The Unified Theory
One problem. One axiom. Six facets. Zero coincidences.
This series began with a single observation dressed in nine costumes (Where Everything Breaks): across medicine, finance, science, organisations, and AI, the same wound keeps opening. Truth has a status, and systems keep failing to track it honestly. This final article names the whole theory in one place and shows that it follows from one axiom and resolves into one question.
The problem, restated
Truth fails in two ways, and an honest system must defend against both.
It is lost by accident — overwritten, never captured, allowed to go stale, refuted without refutation propagation (non-propagating refutation). This is an architecture problem, and immutability is its defence: a record that cannot quietly lose the past (Truth Doesn’t Change).
It is refused on purpose — bent to fit an appetite, or buried because it is inconvenient. This is a human problem, and the same immutability is its safeguard: you cannot delete what was never deletable, and positional power can ignore epistemic authority but cannot erase it from the record.
Every mechanism in this work serves one or both of these defences.
The axiom
Truth is immutable, granular, and derivable upward but not downward — and every claim about it carries a strength, a commitment, and a standing that an honest system tracks.
The first three properties are about truth itself. The last clause is about our relationship to it.
Immutable. What happened, happened. The past does not change. A system that allows the past to be modified is not modelling reality — it is constructing a more convenient fiction, at the cost of the information replaced.
Granular. Truth exists at the finest grain at which events occurred. Bundling events together produces a lossy compression. The bundle can be derived from the atoms. The atoms cannot be recovered from the bundle.
Derivable upward but not downward. You can always produce a summary from the details; you cannot recover the details from the summary. The data processing inequality proves this formally: I(X;Z) ≤ I(X;Y) for any processing chain. Information flows in one direction.
Tracked honestly. Every statement carries an epistemic status whether or not anyone records it. A system that genuinely cares about truth attaches and maintains that status — the tier of the claim, whether it was committed, and the standing of whoever committed it. Not as a claim to certainty, but as a commitment to honesty about how well we know what we claim.
What the axiom generated
Tier-one substrate — two ways the regress stops. Commit captures (L0 when bound): the regress terminates by nature — the act of committing and the fact of the commitment are identical (Atoms of Truth). Measure captures: designated observations at an input boundary — real transductions whose chain could be pursued further, but the system stops by convention (Captured Evidence). Assert and formalise enter tier one without binding. All tier-one captures are substrate beneath the claim ladder — approximations of tier zero, the Real. Ground truth names tier zero only; calling tier-one records “ground truth” collapses capture into the Real. L2 probabilistic inferences stay pre-bind until someone commits; deterministic derivations rebuild from captured inputs.
The epistemic ladder. Above the substrate, every claim sits on one of twelve tiers, from logically necessary (1) through expert judgment, hypothesis, and personal assessment, down to refuted (12) (The Epistemic Ladder). The open question (11) is a first-class tier; an organisation can commit to one. The refuted claim (12) is never deleted, so refutation can propagate to everything that relied on it. Tier is tracked, decays with time, travels by weakest link along a chain, and is monitored for inflation. Without honesty about tier, the path toward truth is random.
Commitment and authority. Pre-bind material crosses the commitment boundary as a commit capture — the moment a named authority binds, with evidence snapshot, policy, and moment recorded (The Moment of Commitment). Principled binds are usually structurally atomic (L0); depth of the informing material is a separate facet (Why Atomicity Matters). But a commitment is only as good as the authority weight behind it: track record, qualification, certification, authorisation, in the relevant domain. Commitment without standing is worthless; standing without commitment is worthless. Positional authority is not epistemic authority, and a healthy system lets weight follow the latter — which is possible only because commitments are recorded and their outcomes can be scored. In organisational decision rooms, the record is only as honest as the principles under which it was written (Decision Principles and the Record).
The facets and their composition. Trustworthiness is not one property but five orthogonal facets (The Facets Composed) — tier, commitment status, authority weight, atomicity (L0–L3 via the L–U shorthand), and epistemic uncertainty (Low–High, decomposed into temporal staleness, probabilistic loss, and cross-stream consistency). They are independent: each can be high while the others are low. They compose into a single judgment by clear rules — commitment gates accountable use, uncertainty and atomicity cap it, tier travels by weakest link, uncertainty compounds multiplicatively — and that judgment is always relative to the stakes of the use. Information theory confirms the architecture rather than inventing it: processing can only lose information, the quality ceiling is set at observation, losses compound.
The decision strategy. Matching how trustworthy an artifact is against how trustworthy it needs to be gives the When the Commitment Boundary Needs Reinforcing: task nature, severity, and frequency determine the approach, under four axiomatic rules — once a D-F algorithm exists, execution needs derivation not judgment; catastrophic severity always requires a human at the commitment boundary; rubber-stamp approval is not review; severity is set by the highest downstream use. The commitment-quality mechanisms — blind peer review, AI stress testing, disclosed model disagreement — strengthen the bind (Strengthening the Commitment Boundary). Legitimacy — certification, authorization, democratic bind (When Social Stability Matters More Than Quality). Meritocracy — the overlap — is motivated separately (Meritocracy — The Sweet Spot, hypothesis).
The applications. LLMs sit at L2-U-high without a harness; the harness, not the model, sets the trust — and a naked model storing output as L0 is The Diagnosis and The Decision No One Made at industrial scale (LLMs at the Commitment Boundary, inference-as-fact). The ML reproducibility crisis is ruptured provenance chain — an architectural problem with an architectural fix (Event Sourced Science). Immutable containers are L1 artifacts (Immutable Infrastructure); deployment decisions are L0; stability propagates upward, and a foundation of atomic immutable events is what makes software cheap to extend in the age of AI. Decision logs with explicit assumptions and falsification triggers extend the same discipline to strategy and architecture commits; We Already Decided names when finality is discipline versus commitment-as-immunity.
The organisational identity. The single writer principle — one authority per invariant — applies as precisely to organisations as to software. Conway’s Law makes it a structural identity, not an analogy. The locking organisation and its cultural twin consensus culture are distributed systems with global locks: O(n²) coordination, negotiated rather than enforced invariants, diffuse accountability — anti-patterns for adaptive work, though consensus can fit stable environments where dynamic capabilities are less demanded. Clear domain ownership is the architectural prerequisite for organisational agility, and the reservation pattern lets it scale across cross-cutting obligations without synchronous coupling.
One axiom generates all of it
Looking across the whole series, no part requires a premise that is not already in the axiom or a consequence of it.
The two regress terminations at tier one — commit by nature, measure by convention — follow from granular and from the capture honesty Captured Evidence demands. The epistemic ladder follows from tracked honestly: if every claim carries a strength, that strength needs a scale. Commitment follows from immutable: the moment a derivation becomes a fact, it is permanent, and must be named, attributed, recorded. Authority weight follows from tracked honestly applied to who commits: standing is only meaningful if track records are kept. The facets follow from derivable upward but not downward: depth measures how far derivation has gone, uncertainty how much was lost, and the data processing inequality is the formal statement of “not downward.” The when-to-reinforce framework follows from matching the facets against the stakes. Organisational decision logs follow from commitment plus tracked honestly: assumptions, triggers, and dissent belong in the artifact, not in hallway memory (Decision Principles and the Record). The single writer principle follows from immutable and granular: if facts are immutable atoms and only one authority commits within a domain, the invariant is locally enforceable without coordination — and diffuse ownership is distributed write authority without coordination, which fails identically at team and service scale.
Nothing requires a new principle. Everything follows from the one.
The series demonstrates its own thesis
The conversation and the thinking that produced this series are the substrate the articles were built from — questions asked, answers given, pushbacks made, corrections accepted, insights reached. These articles are a L2–L3 projection over that record: organised, edited, structured for a different audience than the original exchange. The projection is useful. It is not the substrate. If you disagree with a conclusion, trace it to the reasoning and challenge the step where you diverge.
These articles will eventually be out of date. New cases will arise that the facets do not cleanly cover. New tiers of subtlety will push the ladder further. That is how it should be. A framework that claims to be finished is a monument. This one is a hypothesis under continuous investigation — stated precisely enough to be challenged, and worth challenging because it covers more ground than anything that came before it. It even tells you its own tier: not tier 2, not certain, but a tier-4-to-6 structure, committed to, by authors accountable for it, and open to refutation.
Truth doesn’t change. Systems and organisations that model it correctly — that track its status honestly, defend it against both accident and suppression, and match their confidence to the stakes — are more auditable, more evolvable, more accountable, and more honest than those that don’t. That claim is testable. The testing continues — and where it goes next is the subject of the outlook.
Part IX — Synthesis and Epilogue turns from the unified theory to open frontiers — disposition, harness design, and what remains to build.