R29: the PROBLEM statement is the boundary — a solution outside it needs its own PR
R29 calibration: count runtime decisions; size is its own signal. USAGE slot + enumerated SOLUTION in the brief
R29: split by decision count, not line count
R1j: an unjustified change is a finding even when nothing measurably breaks
R18e/R18f: price the render pass an effect costs; scope listeners to the moment
R2g: an inline names the problem and points at the pattern
R6e: a superseded PR's leftovers need a user, not just a difference
R28: walk the use case end to end — a mechanism that cannot complete it is a different feature wearing its motivation
R28: establish the need before the mechanism; decide the category not the instance
R3j: if the component can apply it, exposing it is the finding — and it found the missing start alignment
R16f: two halves, two verdicts, two bars — restoring a contract is not adding one
R18d: an effect's dep array is a claim about when it fires; resolve the chain and read it against the comment
R27: four tests for what the loop cannot decide, and the one-sentence handover
Pause R26b/R26c: the pitch pattern is unproven; R26's review-never-merge stands
R26c: framework-decision and pitch labels as the register the loop checks in STEP 1
R26b: pitch not gate, collect considerations not just code, and do not route contributors into a noisy tracker
R26b worked example: the layer stack lands half-adopted; six layers still off it
R26b: an owned integration branch turns N rulings into one design decision plus N normal reviews
R26: the gate is unattended merging, not merging — recommend, do not shrug
R26: a new subsystem is reviewed but never merged by the loop; ask the design question, not its six consequences
R13d: refresh a fork, never push content; ruling 3 on the edge-alignment mechanism
R25: public docs are for callers, not for building Astryx
R24: every touched PR exits in a tracked state; approved-and-blocked joins the ladder
R23: a cut change inherits the reasoning; it reviews itself and arms its own merge
R18c-cache: assert the cache, not the clock; the markdown ruling owes a perf test
R22 x R6b: authorship changes what an overlap means; a team draft is precedence, not a peer option
R22: a class is made of LANDED members; a draft settles nothing
What the loop is FOR: goaled on good reviews, not merges
R21: the gate-clearing workflow can itself be gated; an approved unblocked PR gets merged
R22: a feature review settles everything verifiable and hands her one question