Signal Architecture
Clasr is a stack of modules, each with one job. You can trace any signal back to the module that made it. Nothing is hidden.
This is the part that never moves. English manuscripts go in, signal reports come out, and no module gets to turn that into a score, a verdict, or a guess at what a reviewer will say.
Every other layer can change around CORE without touching it. That's the whole point of putting it at the bottom: the system can grow without ever changing its basic promise.
This is what holds a report together: section order, label rules, the Q1/Q2/Q3 gates, expression thresholds, the blocks every report has to include no matter what.
It doesn't drift within a version. You can turn the sensitivity up or down with Q settings, but the structure underneath stays exactly where you left it.
This layer decides how a specific reading should run: which Q tier applies, which field, whether the input is partial, whether it's a revision, which output mode to use.
A partial methods section and a full Q1 submission shouldn't be read the same way, and they aren't. Tier 2 picks the route. It never touches what the system fundamentally does.
This is where the depth lives: argument integrity, figures and tables, reproducibility, sources, hedging, overreach, conclusion fit, reporting standards, domain-specific patterns.
It all runs as one architecture now, every module handing back a signal you can trace, never just an opinion.
Not every manuscript needs the same stack. A theoretical law paper and a clinical trial are asking different questions, so Clasr only switches on what actually applies.
Every signal can be checked against the module that made it, and the version that shaped it. Nothing in here is a black box.
A careful reviewer doesn't read a paper in one pass. They read in layers, argument, then evidence, then structure. This architecture is built to do the same thing.