Review round 3

Review round 3 was the third blind review of #14, the standalone manuscript proposing that dark matter consists of superheavy ’t Hooft–Polyakov monopoles of a hidden gauge sector. It ran on 2026-10-19 on version 3, with a fresh subagent as referee that could execute code. The referee checked every citation against its source, recomputed every number in its own scripts, re-derived the manuscript’s one theorem, looked for constraints the manuscript had left out, and graded it C, fundamentally flawed: the count of monopoles formed in the early universe, on which the predicted mass rested, used a wrong normalization of the Kibble estimate. The adjudicator recomputed the central numbers and agreed. The decisions that followed retired the soliton candidate, made strict sector separation the reading of postulate P4, and set the toy Lagrangian as the next task, and the round wrote two lessons into the program’s method.

Records only: the four pages of this section record one review and make no claims of their own. The statuses the round set are kept in the registers: the retired candidate and the parts of it that died, the soliton candidate, the Kibble normalization, the ψ relic window, Planck-suppressed adjoint mixing and the √N floor, in the graveyard; the corrected overproduction at the benchmark and the W′ overclosure among the claims that stand; and the queue the round changed in open problems and the queue.

Map#

Round 3 was planned in the claude.ai period. A first attempt, dispatched from claude.ai on 2026-10-17, “died partway on a rate limit after reporting ‘most citations and core numbers checked; several omitted-constraint literature checks remaining.’” (HANDOFF.md §6, which still calls the round incomplete). After the move to Claude Code, v3 was cut from round 2’s pending list in review/v3-worklog.md, and the round ran to completion under the five-step protocol: the forecast was committed, a referee given only the manuscript audited it, and the author ruled on each finding, recomputed the central numbers and decided what followed. The pages take the round in that order, with the numerical evidence last.

Round-3 pre-registration comes first because it was committed before the referee was dispatched. It forecast verdict B, the weakest link “the framework does no work”, no surviving numerical indictment and all 23 citations resolving. The page sets out the forecast’s reasoning, how each part scored (a miss on the verdict, the first after two hits), and how the commit’s subject line leaked the forecast into the referee’s view.
Round-3 verdict, rulings and citation audit is what the referee found and how the adjudicator ruled: the grade C, the thirteen rulings with their evidence, the compatibility theorem verified and re-attributed to Freidel–Leigh–Minic, the constraints #14 had omitted, accepted as a to-do (among them supernova SN1987A, dark acoustic oscillations, the Higgs portal, collective plasma effects, reheating and uncited prior art), the audit of all 23 references, and what survives v3.
Round-3 decisions and lessons is what the program did next, under Paul’s delegation and veto: strict sector separation, the retirement of the soliton candidate after weighing three routes, the dropped √N floor, the toy Lagrangian as the next task, the same-day withdrawal and change of gate script, and the two lessons now in the root CLAUDE.md.
Round-3 arbitration and the referee’s constants is the numerical evidence behind rulings 1, 2 and 4: the adjudicator’s script, read line by line, and its output; the referee’s own numbers check, constants and validations; where the two calculations differ; and how the corrected numbers reached the gate.

The round’s files are review/round3-preregistration.md, review/round3-adjudication.md, the referee’s notes, scripts and outputs in review/round3/referee/, and the adjudicator’s script and output in review/round3/adjudicator/. The manuscript is notes/knot-dark-matter-standalone.md (v3, withdrawn; the round’s files cite it at its old path in papers/). The root CLAUDE.md puts the adjudication third in a fresh session’s reading order, after #11 and HANDOFF.md §6.

Boundaries#

The physics the round tested and killed is in the soliton candidate: the overproduction at the benchmark, the W′ overclosure that survives, kinetic mixing between sectors, the hidden sector’s thermal history and the published topological dark matter that the candidate’s viable low-mass version reproduces. The general monopole physics, such as the Kibble mechanism, is in monopoles.
The theorem the referee re-derived is the compatibility theorem, re-tagged [T] with credit to Freidel–Leigh–Minic after the round.
The protocol every round follows, the scale of grades and the forecast’s calibration across rounds are in blind review: the review protocol, pre-registration and the measured referee failure mode. The round before is rounds 1 and 2, the round after round 4.
The script that now gates the round’s corrected numbers is the soliton-obituary script, run by the gate; the rules the round produced are independent anchors and verification checks arithmetic, not the model.
Strict sector separation as a postulate is P4; the toy Lagrangian the round set as the next task is in toy models of broken reciprocity; the referees’ shared objection is the ‘framework does no work’ critique; the dates are in the timeline.