Evidence before assertion
Important conclusions should remain connected to the information supporting them.
IN PRACTICE — Visible source identity, source route/viewer, evidence relationship, and provenance state.
RESEARCH / TALVRIN METHODOLOGY
How Talvrin keeps
research connected
to the evidence.
Start with a question. Discover and inspect the underlying sources. Preserve the dates, jurisdictions, versions, rights and relationships that affect meaning. Build a research view. Keep it connected to the evidence. Reassess when material information changes.
AI can assist discovery, organization, comparison, summarization, explanation and change identification. It does not become the evidence. Research judgment remains human.
Research and market intelligence. No legal advice. No trade execution. No manufactured buy/sell/hold recommendations.
EVIDENCE BEFORE ASSERTION
SOURCE BEFORE SUMMARY
CONTEXT BEFORE CONFIDENCE
TRANSPARENCY BEFORE PERSUASION
AI ASSISTANCE, NOT AI AUTHORITY
COVERAGE BEFORE CLAIMS
JUDGMENT REMAINS HUMAN
SEVEN PRINCIPLES
The methodology in one sentence.
Connect the question to the evidence, the evidence to its context, the context to a research view, and the research view to whatever changes the evidence next.
Source before summary
A convenient explanation must not erase the underlying source.
IN PRACTICE — Source access precedes or accompanies interpretation; source label never hidden behind AI text.
Context before confidence
Dates, jurisdictions, versions, periods, rights, and relationships can materially change meaning.
IN PRACTICE — Publication time, effective/reference period, jurisdiction, version/supersession, and rights state where material.
AI assistance, not AI authority
AI may accelerate research without becoming the authoritative evidence source.
IN PRACTICE — Persistent provenance treatment and separate labels for generated/model-assisted output.
Coverage before claims
Talvrin should describe what is actually supported, not theoretical capability as live coverage.
IN PRACTICE — Coverage registry/status pattern and explicit limitations.
Judgment remains human
Talvrin can improve the information environment; users remain responsible for decisions.
IN PRACTICE — No prescriptive investment recommendation language or "decision certainty" claims.
RESEARCH WORKFLOW
Ask → Discover → Inspect →
Understand → Build → Monitor
→ Reassess.
The seven-step process a research object moves through, from a framed question to a monitored, revisit-able view.
01 Ask
USER JOB
Frame a market, issuer, security, event, policy issue, or research question.
UI / CONTENT BEHAVIOR
Preserve the question as the research object context rather than reducing it to a keyword query.
02 Discover
USER JOB
Find relevant evidence and context.
UI / CONTENT BEHAVIOR
Return governed source candidates with source identity and scope, not only synthesized answers.
03 Inspect
USER JOB
Review the underlying sources.
UI / CONTENT BEHAVIOR
Open source or governed viewer; show publication timing, version, jurisdiction, and rights state where relevant.
04 Understand
USER JOB
See how evidence relates to the question.
UI / CONTENT BEHAVIOR
Expose supporting, contradicting, updating, or contextual relationships without a hidden black-box score.
05 Build
USER JOB
Develop and preserve a research view.
UI / CONTENT BEHAVIOR
Keep evidence, interpretation, assumptions, and user notes distinguishable and revisit-able.
06 Monitor
USER JOB
Continue watching relevant evidence and assumptions.
UI / CONTENT BEHAVIOR
Surface meaningful evidence changes separately from market-price alerts or generic notification noise.
07 Reassess
USER JOB
Return when new evidence changes the basis of the view.
UI / CONTENT BEHAVIOR
Show what changed, when, where, and which prior assumption/evidence relationship may need review.
EVIDENCE ANATOMY
Every piece of evidence carries the
same fields.
Provenance and context fields are preserved on every record — never inferred, never invented.
Source identity
Name the organization, publisher, issuer, regulator, central bank, exchange, or other governed source; never use only "web source."
Source class
Primary / official / licensed / institutional / other governed classification only where supported by the source registry.
Original title
Preserve human-readable source title or document/event name.
Publication time
Display with timezone where material.
Effective / reference period
Separate from publication time when the fact applies to a different period.
Jurisdiction / market
Expose when legally, economically, or market-structurally relevant.
Version / supersession
Identify revision, replacement, withdrawal, or supersession where supported.
Rights / access state
Respect licensing, entitlement, redistribution, and permitted-use rules.
Evidence relationship
Explain whether the evidence supports, contradicts, updates, or contextualizes the research object.
Open source action
Deep link or governed viewer route when permitted; do not fabricate endpoints.
NO SCORE INVENTION
The methodology does not manufacture a numerical confidence score, source-quality score, reliability percentage or fixed weighting model. A future governed scoring model would require its own approved methodology contract before exposure.
INTERPRETATION LAYERS
Source, normalization, analysis, AI
and notes stay visually distinct.
Every layer carries its own ownership and its own required label — never styled as if it were the source itself.
Source evidence
External or governed source; authority depends on source/context.
Source / Official source /
Licensed source
Licensed source
Talvrin normalization
Talvrin structured representation of source material.
Normalized / structured by
Talvrin
Talvrin
Talvrin analysis
Talvrin-authored or product-generated interpretation.
Talvrin analysis /
interpretation
interpretation
AI-assisted output
Model-assisted content under Talvrin policy controls.
AI-assisted; evidence links
remain visible
remain visible
UNCERTAINTY & CONTRADICTION
Explaining uncertainty beats
manufacturing certainty.
Conflicting evidence stays visible as a research condition rather than being collapsed into one answer.
MONITORING & CHANGE
Research should remain connected to
what can change it.
Evidence-change monitoring is distinct from price alerts or generic notification noise.
NEW EVIDENCE
Show source identity, publication time, relationship to the research object, and whether materiality review is complete.
UPDATED EVIDENCE
Preserve prior version or supersession relationship where available; do not overwrite history invisibly.
UNCHANGED EVIDENCE
Do not create noisy alerts solely to demonstrate activity.
SOURCE UNAVAILABLE
Expose availability/rights state; do not silently remove the evidence trail.
MATERIALITY PENDING
Use a pending/review state where a change exists but its significance is not yet established.
RESEARCH REASSESSMENT
Connect the changed evidence to assumptions or parts of the research view that may need review.
MATERIALITY BOUNDARY
Materiality is a governed product concept. The methodology explains its states; it does not invent a formula or imply that every source update is material.
AI BOUNDARY
AI may help organize and explain
the evidence. It does not become
the evidence.
Search and evidence discovery
An authoritative source
Document/source summarization
A guaranteed fact
Comparison across evidence
Investment advice or a buy/sell/hold recommendation
Research organization
A guarantee of completeness or accuracy
Relationship/contradiction surfacing
An invisible layer whose provenance the user cannot inspect
Explanation in accessible language
A mechanism for erasing uncertainty, jurisdiction, version, or rights context
COVERAGE TRUTH
Architecture-ready is not the same as released coverage.
Live coverage is described as released, not inferred from what the platform is architecturally capable of.
MEANING
METHODOLOGY TREATMENT
Deep Coverage
MEANING
High-confidence production-supported research depth for a named market/domain, where approved.
METHODOLOGY TREATMENT
May be described only when registry-approved.
Supported
MEANING
Production-supported but narrower than deep coverage.
METHODOLOGY TREATMENT
Show with scope/limitations.
Limited / Beta
MEANING
Available with explicit limitations.
METHODOLOGY TREATMENT
Limitation label is mandatory.
Planned
MEANING
Roadmap only.
METHODOLOGY TREATMENT
Do not present as current coverage; expose only if roadmap publication is approved.
Architecture-ready
MEANING
Platform architecture could support the category.
METHODOLOGY TREATMENT
Generally internal; does not create a customer-facing coverage claim.
METHODOLOGY GOVERNANCE
The methodology itself behaves like
governed evidence.
Dated, reviewable, versioned, and explicit about material changes — never a cosmetic freshness timestamp.
Methodology version
Human-readable version identifier tied to an approved publication record.
Published date
Date the current version became public.
Last reviewed
Show only when an actual review occurred; no cosmetic freshness timestamp.
Change summary
Explain material methodological changes in plain language.
Owner
Named internal function/role, not necessarily a public individual, responsible for the methodology.
Required reviewers
Research/editorial, product, legal/compliance, data rights, AI governance, accessibility/SEO as applicable.
Prior versions
Provide access or a summary trail where publication governance permits.
Open dependencies
Flag unresolved methodology/product contracts rather than hiding them.
REVIEW CADENCE
The source documents do not define an exact review cadence. Monthly, quarterly, annual or event-driven review is never claimed as current fact unless separately approved.
APPLIED METHODOLOGY WALKTHROUGH
The method, demonstrated with a
neutral research question.
Illustrative walkthrough. It does not imply current live coverage unless the relevant source/coverage registry approves it, and carries no investment recommendation.
01
Question
"What evidence has changed around a public-market policy or macro question?"
02
Discover
Surface governed source candidates and clearly identify source identity and timing.
03
Inspect
Open source or governed viewer; show original title, publication time, reference period, jurisdiction, version, and rights state where material.
04
Understand
Explain how each source supports, contradicts, updates, or contextualizes the question.
05
Build
Preserve the researcher’s view and assumptions separately from source evidence and AI-assisted interpretation.
06
Monitor
Show new/updated/unchanged/source-unavailable/materiality-pending states without generic alert noise.
07
Reassess
Highlight which assumptions or evidence relationships deserve review when material new evidence arrives.
ANSWER-FIRST FAQ
Frequently asked, answered first.
An evidence-first research method that keeps questions, source evidence, context, interpretation, research views and ongoing monitoring connected so users can reassess a view when material evidence changes.
No. The methodology does not manufacture a numerical confidence score, source-quality score, reliability percentage or fixed weighting model. Evidence remains inspectable and its context, provenance and relationships remain visible.
AI may assist with evidence discovery, summarization, comparison, organization and accessible explanation. AI-assisted output remains distinct from source evidence, with evidence links and provenance kept visible.
Conflicting evidence remains visible as a research condition. Sources can be represented as supporting, contradicting, updating or contextualizing the research question rather than being silently collapsed into one answer.
Publication time describes when a source became public. The effective or reference period describes when the underlying fact, policy, event or measurement applies. They can be materially different and should not be conflated.
Updates and supersession relationships remain part of the evidence trail where supported. Prior versions are not silently overwritten when publication governance permits a version or summary history.
Monitoring keeps the research view connected to meaningful evidence changes. New, updated, unchanged, source-unavailable and materiality-pending states can be surfaced without turning every source change into generic alert noise.
The methodology does not turn research evidence into investment advice or buy/sell/hold recommendations. Comparisons and evidence relationships remain distinct from an investment recommendation.
Coverage is described according to released and registry-approved scope. Architecture-ready capability is not treated as current live coverage, and geographic or market coverage should be stated only where approved.
Where permitted by source rights and access rules, users can open the underlying source or a governed viewer and inspect provenance, publication timing, version, jurisdiction and other relevant evidence context.
RELATED RESEARCH & TRUST
Continue into evidence, trust
and coverage.
Only through approved routes — never a generic "recommended for you" ranking.
Research
Return to the Research hub for source-linked public-market intelligence.
Evidence Standards
How evidence is sourced, classified and presented, in full.
AI Principles
AI boundaries, provenance and human-verification expectations.
Market Coverage
Canonical released-coverage matrix and market status.
Policy & Regulation
See the methodology applied to policy and regulatory evidence.
Build research that remains
connected to the evidence.
Explore how Talvrin can help you discover evidence, inspect sources, build a defensible research view, and monitor what changes next.
Research and market intelligence. No legal advice. No trade execution. No manufactured buy/sell/hold recommendations.