Cross-Domain Evidence Matrix¶
The evidence families differ in observation unit, spatial meaning, temporal meaning, geographic reach, and publication maturity. Their coexistence in one atlas enables comparison; it does not make them interchangeable.
Domain Matrix¶
| Domain | Observation unit | Spatial meaning | Temporal meaning | Publication role | Principal limit |
|---|---|---|---|---|---|
| LandClim | pollen sequence or model-grid context | site and modelled landscape context | sequence- or model-specific | primary palaeoenvironmental context | model grids are not direct sample observations |
| Neotoma | palaeoecological site and dataset | named site coordinates | site-specific chronology and resolution | primary pollen context | coverage and age control vary by site |
| SEAD | environmental-archaeology site or record | archaeological place | heterogeneous archaeological chronology | contextual archaeology | access, normalization, and temporal comparability remain uneven |
| RAÄ | Swedish heritage record or density surface | Sweden-specific registered location | record-specific and often broad | contextual archaeology | national scope and heterogeneous dating |
| boundaries | administrative geometry | product scope and clipping | none | geographic framing | carries no independent scientific weight |
| SMHI SVAR | registered Swedish water body | lake identity and geometry | current registry state | sampling context and lake prioritization | registry presence is not coring suitability |
| AADR | release-versioned human sample metadata | sample location at declared precision | metadata interval or cultural context | direct human aDNA evidence | genotype processing is outside current scope |
| animal aDNA | curated sample linked to project and literature | sample-owned locality and coordinate posture | sample-owned chronology posture | direct evidence when admitted | project recovery and sample support remain uneven |
| fieldwork | dated repository visit | declared visit coordinates | exact visit date | direct visit evidence | one visit is not representative coverage |
Valid Comparisons¶
Cross-domain comparison requires an explicit bridge:
flowchart LR
A["evidence family A"] --> RoleA["role, place, time, precision"]
B["evidence family B"] --> RoleB["role, place, time, precision"]
RoleA --> Question["declared comparison question"]
RoleB --> Question
Question --> Result["qualified comparison or refusal"]Geographic proximity is only one input. A valid interpretation also accounts for temporal overlap, observation unit, source maturity, and whether each layer is direct evidence, context, framing, or decision support.
Comparison Readiness Has Four Boundaries¶
Before comparing two families, evaluate four independent boundaries for each input:
| Boundary | Question | Failure posture |
|---|---|---|
| materialization | are the exact contracted source and normalized artifacts present? | source known, captured only, or normalized authority missing |
| evidence semantics | are object identity, unit, spatial meaning, time meaning, and missingness explicit? | ambiguous, contextual-only, or unresolved claim |
| review | has the required source- or claim-specific review artifact been materialized? | unreviewed current state, even when a retained product exists |
| product admission | does the named product permit this member, role, and precision? | qualified, refused, excluded, deferred, or outside scope |
These boundaries cannot substitute for one another. A retained map demonstrates that a product was materialized; it does not prove that the current normalized authority or source-specific review remains present. Conversely, a complete capture can be valuable for recovery while remaining ineligible for the comparison being requested.
flowchart TD
Input["candidate family member"] --> Material{"contracted authority materialized?"}
Material -->|no| Refusal["bounded refusal or context-only result"]
Material -->|yes| Meaning{"identity, unit, space, and time explicit?"}
Meaning -->|no| Refusal
Meaning -->|yes| Review{"required review materialized?"}
Review -->|no| Refusal
Review -->|yes| Admission{"named product admits role and precision?"}
Admission -->|no| Refusal
Admission -->|yes| Bridge["eligible cross-domain bridge input"]Relationship Contract¶
Cross-domain publication joins records for a declared question; it does not merge their identities or evidence roles.
flowchart TD
Direct["direct evidence record"] --> Bridge["declared bridge"]
Context["context record"] --> Bridge
Framing["scope or boundary"] --> Bridge
Bridge --> Relation["derived relationship"]
Relation --> Claim["qualified result"]
Direct --> DirectAuthority["direct-record authority"]
Context --> ContextAuthority["context authority"]
Framing --> FramingAuthority["scope authority"]The bridge may be a distance band, interval-overlap rule, country predicate, species relation, or fieldwork criterion. Its result is a new derived relation, not new source evidence. Removing the bridge must leave each input record and its authority intact.
Every derived relationship should preserve the identities of its members, their roles and precision, the rule and parameters, the product scope, and an outcome of supported, boundary-sensitive, contextual-only, or refused. A reader can then challenge the relationship without having to challenge the existence of its source records.
Comparison Protocol¶
- State the scientific or operational question without naming the desired outcome.
- Identify the observation unit and governing identity in each family.
- Declare each family's evidence role for this question.
- Normalize spatial and temporal precision only to the weakest supported common level.
- Record source version, scope, missingness, and inclusion rules.
- Decide whether the bridge supports comparison, context, prioritization, or refusal.
- Preserve the member-level inputs and qualifications with the result.
The bridge is part of the result. A distance threshold, temporal-overlap rule, administrative containment, species relation, or sampling criterion must be named rather than implied by a combined map.
Count Units And Denominators¶
Counts are comparable only after their observation units and transformations are declared. The checked-in collection contains several instructive cases:
| Published quantity | Unit | What must not be inferred |
|---|---|---|
| 2,195 SEAD inventory rows | captured inventory row | 2,195 distinct normalized map points |
| 2,172 normalized SEAD records | published site point | complete or uniformly dated archaeology evidence |
| 761,917 RAÄ records | source registry record | equivalent density or recording effort across space |
| 106 RAÄ density cells | one-degree aggregate cell | 106 archaeological sites |
| two fieldwork pages | documentation surface | two independent visits |
| one fieldwork feature | dated visit event | representative lake or regional coverage |
The SEAD difference reflects transformation from source inventory to normalized spatial records. The RAÄ difference reflects deliberate aggregation. The fieldwork difference separates documentation pages from the single event they describe. None is an error when the unit remains attached; all become misleading when reduced to an unlabeled total.
For any rate or comparison, publish the numerator, eligible denominator, excluded or unresolved count, and unit of observation. “Evidence-rich” is not a defensible substitute for those quantities.
Comparisons To Refuse¶
| Proposed inference | Why it fails |
|---|---|
| map co-location proves contemporaneity | spatial overlap carries no temporal equivalence |
| a project location is an exact sample location | project and sample are different observation units |
| more registered sites means greater historical activity | registry coverage, preservation, and recording effort differ |
| a contextual layer confirms a direct-evidence claim | evidence roles are not interchangeable |
| absence from one product proves source absence | scope, admission, recovery, or precision may explain exclusion |
| counts across families measure the same abundance | denominators and observation units differ |
Refusal does not make the families useless together. It defines the narrower question they can answer without manufacturing comparability.
Example: Lake Prioritization¶
A registered lake can supply sampling identity and geometry. Nearby archaeological records can describe contextual density; pollen sources can describe palaeoenvironmental context; and admitted ancient-DNA samples can provide direct evidence for their own governed specimens. A ranking may combine declared distance bands and sensitivity scenarios to prioritize fieldwork.
That result supports a decision about where to investigate. It does not show that contextual records originated at the lake, that the layers are contemporaneous, or that a high-ranked basin will yield a successful core. Those limits travel with the ranking.
Maturity Is Not A Single Rank¶
A family can have reliable acquisition but weak temporal comparability, or strong sample identity but unresolved coordinates. The live matrix therefore tracks evidence dimensions separately instead of collapsing them into a single confidence score.
The repository cross-domain evidence matrix reports current family posture. The atlas input audit reports which governed inputs reach publication. Source-specific interpretation begins with the source-family matrix.