Signal Architecture
Clasr is built in layers, each with a specific role. Signals are structured by the module that surfaces them and carried into a traceable report.
This is the part that does not change. English-language manuscripts go in, signal reports come out, and no module can turn that reading into a score, a decision, or a prediction of review outcome.
Other layers can evolve without changing this boundary. CORE keeps the system anchored to the same role: surfacing signals, not making decisions.
This layer defines how every report is structured: section order, label rules, Q1/Q2/Q3 calibration, sensitivity thresholds, and the fields each report must contain.
The rule set stays fixed within a version. Q settings can change sensitivity, but not the underlying report structure.
This layer configures each reading: target tier, field, input type, revision status, and how the report is presented.
A partial methods section and a full Q1 manuscript should not be handled identically. Tier 2 sets the route without changing the system’s core boundary.
This is where depth is added: argument integrity, figures and tables, reproducibility, sources, hedging, overreach, conclusion fit, reporting standards, and field-specific patterns.
Active modules work together to return structured signals rather than free-form commentary.
Not every manuscript needs the same checks. A theoretical law paper and a clinical trial raise different questions, so Clasr activates the modules that apply.
Signals are returned in a consistent structure, with their location, type, and severity preserved across modes.
Clasr separates the manuscript into layers — argument, evidence, structure, and context — then brings those signals back together in one report.
The underlying detection stays the same across Author, Reviewer, and Editor modes. Only the presentation changes.