RMR
Review-friendly
Preserves structure for human review, provenance checks, and source-facing QA.
TQT
Measurement-clean
Normalizes text for statistical work so metric behavior is not driven by formatting artifacts.
Separation
Review vs stats
Keeps public claims tied to the right layer of the pipeline.
Evidence Frame
RMR
RMR is the review-friendly transform.
It preserves more readable structure so a reviewer can inspect flow, segmentation, and preparation choices inside a controlled environment.
RMR is not a public source publication. It is the layer that helps explain how material moved through preparation without turning the site into a corpus dump.
TQT
TQT is the measurement-friendly transform.
It is used for metric consistency, reduced punctuation interference, tokenizer analysis, and statistical comparison.
TQT makes the corpus easier to compare across tokenizers, slice sizes, and metric families by reducing formatting noise before quantitative analysis.
Why both exist
RMR and TQT answer different questions.
RMR helps preserve review readability.
TQT helps stabilize quantitative analysis.
Together they let PRM be reviewed as both writing and measurable language data without merging those jobs into one unstable layer.
How they fit the pipeline
The pipeline uses transforms to keep review and measurement separate. RMR supports controlled human review. TQT supports repeatable metric work.
That separation matters because a public-safe website cannot expose protected source text, private manifests, or source-level mappings. The transform layer lets the project describe the process while raw material remains inside the controlled review boundary.
What the comparison does not mean
RMR and TQT are not competing claims about which version is “real.” They are different working views of the same underlying material.
The public page does not ask readers to reconstruct the corpus from either transform. It explains why two transform layers exist and how they support reviewability, measurement consistency, and public-safe reporting.
How to read the charts
Start with deep metric coverage to see why transform discipline matters. The chart shows that RMR and TQT are not cosmetic versions of the text; they sit upstream of a large measurement surface involving workbooks, metric families, tokenizers, segments, and slice sizes.
Tokenizer robustness is the clearest TQT-facing check: it asks whether normalized measurement text still produces stable results across tokenizer systems. Raw trends, the metric-slice heatmap, and the 65536 macro lens then show whether the transform choices support large-scale aggregate behavior instead of only small-excerpt effects.
Public-safe limits
This page describes transform roles and aggregate outputs. It does not publish raw writing, protected excerpts, private transform files, source-level mappings, third-party source labels, artist names, album titles, or song titles.
Public-safe boundary
Public pages show aggregate evidence, metric behavior, method provenance, and corpus structure. Protected text, identities, source titles, and reconstructable mappings stay private.