should be reviewed explicitly rather than absorbed into a general impression. A simple rubric can record directness, method transparency, recency, relevance, independence, completeness and whether the source is primary or secondary.
- Directness: does the source directly address the research question or only an adjacent issue?
- Transparency: can the method, sample, assumptions or calculation be inspected?
- Recency: is the evidence current enough for the question?
- Relevance: does the population, geography and context match the research need?
- Independence: does the source have incentives that should be considered?
- Completeness: are important limitations or missing data visible?
- Provenance: can the original evidence be located?
The rubric is not a universal score. A source can be excellent for one question and weak for another. The purpose is to make the researcher’s judgment inspectable.
Primary Sources vs Convenient Summaries
Super Intelligence makes summaries easy to consume. This can create a bias toward secondary material because it is easier to read. For important claims, return to the primary source where possible.
A press release about a study is not the study. A blog describing a law is not the legislation. A product comparison quoting vendor claims is not independent evidence. SI should help the researcher move from summary to source rather than stop at the summary.
The Evidence Directness Test
Ask how many inference steps sit between the source and the conclusion. A source that directly measures the target outcome is stronger for that question than one that measures a proxy.
SI can help label proxies and inferential steps so the researcher does not accidentally treat indirect evidence as direct.
The Research Scope Ledger
Maintain a short record of scope decisions: what counts as relevant, what is excluded and why. This prevents the search process from silently expanding into adjacent topics.
If the question changes materially, create a new version of the research brief rather than mixing two objectives into one evidence table.
The Search Strategy Ledger
Record important search terms, databases, websites, date filters and source classes already examined. This helps avoid repeating the same search and gives the researcher a way to explain coverage.
SI can expand the ledger with synonyms and adjacent technical terms, but the human should decide which terms remain in scope.
The Negative Search
Researchers often search only for evidence supporting the working hypothesis. A negative search deliberately looks for failures, contradictions, null results, adverse outcomes and critiques.
SI is especially useful here because it can suggest alternative phrasings that a researcher anchored on one conclusion might not think to use.
The Unknown-Unknown Search
Ask SI what categories of evidence might matter that are absent from the brief: implementation cost, user behaviour, legal context, technical constraints, selection effects or long-term outcomes.
Treat the suggestions as search directions, not as facts.
The Historical Search
Current questions sometimes require historical context. SI can help identify earlier terminology, previous policy names, older product generations and prior research eras.
Historical evidence should remain clearly dated and should not be presented as directly representative of current conditions.
The Jurisdiction Search
For policy, legal, education, finance and compliance questions, identify the relevant jurisdiction explicitly. SI can help compare how definitions or rules differ across countries or regions.
The output should not collapse different legal or policy regimes into one generic answer.
The Language Search
Important evidence may exist in multiple languages. SI can help translate search terms and summarise retrieved sources, but the researcher should verify critical translations and preserve original context where meaning is sensitive.
The Grey-Literature Search
Not all useful evidence appears in peer-reviewed journals. Government reports, standards, working papers, conference papers, technical reports and organisational datasets may be important.
Classify source type and limitations rather than treating publication venue as a simple quality switch.
The Community-Evidence Search
Forums, Reddit, practitioner communities and user reviews can reveal lived experience, failure modes and practical sentiment. They are useful for experience questions but weaker for population-level factual claims.
SI can cluster recurring themes while preserving the distinction between anecdote and general evidence.
The Expert-Interview Workflow
For primary qualitative research, SI can help prepare interview questions, transcribe or summarise interviews where appropriate, code themes and compare viewpoints.
The human researcher should preserve consent, context, minority views and the difference between participant claims and verified facts.
The Survey Workflow
SI can help draft survey items, identify ambiguous wording, group open-text responses and prepare descriptive summaries.
Sampling, questionnaire validity, response bias and statistical inference remain methodological responsibilities.
The Case-Study Workflow
SI can help assemble chronology, stakeholders, decisions, evidence and outcomes. The researcher should be cautious about generalising one case to a broader population.
The Experiment Workflow
SI can help document hypotheses, treatment conditions, measures and analysis plans. Exact randomisation, statistics and causal claims should use appropriate methods and deterministic software.
The Observational-Study Workflow
SI can help identify variables, confounders and alternative explanations. The researcher remains responsible for understanding what observational data can and cannot establish.
The Systematic-Review Workflow
SI can accelerate search-term expansion, screening support, metadata extraction and evidence tables. A rigorous review still requires transparent inclusion criteria, reproducible search and human oversight of study eligibility and quality.
The Meta-Analysis Boundary
A true meta-analysis involves statistical synthesis that should be performed using appropriate quantitative methods. SI can help prepare data and explain results, but the calculations and eligibility logic should be independently reproducible.
The Data-Extraction Validation Plan
For repeated extraction, define a validation sample. Compare SI-extracted fields with human-verified values. Track field-level correction rather than only document-level success.
Some fields may be easy while one nuanced field drives most error. Field-level analysis tells the team where human review remains necessary.
The Evidence Table Version
Evidence tables should be versioned when the research is maintained over time. New sources, corrected fields or changed definitions should be traceable.
This prevents the final synthesis from depending on an unexplained change in the underlying evidence.
The Source Deduplication Problem
The same finding may be repeated across press releases, news articles and reports. SI can help identify duplicate chains so the researcher does not count one underlying result as multiple independent sources.
The Citation Chain
When one source cites another for a key claim, trace the chain back to the original evidence where practical. Secondary sources can simplify or distort the original result.
The Evidence Weighting Problem
Synthesis is not democratic voting among sources. One strong study may deserve more weight than ten weak anecdotes. The researcher should explain why some evidence is more persuasive.
SI can propose weighting criteria, but the final evidentiary judgment belongs to the human research process.
The Null-Result Problem
Negative or null results may be underrepresented in available sources. Researchers should search for them deliberately when evaluating whether an effect is robust.
The Publication-Bias Problem
Published evidence can overrepresent positive or novel findings. SI can flag publication bias as a concern, but domain-specific methods may be required to evaluate it rigorously.
The Selection-Bias Problem
Surveys, user reviews and observational datasets often reflect who chose to participate. The research output should state how selection could affect interpretation.
The Survivorship-Bias Problem
Research can over-focus on successful companies, products or interventions because failures are less visible. Ask what disappeared from the sample and whether it matters.
The Recency-Bias Problem
SI and web search may overweight recent sources because they are prominent. Historical evidence can still matter, especially for long-run effects. Balance recency with relevance.
The Authority-Bias Problem
Prestigious institutions can produce high-quality work, but authority should not replace reading the method or scope. Treat reputation as one signal, not proof.
The Model-Agreement Trap
Asking multiple models the same question and receiving similar answers does not create independent evidence if they rely on overlapping training or the same retrieved sources.
Research confidence should come from source independence and method, not model consensus.
The Research Agent
A bounded research agent can search approved sources, collect metadata, populate evidence tables and prepare update briefs. It should have explicit scope, source rules, stopping conditions and provenance requirements.
The agent should not silently expand the question or publish conclusions without review.
Research Agent Stop Conditions
- Question becomes ambiguous.
- Source cannot be retrieved.
- Evidence conflicts materially.
- Required jurisdiction is unclear.
- Paywalled or inaccessible source blocks verification.
- Data extraction confidence is insufficient.
- New evidence changes the research scope.
- Sensitive data or restricted sources appear.
Stop conditions are evidence discipline, not failure.
The Research Agent and Browsing
Browsing agents can collect many sources quickly. The system should keep a search log, avoid duplicate citation chains and preserve the difference between retrieved evidence and model synthesis.
The Research Agent and PDFs
PDF research can involve tables, footnotes and layout-sensitive information. Extraction should preserve page or section reference so reviewers can inspect the original material.
The Research Agent and Spreadsheets
Evidence tables often live in spreadsheets. SI can help populate and explain them, but formulas, filters and calculations should remain reproducible.
The next article in this series covers spreadsheet and data-analysis workflows in depth.
The Research Agent and Connected Knowledge
For internal research, an agent may need company documents, project history or private datasets. Access should follow organisational permissions and data rules.
The system should not retrieve sensitive information merely because it is technically available.
The Research Agent and External Publication
Automatic publication should be separate from research. A system can assemble evidence and draft a brief without having authority to publish claims publicly.
The Research Review Layers
- Extraction review: are fields copied correctly?
- Source review: are sources appropriate and current?
- Method review: does the design support the claim?
- Synthesis review: are stronger sources weighted appropriately?
- Scope review: does the conclusion stay within population, time and geography?
- Decision review: is the implication justified?
- Publication review: are citations, caveats and sensitive claims acceptable?
Not every low-risk research task needs all seven layers, but high-consequence research may.
The Research Decision Brief
When research supports a decision, convert the synthesis into a decision brief: short answer, strongest evidence, major uncertainty, options, recommendation and what would change the recommendation.
This keeps the research connected to its purpose.
The Research Update Brief
For recurring topics, an update brief should focus on what changed since the last review, why it matters and whether the previous conclusion still holds.
SI is especially useful for delta analysis because it can compare old and new evidence states.
The Research Archive
Keep final evidence tables, accepted briefs, source registers and change logs. Research that is likely to recur should become organisational memory rather than disappear into personal chat history.
The Research Knowledge Loop
When recurring questions reveal missing internal documentation, convert the answer into maintained knowledge where appropriate. Research should reduce future search cost.
The Research Transfer Test
A research workflow that works in one domain should be reassessed before transfer. Legal, scientific, market and user research have different source hierarchies and evidence standards.
Transfer the method—question, provenance, extraction, comparison—not the assumption that every domain has the same rules.
The Research Independence Test
Have another researcher reproduce the evidence path from the brief. If they cannot find the supporting sources or understand why evidence was included, the workflow depends too much on private context.
The Research Reproducibility Test
For important research, preserve enough detail that another person can repeat the search, extraction and analysis at least approximately.
SI makes reproducibility easier when it records search and source state explicitly.
The Research Currentness Test
For fast-moving topics, mark the date through which the evidence was reviewed. A research brief without a currentness boundary can be mistaken for a live answer long after the environment changes.
The Research Sensitivity Test
Ask whether the conclusion changes when one weak source is removed, one assumption changes or one population is excluded. Conclusions that collapse under small changes deserve weaker confidence.
The Research Scenario Test
For strategic research, explore multiple future scenarios rather than one forecast. SI can help generate internally consistent scenarios while the researcher sets plausible assumptions.
The Research Decision-Threshold Test
Sometimes the exact answer is less important than whether evidence crosses a decision threshold. Define what level of evidence would cause action.
This keeps research proportionate to the decision.
The Research Cost Rule
Research depth should match consequence. A low-risk internal choice may justify a quick scan. A major investment, legal conclusion or public claim may justify deeper review.
SI lowers the cost of research, but cheap research should not become endless research.
The Research Stop Rule
Stop when additional evidence is unlikely to change the decision or materially reduce uncertainty. Super Intelligence can make search infinite; the workflow still needs closure.
The Research Escalation Rule
Escalate to a domain expert when source interpretation, professional authority or methodological nuance exceeds the researcher’s competence.
The Research Privacy Rule
Internal research may involve personal, confidential or commercially sensitive material. Use approved environments and minimise data access.
The Research Security Rule
External documents and websites are untrusted content. Tool-using research agents should not let retrieved text override system rules or access unrelated internal data.
The Research Attribution Rule
Preserve authorship and attribution for external ideas. SI synthesis should not erase the difference between source argument and the researcher’s conclusion.
The Research Copyright Rule
Use short necessary quotations and prefer paraphrase for synthesis. Avoid reproducing large portions of protected material simply because a model can copy them.
The Research Integrity Rule
Do not fabricate data, citations or methodology to fill gaps. Unknown should remain unknown.
The Research Portfolio
Teams running many research projects can maintain a portfolio of active questions, source owners, update dates and decision relevance. SI can help identify overlap and prevent duplicate research.
The Research Prioritisation Rule
Prioritise questions that unblock decisions, affect many workflows or reduce recurring uncertainty. Avoid producing research because the topic is interesting but operationally unused.
The Research Time-to-Value Rule
A fast research workflow is valuable when the evidence can be trusted soon enough to affect the decision. If source validation takes longer than the decision window, narrow the question or use a more conservative conclusion.
The Research Handoff Packet
- Question
- Scope
- Short answer
- Evidence table
- Strongest sources
- Key disagreement
- Uncertainty
- Recommendation or implication
- Update date
- Owner
This packet lets the decision-maker inspect the evidence without repeating the entire research process.
The Research Control Room
For recurring or high-value research, track active question, evidence cutoff, source status, unresolved gaps and next review date. A spreadsheet is often enough.
The Research Failure Catalogue
- Question too broad
- Source fabricated or unverified
- Population mismatch
- Date mismatch
- Metric mismatch
- Duplicate evidence
- Unsupported causal claim
- Important contradiction hidden
- Missing null evidence
- Extraction error
- Stale conclusion
- Decision not linked to evidence
Classify failures so the workflow improves at the correct layer.
The 30-Day Research Workflow Build
Week 1 — Define
Choose one recurring research question, source classes, extraction fields and output.
Week 2 — Extract
Build the evidence table from representative sources and validate the extraction process.
Week 3 — Synthesize
Compare evidence, map contradictions, draft synthesis and run disconfirmation searches.
Week 4 — Maintain
Create the source register, update trigger and decision brief. Decide whether the topic deserves living monitoring.
The Research Operating Standard
- Question is explicit.
- Scope is bounded.
- Source classes are defined.
- Provenance is preserved.
- Extraction does not guess missing values.
- Disagreement remains visible.
- Method and population are recorded.
- Material claims are traceable.
- Uncertainty is stated.
- Conclusion stays inside the evidence.
- Update date is visible.
- Decision or reader use is clear.
The Final Research Principle
Super Intelligence should make it easier to find evidence without making it easier to forget where the evidence came from.
The research system is strong when machine speed increases coverage and comparability while human judgment still determines what deserves belief, what remains uncertain and what conclusion the evidence can legitimately support.
Applied Playbook: Market Entry Research
A market-entry question should separate market size, customer demand, regulation, competitors, distribution, pricing and operating constraints. SI can help discover sources and normalise evidence into one table.
The synthesis should distinguish measured market data from vendor claims and analyst estimates. A market-size number without method or date should remain a weak input until the source can be inspected.
Applied Playbook: Competitor Intelligence
Build a dated evidence table across product, pricing, positioning, distribution, customer feedback and public strategic signals. SI can monitor official pages, release notes and reputable reporting, but the output should preserve whether each claim comes from the competitor, an independent source or user experience.
Avoid treating absence of evidence as evidence that a competitor lacks a capability. The correct state may be “not verified.”
Applied Playbook: Vendor Due Diligence
Vendor research can include company background, product capability, security materials, data handling, pricing, integrations, customer evidence, incident history and contractual terms. SI can organise the packet and identify missing due-diligence items.
Final procurement decisions should still rely on verified documents and authorised commercial, legal and security review.
Applied Playbook: Product Research
SI can compare specifications, documentation, user reviews, benchmark results and release history. Separate official specifications from independent performance evidence and subjective reviews.
For changing software products, version and date matter. A review written for an older version may no longer be representative.
Applied Playbook: Customer Research
Combine interview transcripts, surveys, support logs and behavioural data carefully. SI can cluster themes and map recurring pain points, but frequency in a small qualitative sample should not be treated as population prevalence.
Preserve dissenting and minority experiences where they may reveal important edge cases.
Applied Playbook: User-Experience Research
SI can help code usability sessions, cluster friction points and compare behaviour across participants. Researchers should distinguish what users say from what they actually do and preserve context around task conditions.
Applied Playbook: Policy Research
Start with current official policy, legislation, regulator guidance and authoritative explanatory material. SI can compare changes across versions and produce a chronology of how the policy evolved.
Always identify jurisdiction and effective date. A policy from one country should not be generalised to another without explicit comparison.
Applied Playbook: Regulatory Research
Regulatory research should preserve source hierarchy: statute or regulation, regulator guidance, official interpretation and secondary commentary. SI can help locate and compare, but qualified professionals may be required for legal interpretation.
Applied Playbook: Education Research
Education research often varies by age, curriculum, intervention length, assessment method and classroom context. SI can create structured evidence tables across these variables.
Do not generalise a result from one age group or subject to all learners without evidence.
Applied Playbook: Workforce Research
When researching productivity, jobs or skills, record occupation, task, worker population, experience level, technology, measurement method and time horizon. Broad headlines about “AI productivity” can hide very different task-level effects.
Applied Playbook: Technology Trend Research
Fast-moving technology research needs strong currentness control. Use official documentation, release notes, standards and recent independent testing. Mark the evidence cutoff date visibly.
SI can monitor new releases, but researchers should decide whether a change is material to the previous conclusion.
Applied Playbook: Security Research
Use authoritative advisories, vendor security notices, standards and reputable technical analysis. SI can summarise vulnerability details and mitigation guidance, but security teams should verify critical actions against the original advisory and local environment.
Applied Playbook: Financial and Economic Research
Preserve unit, currency, inflation basis, period and data source. SI can prepare comparisons and explanations, but exact calculations should be performed reproducibly.
A chart or narrative should never silently mix nominal and real values or annual and quarterly periods.
Applied Playbook: Scientific Research
SI can accelerate search, extraction and synthesis, but the researcher should inspect experimental design, sample size, statistical method, replication, uncertainty and whether the conclusions exceed the results.
Applied Playbook: Historical Research
Historical research requires source context and caution about present-day interpretation. SI can organise chronology and compare accounts, but source provenance and period-specific terminology should remain visible.
Applied Playbook: Internal Company Research
Internal research can combine private project files, decisions, metrics and interviews. SI can create continuity across fragmented information, but permissions and confidentiality should be respected.
Internal evidence should distinguish official state from personal notes and recollection.
Applied Playbook: Operations Research
Use process data, queues, incident history, handoff records and performance metrics to identify repeated causes and bottlenecks. SI can cluster incidents and summarise patterns.
Root-cause conclusions should remain evidence-based and should not assign blame from correlation alone.
Applied Playbook: Strategic Research
Strategy research should connect evidence to choices. Map market, capability, competitor, regulatory and operational evidence to scenarios and decision thresholds.
SI can generate alternative futures, but the organisation should own the assumptions and risk appetite.
Applied Playbook: Research for Writing
When research supports a report or proposal, maintain a source table behind the document. Draft from that table rather than directly from search snippets.
This creates a clean transition into the Professional Writing workflow.
The Research Decision Matrix
- Evidence strong, consequence low: act with ordinary review.
- Evidence strong, consequence high: act through the appropriate decision authority.
- Evidence weak, consequence low: run a bounded experiment or gather more data.
- Evidence weak, consequence high: delay, narrow or escalate rather than force a conclusion.
This keeps research proportionate to both evidence and decision consequence.
The Research Recommendation Test
A recommendation should state which evidence supports it, which assumptions it depends on and what new evidence would change it.
SI can help generate the reversal conditions, which often expose hidden assumptions.
The Research No-Decision Outcome
Sometimes the correct output is “evidence is insufficient to decide.” This is a legitimate research result.
Super Intelligence should not be rewarded for always producing a confident recommendation.
The Research Update Trigger
- New authoritative dataset released.
- New regulation or guidance takes effect.
- Important study published.
- Product or technology version changes.
- Material market event occurs.
- Previous source corrected or retracted.
- Decision assumptions change.
- New population or geography enters scope.
Update triggers convert one-off research into maintainable organisational intelligence.
The Living Brief Workflow
- Maintain canonical research question.
- Maintain evidence table.
- Monitor approved sources.
- Flag material changes.
- Compare new evidence with prior conclusion.
- Update uncertainty and recommendation.
- Record change date and owner.
The brief should show what changed rather than rewriting everything each time.
The Research Alert Rule
Alert only when new evidence is material to the question, not whenever the topic is mentioned. High-volume generic monitoring creates noise.
The Research Delta Summary
A delta summary answers: what new evidence appeared, how strong it is, whether it changes the conclusion and what action follows.
This is one of the strongest SI uses for recurring research because comparison across versions is information-intensive and highly structured.
The Research Confidence Register
For important conclusions, record confidence and why: source strength, agreement, directness and remaining gaps. The label can be qualitative—low, moderate, high—if the reasoning is explicit.
Do not confuse confidence with certainty.
The Research Assumption Register
Strategic and analytical research often depends on assumptions. Record them separately so later evidence can update the conclusion efficiently.
The Research Reversal Conditions
State what would cause the conclusion to change. A new price threshold, stronger study, regulation change or failure of a key assumption can become a monitoring target.
The Research Review Meeting
For consequential research, review source quality, disagreement, methodology, conclusion and recommendation with the appropriate domain experts or decision-makers.
The meeting should challenge the evidence rather than merely approve the prose.
The Research Peer-Review Prompt
Ask SI to simulate different reviewer perspectives—methodologist, skeptic, operator, finance, legal, customer—and list questions they might ask. Use this to prepare human review.
The model-generated objections are prompts for scrutiny, not substitute reviewers.
The Research Reproducibility Pack
- Research brief
- Search terms
- Source register
- Inclusion and exclusion criteria
- Evidence table
- Extraction schema
- Analysis notes
- Key calculations
- Synthesis draft
- Final conclusion
- Evidence cutoff date
A second researcher should be able to follow the pack and understand how the conclusion was reached.
The Research Handoff to Writing
Before drafting, freeze the evidence table for the version of the document being written. Changes to evidence after drafting begins should be logged and assessed deliberately.
The Research Handoff to Data Analysis
If sources produce structured data, move calculations and charts into reproducible spreadsheet or analytical workflows. The narrative should read from those outputs rather than manually retyping figures.
The Research Handoff to Planning
A research conclusion may create a plan. Separate what the evidence suggests from the schedule, resources and commitments required to implement it.
The Research Handoff to Decision
The decision-maker should receive enough evidence to challenge the conclusion without repeating the whole research process.
The Research Handoff to Knowledge
Recurring evidence, definitions and decisions should move into maintained organisational knowledge where they can reduce future research cost.
The Research-Human Skill Boundary
Researchers still need enough domain knowledge to recognise implausible claims, weak methods and missing variables. SI can accelerate the work but does not remove the need for methodological literacy.
The Research Learning Loop
Track where SI extraction or synthesis repeatedly fails. Update the schema, source rules or review checklist instead of correcting the same issue indefinitely.
The Research Cost Model
Include human review, source access, software, agent time and maintenance. Compare with time saved, improved evidence coverage and decision value.
Research that becomes faster but much harder to verify may not be cheaper overall.
The Research Queue Model
Teams can maintain queues for questions to investigate, sources to review, gaps to resolve and briefs to update. SI can triage these queues by decision relevance and deadline.
The Research Portfolio Review
Periodically ask which recurring research products are actually used, which duplicate one another and which should be retired. Cheap generation can create report sprawl.
The Research Retirement Rule
Retire living briefs when the question is no longer decision-relevant. Archive them with an evidence cutoff so they are not mistaken for current analysis.
The Research Quality Checklist
- Question is precise.
- Scope and decision use are explicit.
- Search strategy is broad enough.
- Authoritative sources are included.
- Primary sources are used where important.
- Provenance is preserved.
- Definitions match across comparisons.
- Population and dates are recorded.
- Method limits are understood.
- Contradictions remain visible.
- Null or adverse evidence has been sought.
- Material claims trace to sources.
- Uncertainty is explicit.
- Conclusion fits the evidence.
- Evidence cutoff date is visible.
The Research Agent Checklist
- Objective is bounded.
- Approved source classes are defined.
- Search log is preserved.
- Duplicate sources are detected.
- Extraction schema is explicit.
- Missing data is not guessed.
- External content cannot change trusted instructions.
- Tool access is scoped.
- Provenance is attached.
- Stop conditions are defined.
- Human review occurs before consequential conclusion or publication.
The Research Decision Checklist
- Is the evidence strong enough for the consequence?
- Are opposing findings represented?
- Are key assumptions visible?
- Would a different population change the answer?
- Is the conclusion still current?
- What evidence would reverse the recommendation?
- Who has authority to act?
The Final Research Operating Standard
A mature SI research workflow produces more than a summary. It produces a traceable evidence state: question, sources, methods, findings, disagreements, gaps, synthesis, uncertainty and update date.
That state can then support professional writing, data analysis and planning without forcing the next user to trust a black-box answer.
The Final Research Rule
Research with Super Intelligence should increase both the speed of evidence processing and the visibility of the evidence trail.
When those two goals move together, SI makes the researcher more capable. When speed rises while provenance disappears, the workflow becomes faster but less trustworthy.
Research Failure Mode: The Answer-First Workflow
The researcher begins with a desired conclusion and uses SI to gather supporting evidence. This produces efficient confirmation bias.
Repair by writing the question, scope and decision criteria before searching. Run a negative search and require opposing evidence where available.
Research Failure Mode: Search-Snippet Research
The workflow relies on snippets or generated summaries without opening the underlying sources. Important qualifiers, dates and methods disappear.
Repair by requiring source retrieval for every material claim and preserving provenance in the evidence table.
Research Failure Mode: Citation Laundering
A secondary article cites another source, which cites a third source, but the workflow treats the secondary article as proof. Repetition creates an illusion of independent confirmation.
Repair by tracing citation chains to the original evidence where practical.
Research Failure Mode: Evidence Inflation
The system collects many low-quality sources and the final brief looks comprehensive. Volume creates false confidence.
Repair by ranking relevance and quality, then writing the synthesis from the strongest evidence rather than from source count.
Research Failure Mode: Silent Scope Shift
The question begins with one population, period or geography and gradually expands because adjacent evidence is easier to find.
Repair by maintaining the scope ledger and flagging when evidence falls outside it.
Research Failure Mode: Method Flattening
A survey, experiment and anecdote are summarised as though they offer the same type of evidence.
Repair by recording method and the inference each method can legitimately support.
Research Failure Mode: Hidden Missingness
The model fills gaps in an evidence table with plausible values or implied categories. The table becomes cleaner and less truthful.
Repair with the no-guess rule and explicit missing-value states.
Research Failure Mode: Lost Minority Evidence
Clustering compresses uncommon viewpoints into a dominant theme. Important edge cases disappear.
Repair by preserving outliers when they may affect risk, design or fairness even if they are infrequent.
Research Failure Mode: Overconfident Synthesis
The final prose becomes more certain than the sources. Words such as proves, causes or always appear despite weaker evidence.
Repair with claim-type tagging and confidence language matched to method.
Research Failure Mode: Stale Living Brief
A brief is treated as current long after the source environment changed.
Repair by displaying evidence cutoff, owner and update trigger prominently.
Research Failure Mode: Research Without Decision Use
The organisation produces reports that no decision or learning process consumes.
Repair by connecting each research product to a reader, decision, maintained knowledge asset or explicit learning objective.
Research Failure Mode: Endless Search
Because SI lowers search cost, the team keeps expanding the corpus without changing the conclusion.
Repair with a stop rule based on diminishing decision value.
Research Failure Mode: Model-as-Expert Substitution
The system’s synthesis is treated as professional authority in law, medicine, finance, safety or another specialist domain.
Repair by using SI for preparation while retaining qualified domain review where required.
The Research Review Matrix
- Low consequence + familiar domain: light source verification may be sufficient.
- Low consequence + unfamiliar domain: stronger source and terminology checks.
- High consequence + strong direct evidence: qualified review and explicit uncertainty.
- High consequence + weak or conflicting evidence: preserve uncertainty and avoid forced recommendation.
- Fast-changing topic: currentness and monitoring become central.
- Regulated topic: jurisdiction and applicable professional controls become central.
The Research Red-Team Pass
Before finalising important research, ask a reviewer or SI critic to attack the conclusion. Which sources are weak? Which population mismatch matters? Which assumption is hidden? Which contradictory evidence was excluded? What alternative conclusion could fit the same facts?
The goal is not to create false balance. It is to find weaknesses that would matter to a reasonable skeptical reader.
The Research Blue-Team Pass
After the red-team critique, strengthen the evidence where possible and explain why remaining objections do or do not change the conclusion.
This produces a more defensible synthesis than merely adding caveats at the end.
The Research Expert Review Packet
- Research question
- One-paragraph conclusion
- Top five sources
- Evidence table
- Main contradictions
- Key methodological limits
- Assumptions
- Unresolved questions
- Decision implication
The packet lets a domain expert focus on the parts that require expertise instead of rereading the entire search history.
The Research Decision Owner
Research can inform but does not automatically own the resulting decision. Record who has authority to act on the evidence and what additional factors they must consider.
This is especially important when research crosses from evidence into policy, hiring, finance, health or strategy.
The Research Decision Log
When a decision is made from the research, record which conclusion was used, which evidence mattered, what assumptions were accepted and what would trigger reconsideration.
The log creates a return path when new evidence appears.
The Research-to-Action Gap
A strong research system should make it easy to see what action follows and what evidence is still required. Research can be correct yet operationally unused if the handoff into decision-making is weak.
The Research-to-Experiment Path
When evidence is weak but the consequence allows learning, convert uncertainty into a bounded experiment. Define hypothesis, measure and stop condition.
This can be better than continuing desk research indefinitely.
The Research-to-Monitoring Path
When the answer depends on a future change, stop researching the present and define the condition to watch: regulation issued, price threshold crossed, study published or product capability released.
SI can then monitor a narrow condition rather than repeat a broad search.
The Research-to-Knowledge Path
If the conclusion is likely to recur, convert it into maintained organisational knowledge with source links and an update date.
Do not leave durable knowledge only inside a one-off report.
The Research-to-Training Path
When research reveals a recurring conceptual gap among employees, convert the evidence into a training module or reference guide. SI can help adapt the same evidence for different roles.
The Research-to-Product Path
User and market research may create product decisions. Preserve the original evidence and user language so downstream teams can distinguish observed need from internal interpretation.
The Research-to-Policy Path
Research supporting policy should show scope, source strength, affected groups, trade-offs, uncertainties and implementation implications.
SI can draft the bridge, but policy authority remains with the appropriate governance process.
The Research-to-Finance Path
Research that affects budgets or forecasts should transfer structured assumptions and data into reproducible financial models rather than rely on prose alone.
The Research-to-Operations Path
Operational research should identify which process state, threshold or handoff should change. Conclusions without an operating lever may not improve the system.
The Research Update Review
When new evidence arrives, do not rewrite the entire synthesis automatically. First ask whether the new source changes confidence, scope or recommendation.
A material-change gate prevents living briefs from becoming noisy news feeds.
The Research Monitoring Dashboard
- Research question
- Evidence cutoff
- Last material update
- Current conclusion
- Current confidence
- Open gaps
- Monitored sources
- Reversal conditions
- Owner
- Next review date
A simple spreadsheet can manage this for a small portfolio.
The Research Portfolio Prioritisation
Rank recurring research by decision impact, uncertainty, update frequency and cost of being wrong. High-impact fast-changing questions deserve more active monitoring than stable background topics.
The Research SLA
Some questions require answers in hours; others justify weeks of evidence gathering. Define the decision deadline so the research depth matches available time.
SI can accelerate work, but it cannot create evidence that does not exist.
The Research Triage Rule
- Known answer, current source: retrieve and cite.
- Known method, new data: update the evidence table.
- New question, low consequence: rapid bounded research.
- New question, high consequence: deeper review and specialist input.
- Future-dependent question: monitor condition.
- Unresolvable with available evidence: state uncertainty.
The Research Automation Boundary
Automate repetitive discovery, extraction, deduplication and update checks. Keep source judgment, methodology, interpretation and consequential recommendation under meaningful human review.
The Research Agent Permission Boundary
A research agent may need web access, internal files or databases. Grant only the sources required for the question and separate research access from unrelated operational tools.
A research agent usually does not need permission to send messages, edit customer records or take other business actions.
The Research Agent Memory Boundary
Persistent memory can help continuity but may preserve stale assumptions. Important conclusions should be reconstructed from maintained evidence, not invisible memory alone.
The Research Agent Time Boundary
Long-running agents should checkpoint scope and currentness. The original question may change while the agent is still searching.
The Research Agent Cost Boundary
Deep autonomous search can consume significant compute or paid source access. Set limits based on decision value and stop when marginal evidence becomes small.
The Research Agent Transparency Boundary
The system should show which sources were searched, which were included, which failed and why. Hidden search history makes audit and reproduction difficult.
The Research Human-Agency Rule
Researchers should remain able to change scope, reject sources, correct extraction, weaken conclusions and stop the process. SI should increase research capacity without making the human a passive recipient of synthesis.
The Research Training Checklist
- Can the user formulate a bounded question?
- Can the user distinguish source types?
- Can the user judge whether a citation supports a claim?
- Can the user recognise method limitations?
- Can the user preserve uncertainty?
- Can the user identify currentness needs?
- Can the user challenge an SI-generated conclusion?
These skills are more durable than memorising one research prompt.
The Research Quality-Control Checklist
- Question matches output.
- Scope remained stable.
- Search covered key source classes.
- Primary sources were checked for material claims.
- Duplicate citation chains were removed.
- Extraction was validated.
- Definitions were normalised.
- Population and time differences were preserved.
- Contradictions were analysed.
- Disconfirming evidence was sought.
- Conclusion language matches method.
- Citations support nearby claims.
- Evidence cutoff is visible.
- Decision implication is explicit.
The Research Publication Checklist
- Material claims are source-grounded.
- Quotes are accurate and necessary.
- Numbers and units are verified.
- Jurisdiction and population are explicit.
- Uncertainty is not hidden.
- Conflicts of interest are considered where relevant.
- Sensitive or private sources are handled appropriately.
- Conclusion does not exceed evidence.
- Update date or evidence cutoff is shown.
The Research Final Review Questions
- What is the strongest evidence?
- What is the strongest evidence against the conclusion?
- What assumption matters most?
- What population is not represented?
- What would change the recommendation?
- What remains unknown?
- What decision does this research support?
- When will the answer become stale?
The Research Standard
A research system is ready when another competent person can inspect the evidence path and understand why the conclusion exists.
Super Intelligence should make that path shorter to build, not harder to see. The speed of synthesis matters, but the visibility of evidence, scope and uncertainty matters more.
The Research Audit Trail
For important research, preserve the path from question to conclusion. The audit trail should make it possible to reconstruct what was searched, which sources were included, what extraction changed, how contradictions were handled and when the evidence cutoff occurred. This does not require saving every exploratory thought. It requires enough evidence that another competent person can see why the final claim exists.
The Research Change Ledger
Maintain a short ledger for material changes: a new source, corrected statistic, revised definition, changed jurisdiction, new methodology, retracted paper or altered conclusion. Living research can drift silently if the evidence base changes without a visible record. A ledger makes downstream writing and decisions easier to update because the team knows exactly what moved.
The Research Handoff Standard
- Question and scope
- Short answer
- Strongest supporting evidence
- Strongest contradictory evidence
- Important methodological limits
- Open gaps
- Evidence cutoff date
- Recommended next action
- Owner and next review date
A handoff built this way lets a writer, manager or decision-maker use the research without accepting a black-box conclusion. The receiver can see what is known, how strongly it is known and where further investigation may still be required.
The Research Receiver Test
Ask the intended receiver whether they can identify the strongest evidence, the most important uncertainty and the action implication without reopening the entire search process. If not, the synthesis is still too diffuse. Good research compresses complexity without hiding the route back to evidence. The receiver should be able to drill down only where the decision requires it.
The Research Independence Test
Have another competent person reproduce one or two material claims from the evidence pack. Can they locate the source, understand the definition, identify the population and see how the conclusion follows? If not, the workflow may still depend too heavily on hidden researcher context. Independence is a useful sign that the research belongs to the organisation rather than one person’s private conversation with SI.
The Research Freshness Test
Before reuse, ask whether the topic can change faster than the brief is updated. Fast-moving software, regulation, markets, security and public policy require stronger currentness control than stable historical or conceptual questions. A well-researched answer can become wrong because the world changed. Show the evidence cutoff and define the event that should trigger reassessment.
The Research Peer-Review Handoff
When another researcher or domain expert reviews the work, give them the evidence table rather than only the final prose. Review should focus on source quality, missing variables, method limits, causal claims and whether the synthesis gives appropriate weight to competing evidence. Super Intelligence can prepare a review checklist and highlight claims with the weakest source support, allowing human reviewers to concentrate attention where it matters most.
The Research Decision-Use Test
Before closing the project, ask the intended decision-maker which parts of the research actually change the decision. If no evidence affects the choice, the research may be overbuilt or disconnected from the real question. This test can also reveal when the evidence is sufficient even though more sources remain available. Decision relevance is a legitimate stopping condition.
The Research Reuse Test
Ask whether the work created reusable assets: a source register, extraction schema, evidence table, search vocabulary, monitoring rule or decision framework. Reusable assets make the next related question cheaper and more consistent. If all learning remains inside one final report, the organisation may repeat the same discovery and validation work later.
The Research Maintenance Test
If the brief will be reused, assign an owner and a review trigger. If it will not be maintained, mark the evidence cutoff clearly and archive it as a dated analysis rather than allowing readers to assume it remains current indefinitely. A maintained conclusion and a historical conclusion are different artefacts and should be labelled accordingly.
The Research Governance Test
For high-impact internal research, confirm who owns the question, who may access the sources, who reviews sensitive or regulated conclusions and who can approve publication or action. A capable research agent does not automatically receive authority over how its synthesis is used. Governance should follow consequence, data sensitivity and professional responsibility.
The Research Privacy Test
Where research uses personal, customer, employee or confidential information, minimise the data supplied to SI and follow the approved environment and access rules. The fact that a field might improve context does not automatically justify using it. Research should use the smallest information set that can answer the question reliably.
The Research Security Test
External websites, PDFs and documents should be treated as untrusted content when tool-using research agents are involved. Retrieved text can inform the research but should not override trusted instructions, reveal unrelated secrets or trigger unauthorised actions. Research access and operational action authority should remain separate wherever possible.
The Research Cost-to-Decision Test
Compare total research effort with the value and reversibility of the decision. A low-consequence choice may need only a rapid evidence scan. A major investment, public claim or professional conclusion may justify deeper review. Super Intelligence should make proportional research easier, not encourage exhaustive research for every question simply because search and synthesis are cheaper.
The Research Update Trigger
Define the conditions that require the brief to be reopened: a new authoritative dataset, material product release, regulatory change, retraction, major market event or failure of a key assumption. SI can monitor those triggers, but a human owner should decide whether the new evidence actually changes the conclusion.
The Research Portfolio Test
Teams with many active research questions should periodically ask which products are used, which overlap, which are stale and which deserve monitoring. Cheap synthesis can create report sprawl. A smaller set of maintained decision-relevant briefs is often more useful than a large archive of one-off generated analyses.
The Research Closure Rule
Research closes when the question has enough trustworthy evidence for its intended use, not when search results stop appearing. Super Intelligence makes more searching cheap, so the workflow needs an explicit stopping condition. The researcher should be able to state: this is what the evidence supports, this is what remains uncertain, this is the decision or next step, and this is when the conclusion should be revisited.
The Research Final Operating Standard
A mature SI research workflow is fast, reproducible and humble about uncertainty. It expands source coverage, reduces extraction burden and sharpens comparison while keeping provenance, scope and method visible.
Machine intelligence should accelerate the path through evidence without becoming the evidence itself. That boundary is what makes research both faster and more trustworthy, and it allows the next document, analysis or decision to inherit the evidence rather than only the model’s summary.