The GANTIÈRE Method

Research the find.Show the evidence.

The method is designed for the exact point where art history, connoisseurship, marketplace noise, signatures, provenance, and incomplete records collide.

Research the find. Show the evidence. Know when not to guess.

Research discipline

Three rules before any answer becomes a conclusion.

Object before story

Observation comes before attribution, provenance narrative, or marketplace language.

Identity before market

Price context waits until the object and maker hypotheses are strong enough to support comparison.

Contradictions stay visible

Conflicting signatures, dates, sources, and exact-image evidence remain part of the case until reconciled.

Observe

Start with the object before the story: photographs, dimensions, materials, construction, marks, labels, inscriptions, surface, damage, and anything visible on the reverse.

Identify

Build artist, maker, period, technique, edition, and object hypotheses without promoting the first plausible answer to fact.

Compare

Find canonical works, institutional references, signature examples, exact-composition matches, and competing identities. Similar is not the same.

Reconcile

Resolve contradictions between listings, auction records, museum data, signatures, technique descriptions, dates, and provenance claims instead of silently choosing one.

Place

Separate sold results from asking prices, compare relevant works, preserve dates and currencies, and make market context conditional on identity.

Report

Show whether each claim is verified, inference, estimate, or unknown; keep competing and contradicted claims visible; and explain why the final confidence band is earned.

Case model

What the object shows is not the same thing as what we conclude.

GANTIÈRE keeps observations, conclusions, evidence references, source references, and applicability separate so a visible fact cannot quietly become an attribution.

Observation

A direct description of this object: descriptive, this-object scoped, and anchored to direct user evidence. An observation is not labeled inference or estimate.

Conclusion

An identity, attribution, provenance, date, market, or external-context claim carries its own status, basis, and evidence/source references.

Applicability

A source remains marked as exact-object evidence, reference context, market comparable, or unknown so relevance is not assumed merely because a source exists.

Signature forensics

A signature is evidence. It is not a shortcut.

The comparison stays evidence-first: several readings can remain alive, reference independence matters, documented variation is allowed, and a real mismatch can defeat an otherwise attractive attribution.

Multiple readings

Ambiguous letters keep more than one reading alive. The first plausible name is a hypothesis, not a locked identity.

Independent references

Qualifying signature evidence should come from at least two independent reference objects. Cross-listed records of the same physical object count once.

Period & medium

Signature variation can be grouped by period and medium when variation is documented or plausible rather than forcing one timeless master signature.

Mismatch matters

A material signature mismatch is high-value negative evidence and cannot be overridden by style, dimensions, edition number, provenance story, or marketplace metadata.

Similarity is not authentication

Signature similarity can support attribution, but it cannot authenticate the physical object by itself.

Unresolved is allowed

When strong reference imagery is unavailable, the comparison stays unresolved instead of manufacturing certainty.

Canonical work matching

Retrieval can be broad. Identity stays conservative.

Metadata is useful for finding candidates. It never gets to skip the direct visual comparison, and even an exact composition still does not prove that two records describe the same physical object.

Metadata retrieves; images adjudicate

Artist, medium, dimensions, signature location, title, date, provenance narrative, and stylistic similarity may retrieve or rank a candidate, but cannot establish exact identity by themselves.

Mismatch stops the match

A direct visual composition mismatch rejects exact identity. Metadata agreement cannot overrule the image.

Similar is not exact

Stylistic or compositional similarity remains similar-work evidence until direct visual comparison verifies the exact composition.

Composition is not object identity

For editions, casts, prints, repeated compositions, and other multiples, the same composition does not prove the same physical object.

Same-object needs object-history evidence

Exact-object status requires a separate verified this-object authentication/provenance claim with qualifying exact-object source support from an object-history-capable source.

Interested listings cannot unlock provenance

Active listings and dealer asking records may locate candidates or describe market context, but they cannot establish same-object provenance continuity.

Source context

Evidence keeps its source class.

GANTIÈRE does not flatten every link, photograph, email, and sale record into the same kind of proof. Source context stays attached so a claim can be judged for what the evidence actually supports.

Object evidence

Front, verso, detail photographs, dimensions, condition, labels, inscriptions, materials, and construction describe the specific object being researched.

Institutional records

Museum, archive, foundation, catalogue, and collection records can anchor artist identity, documented works, exhibitions, techniques, and historical context.

Market records

Auction, dealer, and marketplace records can supply comparison and transaction context, but asking, estimate, and sold evidence remain distinct.

Correspondence

Estate, archive, institution, expert, and owner correspondence can clarify claims later in the case and is preserved with its date, source, and context.

Claim-specific source quality

A strong source can still be the wrong source for the claim.

GANTIÈRE evaluates source authority and claim applicability separately. The question is not simply whether a source looks prestigious, but whether it actually supports this claim about this object.

Authority is not applicability

A source can be highly authoritative about what it records and still be insufficient for the specific claim being tested.

Institutional is not universal proof

An institutional reference can be primary evidence for its catalog record without automatically becoming market-price or authentication evidence.

Exact-object support is stronger

Exact-object applicability can directly support this-object claims; reference-context material stays contextual unless the claim is external context.

High-risk context is capped

For identity, authentication, provenance, market, and date/origin claims about this object, contextual references cannot be promoted beyond low support merely because the source is prestigious.

Market context stays context

A market-comparable source can support a market claim as context, but its support is capped rather than treated as direct object proof.

Ordinal, not pseudo-precise

Source support is expressed as high, moderate, low, or unknown. The system does not invent a fake percentage for source quality.

Evidence is not a license

A source can support research without granting GANTIÈRE the right to reproduce its image or content. Provider terms and reproduction rights remain a separate boundary.

Market discipline

Price evidence waits behind identity.

A high price does not make the identity stronger. A comparable first survives identity and relevance checks before any realized sale can enter the valuation lane.

Identity gate

Critical identity mismatch excludes a comparable. Incomplete critical identity keeps it in market context rather than promoting it into valuation evidence.

Verified sold

A source-verified completed sale is only a realized-comparable candidate until identity and relevance gating prove comparability.

Asking / dealer asking

Active asking prices remain market context. They are not silently treated as completed transaction evidence.

Estimate

Auction or market estimates remain context only because an estimate is not a realized transaction price.

Reproduction

Copy or reproduction evidence is excluded from original-object valuation and retained only as explicit market context.

Provenance & traceability

Provenance is a claim, not atmosphere.

Names, places, labels, exhibitions, and ownership stories can matter. They only become provenance when evidence ties them to the specific object and the claim can be traced back to its source.

Exact-object evidence

A verified provenance claim about this object requires an exact-object source. Visual resemblance or general artist history cannot substitute for that link.

Citation trace

Saved claim provenance keeps the source registry, citation spans, and claim envelope together so important statements can be traced back to supporting material.

Incomplete chain stays incomplete

A label, city, exhibition, collection name, or ownership clue can be relevant without becoming a complete chain of custody. Missing links remain visible.

Context is not ownership

An artist having worked, exhibited, or travelled in the same place does not establish that a specific object passed through a particular owner or collection.

Trust rules

The system is designed to resist a confident wrong answer.

These rules protect the research trail when signatures are ambiguous, compositions repeat, and later evidence changes the direction of a case.

No first-reading lock-in

An ambiguous signature keeps multiple readings alive until external references and the object itself narrow the field.

No composition shortcut

A matching composition, edition, or listing image never substitutes for signature, object, and source verification.

No silent overwrite

New evidence reconciles with the previous case state. Earlier conclusions, contradictions, and rejected leads remain part of the record.

Guided evidence capture

Ask for the photograph that changes the case.

GANTIÈRE can compare the views already supplied with the evidence goals of the research and request the missing angle or detail instead of simply saying that more information is needed.

Guidance stays bounded: at most six requests at a time, with priority, reason, capture instruction, and the evidence goals that the photograph can help resolve.

Signature or inscription

Required

Capture close and sharp without glare.

Identity · marks · dating

Full reverse

Recommended

Capture the full reverse straight on.

Baseline · construction

Label or sticker

Required

Capture the entire label and one close detail.

Provenance labels · identity · dating

Underside or base

Required

Capture the entire base square and in focus.

Identity · marks · dating · construction

Scale reference

Required

Photograph the object with a neutral ruler or known scale.

Physical scale

Condition detail

Required

Capture each issue close and in context.

Condition

Hidden construction and mechanical components are requested only when relevant, with capture guidance that explicitly avoids disassembly.

Execution context

The report remembers the context in which it was researched.

Market locale, currency, target country, language, and evidence availability can affect what a research run sees. GANTIÈRE keeps that context with the version instead of hiding it behind invisible defaults.

Language & market

Preferred language, display currency, and target country stay attached to the research version so later review knows the context in which the report was produced.

Marketplace context

Marketplace ID and locale are versioned rather than treated as invisible defaults. A different marketplace can change which evidence is available without changing the object itself.

Unavailable is not absent

If marketplace or institutional evidence is temporarily unavailable and web research continues, that state is preserved instead of being rewritten as no evidence found.

Not requested is explicit

Evidence that was not requested remains distinguishable from evidence that was requested but unavailable, and from an institutional search that returned no records.

Context changes are diffable

When execution context changes between versions, the version diff records that change alongside source, image, claim, and report changes.

Research footprint

The answer shows how much research actually happened.

GANTIÈRE records the search footprint instead of hiding behind phrases like “extensive research.” Source-pool breadth, direct citations, and structured provider evidence remain separately visible.

Source URLs returned

The research run records how many source URLs were returned by live web research rather than implying an undefined amount of searching.

Distinct domains

The footprint also records how many different domains appeared in the research pool so breadth is visible without pretending that domain count equals source quality.

Direct citations

Sources cited directly in the answer are counted separately from the larger research pool that informed discovery and cross-checking.

Provider footprint

When marketplace or institutional evidence participates, the footprint can report the number and class of those records without upgrading them into stronger evidence than they are.

Server-owned trust layer

The model can propose. It cannot promote itself.

Trust-critical derived fields are rebuilt at the public server boundary. If a supporting classifier or comparability layer degrades, the safe outcome is less certainty or an unavailable layer — not a stronger claim.

Provider output is not the final trust state

Model/provider output can supply research material or relation proposals, but trust-critical evidence semantics are finalized on the server.

Server-owned fields overwrite inbound copies

At the public gateway, guided photography, claim-provenance attestation, source quality, confidence, and market comparability are overwritten with server-owned results rather than trusted from upstream payload fields.

Raw relation proposals stay internal

Confidence relation candidates are deterministically finalized or discarded. Raw provider relation candidates never cross the public boundary.

Degraded layers fail downward

If confidence classification, source-quality projection, or market comparability cannot complete safely, that layer becomes degraded or unavailable instead of manufacturing stronger certainty.

Primary research survives additive-layer failure

Comparability and other trust layers are additive. Unexpected failure in an additive layer must not turn otherwise-successful primary research into a false negative or a server error.

Evidence basis is revalidated

Source quality is computed only after the claim-provenance tuple is revalidated against the result, source registry, citation spans, claim envelope, and extraction metadata.

Versioned research

Re-research extends the case. It does not erase it.

New evidence can justify a new conclusion without pretending the earlier state never existed. Each version remains addressable, and the system can explain exactly what changed between them.

Immutable version

Each saved research version keeps its own report, sources, execution context, claim provenance, guided-photo state, and image paths.

Expected current version

Re-research names the version it expects to extend. A stale request does not get to silently rewrite a newer case state.

Evidence membership

Retained images are trusted only after exact membership is checked against the current validated version; new evidence gets new versioned paths.

Visible diff

The change record can show added/removed sources and images, added/removed claims, status or basis changes, source-support changes, direct-evidence changes, and guided-photo changes.

Confidence is earned

The answer can stay unresolved.

GANTIÈRE should prefer a useful uncertainty over a polished false answer. Exact-image mismatch remains negative evidence. Signature mismatch can defeat a stylistic hunch. Market value waits behind identity. Provenance never appears merely because an artist travelled through the same city.

Claims: verified · inference · estimate · unknown

Confidence: supported · competing · contradicted · unknown

Supported

Available evidence materially favors the claim without erasing its remaining limits.

Competing

More than one viable explanation remains and the case keeps them visible.

Contradicted

Material evidence conflicts with the claim and must remain attached to the record.

Unknown

The available evidence does not justify a stronger confidence state.

Ready when the object is

Real objects. Real answers. Anywhere you are.

Research access not yet openArtist Library