CounterProof Research essay, 14 October 2026. By Luuk Joseph Soons, director and co-founder of CounterProof Research, a practice of Clavestra Capital Limited. Original draft developed with Claude (Anthropic); this revision prepared with Codex (OpenAI). This version was reviewed before publication by Kimi (Moonshot AI), DeepSeek and GLM (Z.ai); the review records are held in our reports store and available on request.
What does CounterProof have to do with counterpoint in music?
In counterpoint, several melodic lines sound together while retaining their own character. In a review, several people or AI models examine the same work. In both cases, it matters whether each voice contributes something of its own.
For CounterProof, this is a practical concern. When several reviewers agree on a finding, have they each found evidence for it, or are they repeating a shared mistake? Music offers a way into that question. It also helps explain why criticism matters to anyone trying to build something others can depend on.
Schelling called architecture frozen music. Goethe, crediting the image to "a noble philosopher", proposed calling architecture music that has fallen silent.
On the bestebank.org site, I placed Josquin des Prez's lament for Ockeghem beside a call to build decentralised structures that serve individual freedom. My interest in Bitcoin belongs to the same concern: how people can cooperate without surrendering their independence. Music offered an image of distinct voices sustaining a common work. The same page credits the frozen-music line to Goethe; as Goethe himself said, the image was another's.
Josquin's Nymphes des bois / Requiem aeternam belongs to the Franco-Flemish Renaissance. Its five vocal parts bring together a French lament and a Latin chant. The voices sometimes imitate one another and sometimes move together, giving the words their shared weight. Their distinctness exists within a composition that binds them to one another. Musical analysis.
Here the analogy reaches a limit. Musical independence does not imply independent origins, judgements or errors. The voices share a composer and a design. When we turn from music to a review panel, we have to ask what each voice contributes to our knowledge.
Johann Joseph Fux's Gradus ad Parnassum (1725) offers a useful starting point. Its rules forbid two voices from travelling together in parallel fifths or octaves. These intervals sound consonant, but voices moving together in them can seem to fuse. The texture appears to retain two lines while the ear hears less distinction between them. David Huron's account of voice-leading (2001) explains the rule through such perceptual principles. It serves a particular musical aim: preserving the identity of the parts. It is not a rule for every kind of polyphony; early parallel organum deliberately used fourths, fifths and octaves. Historical examples.
A review panel can suffer a comparable loss of distinction. Reviewers may agree because each has found sound evidence. They may also agree because they share an assumption, a training influence or an earlier review. What sounds like confirmation may be several expressions of the same mistake.
A second opinion is especially valuable when it can expose a blind spot in the first. Suppose each of three reviewers misses a flaw one time in ten. If those misses are independent, all three miss it one time in a thousand. If their misses coincide completely, all three miss it one time in ten: three signatures under one mistake. Three people giving directions from the same out-of-date map will agree with each other. Their agreement tells you about the map, not the road.
Stemmatics, the study of relationships among textual witnesses, sharpens the distinction. Paul Maas called the errors that reveal descent Leitfehler, or indicative errors. Agreement in a correct reading may tell us little about ancestry: any careful copyist might reach it alone. A shared error can point to a common source, provided it is unlikely to have arisen independently. An ordinary copying mistake carries less weight, although several shared features may together become significant.
For model panels, two questions must therefore be kept apart. Do the reviewers share a source or ancestor? And do they tend to fail together? A common false positive may tell us little about lineage while still showing why agreement deserves less confidence. Evidence of descent requires something more distinctive: a shared idiosyncrasy whose independent origin is improbable.
Deliberately planted markers have demonstrated this possibility. Whether naturally occurring model errors can supply reliable markers of descent remains an open question. For practical review, the immediate task is to establish where reviewers' errors overlap. Different model names cannot answer that question for us.
Counterpoint also offers constructive guidance. Contrary and oblique motion allow one voice to rise while another falls or holds still. A panel can likewise approach a problem through different questions, entry points and assumptions to test. That may expose different weaknesses. It does not establish independence by itself, and asking for disagreement is no substitute for checking the evidence. The value of the arrangement has to be measured.
Bach's Art of Fugue shows another part of the discipline. It was printed in 1751, the year after his death, in an order that differs from the autograph manuscript. The unfinished fugue breaks off in a partially written bar 239. One of its subjects begins with B-A-C-H: the German note names for B flat, A, C and B natural.
A manuscript note, usually attributed to Carl Philipp Emanuel Bach, says that the composer died while working on this fugue. It gave rise to the familiar story that Bach died in the act of writing it. But the note is itself a witness to be assessed. Christoph Wolff argued in 1975 that Bach had worked out the combination of the fugue's subjects in a draft now lost; others have questioned that reconstruction.
Editors compare the surviving witnesses, weigh their authority and explain what remains unknown. A review report owes its reader the same care: which reviewers raised a concern, what is known about their independence, what supports the concern, and where the evidence stops.
This is why criticism belongs inside the work of construction. Before asking others to depend on what we have built, we must give them grounds for confidence and means to challenge it. The convictions that lead us to build must also remain open to correction. A design deserves particular scrutiny where its author most wants it to succeed.
The motto on the bestebank.org site, "Listen to all, follow no one!", expresses that responsibility. Listening requires a fair hearing. Keeping one's judgement requires being willing to revise it. Freedom, fairness and truth become practical commitments when criticism can change what we do.
The Latin contrapunctus comes from punctus contra punctum: note against note. CounterProof's name makes a similar move: a proof set against a proof. The purpose is to discover what can bear examination, including in our own work.
In Fux's first species, every interval between the voices must be consonant. Yet parallel fifths remain forbidden. A pleasing chord does not excuse the way the voices arrived at it. For a review panel, the corresponding question is whether agreement brings further evidence or merely repeats a shared assumption. The finding must stand on its evidence, however many reviewers support it.
When a panel sounds perfectly consonant, check the voice-leading before trusting the chord.
CounterProof is a small practice of Clavestra Capital Limited (Malta, C 113987) that reviews AI-generated code and AI-assisted review processes for organisations where a wrong answer is expensive. We disclose that we have a commercial interest in the adoption of the review method this essay describes.
Cite as: Soons, L. J. (2026). Counterpoint in music, CounterProof in method. CounterProof Research, 14 October 2026. Companion to Agentic Stemmatics, doi:10.5281/zenodo.22030516.