A turn from writing code to writing the rules, checks, and values that code must answer to — and making all three legible enough to be contested.

The vision

For most of three decades I have worked across cognitive science, philosophy, and systems engineering, mostly in private. This manifesto marks the move into public: the deployment of the RSR 2026 (Rhodium Standard Repository) and the body of work that surrounds it.

The thesis is simple to state and hard to live up to. Most of what governs software is left implicit — intent lives in prose, soundness lives in trust, value lives in authority. The metaprogramming turn is the decision to make those governing things explicit, first-class, and inspectable: to write not only the program, but the rules it must obey, the checks that establish it, and the values it encodes, in a form that both a machine and a critic can read and argue with.

I can begin this. I cannot finish it alone — and a manifesto about making value contestable would contradict itself if it stayed a monologue. So from here it is we.

We value, in the precise sense that this whole project is about value:

• explicit, contestable structure over implicit good faith;

• checked certainty over fluent plausibility;

• rigour applied wherever a problem demands it over rigour rationed to the problems we have agreed to call "technical".

The assurance turn

Making rules explicit is not enough. A legible rule can still conceal who authorised it, erase the particular person or circumstance it classifies, or turn a passing check into an inflated claim. The assurance turn follows from the manifesto's commitments: governing statements must be connected to evidence, their boundaries and assumptions must be visible, and the people affected by them must be able to contest both representation and result.

Assurance here is therefore not a badge or a promise of absolute safety. It is a disciplined relationship among a claim, the evidence offered for it, the model in which that evidence has force, the residual uncertainty, and an authority accountable for the decision. Draft, designed, assumed, tested, audited, and proven work are different states. Fluency does not collapse them.

Particularity matters because abstraction always leaves something behind. A type, policy category, or machine-readable record is useful only when its lowering remains faithful to what is materially at stake. Contestability is the corresponding political commitment: an affected person must be able to discover the rule and evidence, challenge a classification or automated decision, and reach a responsible human authority.

This chapter explains the philosophical direction, not its operational rules. The proposed estate constitution defines authority, precedence, exceptions, assurance duties, and change procedure. Until adopted through review, it remains a proposal.

Core pillars

Each pillar is a position, not a slogan. Each is meant to be disagreed with.

1. Machine-readable governance

The rules an estate lives by — licensing, provenance, conventions, who may change what — should be encoded as machine-readable artifacts, not narrated in prose that quietly drifts from the code it describes. A human-centric README cannot be checked; governance-as-data can. The 6SCM pattern is how this estate encodes those rules so that tooling, and AI agents, can verify conformance and flag drift rather than take a paragraph’s word for it.

What this rejects: the prose README as the source of truth about how a system is meant to behave.

2. Neurosymbolic verification

Probabilistic generation may propose; only symbolic method may dispose. Where correctness matters, an unproven output from a language model is a hypothesis, not a result, and treating fluency as evidence is a category error. The discipline is to pair generation with mechanised checking so that the part which can be certain has the last word.

What this rejects: plausibility mistaken for proof. (hypatia is the clearest concrete instance.)

3. Post-disciplinary praxis

The rigour we reserve for compilers belongs equally to consent, to policy, to accountability. The line between a "technical" problem that deserves formal treatment and a "social" one left to good faith is itself a choice — and often an alibi. Post-disciplinary means reaching for whatever discipline the problem needs; praxis means refusing to separate the theory from the practice that tests it. The same instinct runs from labour-union policy to RISC-V SSGs.

What this rejects: the division of labour that keeps rigour out of the rooms where power actually operates. (avow-protocol, consent-as-code, is the most developed example.)

The wager: value made legible

Underneath all three pillars is one bet. Deciding what a system should do, whether an output counts as correct, who consented, and what matters are all judgments of value — and they are usually smuggled in implicitly and backed by authority: trust me, the spec says so, the model said so. Axiology is the study of value; the metaprogramming turn applies it operationally. Value is not asserted, it is structured — and structure can be inspected, checked, and argued with.

This is also why the work is, at root, political as well as technical. A value that is explicit and inspectable can be contested by the people it bears on; a value backed only by authority cannot. Making value legible is therefore a precondition for accountability — in a compiler, in an AI system, and in a union. It is the point at which formal methods and solidarity economics turn out to want the same thing: that the rules be legible, and the powerful be answerable to them.

It is the same instinct that runs through the technical research — identity defined by equivalence rather than asserted, a value justified by the construction that produced it rather than taken on faith. The manifesto only generalises that instinct, from values inside a program to values inside our systems and institutions.

Reading order

• EXPLAINME.adoc — the canonical "what is this" entry-point, consumed by AI agents and by the hyperpolymath/standards governance tooling.

• docs/ — supporting essays, position papers, and ADRs that situate the manifesto in context.

• MAINTAINERS.adoc — current stewards.

Status

• Licence: MPL-2.0 (migrated from PMPL-1.0-or-later on 2026-05-26 per the estate licence-debt audit, hyperpolymath/standards#196).

• Maturity: a work in progress; the manifesto is a living document. The version number on the title line is incremented on each substantive revision.

• Audit findings: see docs/tech-debt-2026-05-26.md if present (added by the 2026-05-26 estate tech-debt scan).

Engaging with this manifesto

Collaboration here is not decoration; by the argument above it is constitutive. A manifesto that makes value contestable earns its keep only by being contested, adopted, and extended. Three ways in:

• Disagree — open an Issue. Manifestos thrive on being argued with, and a pillar that no one challenges has not yet been tested.

• Adopt a thread — fork, riff, cite. The MPL-2.0 licence makes derivative manifestos easy.

• Contribute editorially — see CONTRIBUTING for the (intentionally light) editorial conventions.

Whether you are a researcher, a maintainer, a journalist, or a student, the door is open — and the NUJ and solidarity-economics threads here mean I take "we" literally.

Companion repositories

• hyperpolymath/standards — the RSR 2026 standards that the first pillar references.

• hyperpolymath/hypatia — the neurosymbolic analyser that is the second pillar’s clearest concrete instantiation.

• hyperpolymath/avow-protocol — the consent-as-code stack that is the third pillar’s most-developed example.