The Epistemic Ladder
Fourteen levels from the Real to the refuted claim — and the rules for climbing honestly
If every statement carries a tier, we need a scale to put them on. Not a vague sense of “stronger” and “weaker” — a small, stable, named set of levels that anything can be placed on, from a mathematical theorem to a rumour to a claim that has been proven false.
This is that ladder. It does not start at our records; it starts at the truth those records reach for. At the bottom is the Real (tier zero — Rᶜ contingent and Rⁿ necessary). Just above: captured evidence (tier one) — commit, assert, measure, formalise captures. Everything above is a claim about the Real, ranked by support. (The Real develops tier zero; Four Capture Modes develops tier one.)
Tier 0 — the Real
Tier zero is not a record and not a claim. It is the truth itself: what is the case, independent of whether anyone measures, records, or formalises it. This is not only mind-independent reality — the mutual gravitation drawing earth and apple toward each other, the patient’s actual condition, the real structure that mathematics reaches toward — but also the brute fact of real acts: that a commitment was genuinely made is part of what happened, whether or not it was ever written down. All of it is real whether or not any system holds it, and whatever frame we later put on it.
No system contains tier zero. (The Real develops tier zero in full: necessary and contingent truth, Gödel, fallibilism, lost certainty.)
(A second, deliberately reserved meaning sits at tier zero too, and it is no accident that it shares the floor with the Real: the rare statement that is both verifiably true and genuinely funny — the laugh of recognition that comes from touching what is so cleanly that it tips into wisdom. Wisdom is not more knowledge piled higher up the ladder; it is clear contact with the Real at the bottom. We keep the seat. The load-bearing meaning of tier zero, for everything that follows, remains the Real itself.)
Tier 1 — captured evidence
Tier one is captured evidence — not ground truth in the tier-zero sense. Four capture modes enter the permanent log:
These are not claims competing for strength. They are fallible substrate — events reasoning is about. Every claim above must trace to tier one or float free.
We do not relitigate tier one literally — that this instrument reported these values at these times. We argue about interpretation. Capture gaps (coercion, forgery, chain distortion, stream-to-summary collapse) are mode-specific; see Captured Evidence and Thinking by Writing.
The twelve claim rungs
Above the substrate, the rungs classify how strongly a claim is supported. There are twelve of them, numbered 2 through 13, running from the necessarily true down to the explicitly refuted. Each carries a characteristic source of error — the way claims at that level typically go wrong — because a level is only honest if it names its own failure modes.
A few of these deserve more than a row.
Tier 2 — logically necessary given evidence. A tier-2 claim is logically necessary given tier-one captured evidence — a statement that is true if and only if the substrate it rests on is true. No empirical leap, no expert gap, no model assumption beyond what the captures already contain: the claim cannot diverge from its evidence without a logical error. Examples include deductive restatements of committed facts, arithmetic on captured values, and formal theorems whose premises are tier-one records (axiom adoptions as commit/formalise, proof steps in the log). A theorem in ZFC or Peano Arithmetic is tier 2 relative to those tier-one formalisations — not a direct grasp of Rⁿ at tier zero.
This is the strongest rung a claim can reach, but it is not stronger than its substrate. Tier-one captures are fallible (coercion, forgery, instrument error, stream collapse); tier 2 inherits every gap in what it was given. Gödel still applies where formal systems are rich enough: truth in the Real can outrun what any fixed proof system captures. Tier 2 means the claim is locked to its evidence — not that the evidence is locked to the Real. Its characteristic misuse is tier laundering: dressing an empirical inference or a model output as logically necessary, or borrowing tier-2 authority for claims that are only true if the evidence is true, not iff.
Tier 3 vs Tier 4. The line between established and supported is the consensus threshold. Tier 3 is germ theory and evolution: replicated, peer-reviewed, no credible contradiction. Tier 4 is much of working nutrition science, psychology, economics — well-supported but still moving. Putting a tier-4 finding forward as tier 3 is the most common form of scientific overreach.
Tier 5 — expert consensus (global, not local). Tier 5 is field-level agreement among domain experts — specialty-society guidelines, engineering codes, formal expert panels, published consensus statements — when decisive tier-3 or tier-4 evidence is not yet available or does not fully settle the question. The warrant is what the relevant expert community converges on, not what one credentialed person in the room thinks, and not what this organisation prefers.
A lone radiologist’s read, a senior analyst’s forecast, or “our architects agree” is not tier 5. Those are tier 10 (personal or local assessment) unless they trace to a documented global expert synthesis. Local consensus without field standing is closer to tier 9 — consensus culture and homophilic closure often produce tier-9 comfort mistaken for tier 5.
When expert consensus rests on tier-3 science, assign the empirical tier, not tier 5. Tier 5 is for the gap where experts must judge under uncertainty — valuable, but not replication and not one voice.
Tier 6 — model-based. A claim from a model is valid within the model’s assumptions and not automatically valid in the world. A financial projection, a climate scenario, an epidemiological forecast. Its error sources are misspecification and extrapolation beyond the calibrated range — the model used outside the conditions it was built for.
Tier 5 vs Tier 9 vs Tier 10. Three different “we agree” shapes:
Marking a personal preference as tier 10 is honest. Dressing it as tier 5 — because the speaker has credentials — is tier inflation. Dressing org comfort as tier 5 is worse: local agreement laundered as field consensus.
Tier 12 — the open question is a first-class citizen
Most classification schemes have no place for not knowing. The epistemic ladder makes it a level.
Tier 12 is explicit, acknowledged uncertainty: we do not know whether X holds under condition Y. It is not a failure state. It is a starting point — the honest marking of a gap. And it is the level that most distinguishes a truth-seeking system from a confident one, because a system that cannot represent its own ignorance will fill every gap with a guess and forget it did.
There is a question worth settling plainly: can an organisation commit to an open question? Yes. A committed open question is a genuine and valuable act — an authority placing on the permanent record that we officially do not know this, and we are accountable for not knowing it. It is a tier-12 claim that has crossed the commitment boundary: not a gap left lying around, but a gap the organisation has formally acknowledged, dated, and made answerable. Open questions at tier 12 are then prioritised for resolution by how many high-stakes decisions depend on them — ignorance ranked by what it is costing.
Tier 13 — refuted, never deleted
When a claim held at any tier is disproven, it does not vanish. It moves to tier 13 and stays in the record, archived with the evidence that refuted it, the date of refutation, and the decisions that relied on it while it stood.
This is the institutional memory of the system’s own errors, and it is the direct answer to non-propagating refutation — the failure introduced as The Retraction in Where Everything Breaks. The reason a retracted finding keeps poisoning downstream work is that systems treat refutation as deletion — quietly removing the claim and leaving everything built on it untouched. Tier 13 does the opposite: it keeps the refuted claim connected, so refutation propagation can reach everything that ever depended on it. A claim demoted to tier 13 drags its dependents into review. Refutation that cannot propagate is not refutation; it is forgetting with extra steps.
The rules for climbing honestly
A ladder of levels is only as good as the discipline around assigning and maintaining them. Five cross-cutting rules govern the whole structure.
Provenance. Every claim above substrate must trace to tier-one captured evidence — which answers to tier zero (the Real).
Recency and decay. Tiers are not permanent. A tier-3 finding can decay as its evidence ages or its context shifts; a tier-7 hypothesis can be promoted as tests accumulate. A claim’s tier carries a timestamp, and a stale high tier is exactly The Assumption That Calcified — a strength assigned years ago and never rechecked.
Weakest link. A conclusion built on a chain of reasoning inherits the tier of its weakest step. A tier-3 fact and a tier-10 hunch combined into a single inference produce, at best, a tier-10 conclusion. You cannot launder weak premises into a strong conclusion by reasoning carefully from them.
Inflation monitoring. The natural drift of any organisation is upward: hypotheses harden into facts, preferences acquire the tone of expertise. Honest tiering requires actively watching for tier inflation — claims sitting higher than their evidence warrants — and demoting them. This is the mechanism The Confident Deck and the fluent-but-unfounded model output are designed to defeat, and the one a healthy system runs continuously.
Scope by tier. Not every claim earns the same reach. Strongly supported claims (roughly tiers 2–4) may underpin wide reasoning — shared assumptions in policy, analysis, and design. Mid-tier claims (5–7) should stay scoped to the decision, domain, and assumptions that produced them: useful where they were formed, hazardous as organisation-wide defaults. Weak, subjective, or speculative claims (8–13) must still be tracked, but not allowed to silently become load-bearing in chains that treat them as established fact. Scope discipline — together with weakest link and tier-13 refutation propagation — is how the ladder stops weak claims from travelling as if they were strong.
One facet among several
The ladder answers one question with care: how strongly is this claim supported? It does not answer whether anyone committed to the claim, how much that committer’s standing weighs, how far the claim sits from ground truth in processing steps, or how fresh it is right now. Those are the other facets — and the central architectural insight of this part is that they are independent, and that trustworthiness is what you get when you combine them.
A tier-3 claim can be stale, uncommitted, and three lossy transformations from the substrate — and therefore unsafe to act on despite its strength. A tier-7 hypothesis, freshly committed by a high-weight authority for a low-stakes choice, may be exactly the right thing to act on. Tier is necessary. It is not sufficient. The next articles take up the facets it leaves out — atomicity and uncertainty — before showing how all of them compose into a single judgment of how much to trust.