Integrity & Provenance
Integrity & Provenance
PRM maintains separate public, protected, and controlled-review layers so aggregate findings can be inspected without exposing source text or reconstructable mappings.
PRM maintains separate public, protected, and controlled-review layers so aggregate findings can be inspected without exposing source text or reconstructable mappings.
Evidence boundary
Defines which evidence is public, which source materials remain protected, and which records are available through controlled review.
Aggregate evidence
Publishes metric behavior, charts, method records, and reproducible aggregate outputs.
Source audit
Preserves protected text, identities, titles, manifests, and reconstructable mappings for controlled verification.
Public Provenance Records
Public Provenance Records
Authorship and writing-condition documentation
First-draft and unaided-writing constraints
Versioned analytical outputs
Chart-generation and export records
Platform-side history when relevant
Reproducibility documentation
Hash and version tracking
Controlled disclosure procedures
Authorship and writing-condition documentation
First-draft, unaided-writing constraints
Versioned analytical outputs
Chart-generation and export records
Platform-side history when relevant
Reproducibility documentation
Hash and version tracking
Controlled disclosure procedures
Public Provenance
Public Provenance
Public provenance traces each reported finding from defined methods to versioned aggregate outputs and chart assets. It identifies metric families, preprocessing conditions, page-to-output mappings, and documented exceptions. Exceptions remain part of the result record and define the exact scope of each claim.
Public provenance traces each reported finding from defined methods to versioned aggregate outputs and chart assets. It identifies metric families, preprocessing conditions, page-to-output mappings, and documented exceptions. Exceptions remain part of the result record and define the exact scope of each claim.
Integrity Framework
Integrity Framework
PRM separates public aggregate evidence, protected source material, and controlled-review records into distinct layers. The public layer documents methods, results, and provenance. Source text, private manifests, mapping tables, and audit packets remain in the protected and controlled-review layers.
PRM separates public aggregate evidence, protected source material, and controlled-review records into distinct layers. The public layer documents methods, results, and provenance. Source text, private manifests, mapping tables, and audit packets remain in the protected and controlled-review layers.
Private Review Materials
Private Review Materials
Complete source manifests
Private mapping tables
Provenance and version records
Source-level documentation
Protected corpus material
Reproducibility and technical audit packets
Complete source manifests
Private mapping tables
Provenance and version records
Source-level documentation
Protected corpus material
Reproducibility and technical audit packets
Review boundary
Review boundary
Public pages document methods, aggregate results, corpus structure, and provenance. Controlled review supports source-level verification without turning the website into a repository for raw writing, third-party mappings, or reconstructable manifests.
Public pages document methods, aggregate results, corpus structure, and provenance. Controlled review supports source-level verification without turning the website into a repository for raw writing, third-party mappings, or reconstructable manifests.
Controlled review path
Controlled review path
Technical diligence, research review, and licensing discussions continue through controlled review. Depending on the review scope, this can include manifests, provenance files, version records, reproducibility documentation, audit materials, and source-level verification. Training, fine-tuning, benchmark use, and dataset licensing are separately defined rights.Technical diligence, research review, and licensing discussions continue through controlled review. Depending on the review scope, this can include manifests, provenance files, version records, reproducibility documentation, audit materials, and source-level verification. Training, fine-tuning, benchmark use, and dataset licensing are separately defined rights.
Technical diligence, research review, and licensing discussions continue through controlled review. Depending on the review scope, this can include manifests, provenance files, version records, reproducibility documentation, audit materials, and source-level verification. Training, fine-tuning, benchmark use, and dataset licensing are separately defined rights.Technical diligence, research review, and licensing discussions continue through controlled review. Depending on the review scope, this can include manifests, provenance files, version records, reproducibility documentation, audit materials, and source-level verification. Training, fine-tuning, benchmark use, and dataset licensing are separately defined rights.
Layered Verification
Layered Verification
PRM keeps the claim, aggregate evidence, and source-level verification connected without collapsing them into one disclosure layer. Public findings remain inspectable, while protected materials retain their evidentiary and commercial value inside controlled review.
PRM keeps the claim, aggregate evidence, and source-level verification connected without collapsing them into one disclosure layer. Public findings remain inspectable, while protected materials retain their evidentiary and commercial value inside controlled review.
How to read the charts
How to read the charts
Begin with the coverage dashboard, which maps the analytical surface behind the public site across workbooks, metrics, tokenizers, segments, slice sizes, and chart exports. Next, examine the exception count as part of the provenance record. Tokenizer robustness and the metric-slice heatmap show whether results persist across processing choices and measurement windows. The aggregate gauntlet dashboard connects those outputs to the primary comparison environment.
Begin with the coverage dashboard, which maps the analytical surface behind the public site across workbooks, metrics, tokenizers, segments, slice sizes, and chart exports. Next, examine the exception count as part of the provenance record. Tokenizer robustness and the metric-slice heatmap show whether results persist across processing choices and measurement windows. The aggregate gauntlet dashboard connects those outputs to the primary comparison environment.
Protected Materials
Protected Materials
Raw writing, protected excerpts, third-party source text, private manifests, source-level mappings, and protected identities and titles remain outside public disclosure.
Raw writing, protected excerpts, third-party source text, private manifests, source-level mappings, and protected identities and titles remain outside public disclosure.
Disclosure Boundary
Disclosure Boundary
Public pages disclose aggregate evidence, metric behavior, method provenance, corpus structure, and documented exceptions. Protected text, identities, source titles, and reconstructable mappings remain within controlled review.
Public pages disclose aggregate evidence, metric behavior, method provenance, corpus structure, and documented exceptions. Protected text, identities, source titles, and reconstructable mappings remain within controlled review.