LANSKRIT
← Back to Atlas
Evidence and KnowledgeUnder reviewInteractive graftControversy: low

Evidentiality

How do we know this?

A grammatical or lexical distinction that marks the source of a speaker's information — whether directly observed, reported by another, inferred from evidence, or assumed.

Open in Evidentiality Graft LabExecutable Lab →
Graft transformationField record → structural reading → speculative construct → running machine
  1. T1 · Documented

    Attested in 5 languages: Tuyuca, Tariana, Turkish…

    What linguistic research records.

  2. T2 · Interpreted

    ordinary values usually retain only identity, value and type

    How Lanskrit reads the code deficit.

  3. T3 · Proposed

    typed value (Inferred<T>, Observed<T>, Reported<T>)

    Speculative constructs — not executable code.

  4. T4 · Executed

    Deterministic runtime available — Evidentiality Graft Lab.

    Deterministic runtime — the graft as running machinery.

    Open executable lab →
§ EvidenceFour tiers of verifiability

Each tier is a different kind of claim. We keep them visually distinct so a reader can never confuse a field record with a deterministic run.

T1Documented

Documented

What linguistic research records.

  • origin of a claim
  • directness of observation
  • identity of a reporting source
  • basis for inference
  • degree of speaker commitment
T2Interpreted

Interpreted

How Lanskrit reads the code deficit.

  • ordinary values usually retain only identity, value and type
  • provenance is discarded or reduced to a free-text field
  • systems cannot distinguish observation, report, inference and assumption structurally
T3Proposed

Proposed

Speculative constructs — not executable code.

  • typed value (Inferred<T>, Observed<T>, Reported<T>)
  • annotation (@inferred, @observed)
  • wrapper object (Evidence.inferred(...))
T4Executed

Executed

Deterministic runtime — the graft as running machinery.

  • Runtime
    lanskrit-evidence-runtime-v0.1
  • Fingerprint
    sha-256 · canonical form
  • Receipts
    reproducible on any machine
Open executable Lab →
Computational opportunities
  • provenance-aware values
  • evidence types as part of a type system
  • source-sensitive validation
  • confidence propagation across expressions
  • structural conflict detection between claims
Warnings against overgeneralisation
  • Not every language marks evidentiality grammatically; many express it through discourse or vocabulary.
  • Evidentials describe the speaker's stance, not an objective measure of truth.
Example construct

Conceptual syntax — not executable code

typed value
let risk: Inferred<Decimal> = infer(model_v3, tenant_history);
Documented in languages

Cited as examples of the feature, not as claims about all speakers of these languages.

Addresses code opportunities
Related capabilities
Research status

Under review

Research source pending verification.

Central Phase 1 capability and the basis for the current interactive Graft Lab.

Version
Record version
0.2
Last reviewed
2026-07-21
Controversy level
low
Corrections
Full log →
  1. COR-2026-0001
    2026-07-22
    reclassification

    Separated reliability and confidence from evidence kind; kind is now immutable once constructed.

    Volume V white-paper alignment: evidence kind should describe how a claim was obtained, not how much we trust it.

    cites lens.evidentiality.v1 v1.0.0

  2. COR-2026-0002
    2026-07-22
    reclassification

    `disputed` reclassified from an evidence kind to a typed relation between two claims.

    Volume V: disagreement is a property of the graph, not of either claim in isolation.

    cites lens.polyphonic-conflict.v1 v1.0.0