Face B0: Explicit Confidence Tiers (Draft)
Label what you know by how you know it.
Logical derivations, convergent evidence, and personal testimony are different kinds of warrant. They deserve different kinds of confidence. The honest thinker keeps them clearly labeled. Collapsing them into one register — presenting everything with the same certainty — is a form of intellectual dishonesty, even when it makes the argument feel stronger.
Not all knowledge is the same kind of knowledge. A theorem derived by valid proof carries a different warrant than a claim supported by strong convergent evidence from independent sources, which carries a different warrant than a claim based on personal experience and reflection. These are different in kind, not just in degree. The rigorous thinker maintains all three distinctions explicitly, in public, as a discipline of honest communication.
The failure mode this prevents: false confidence. It is easy to present a claim with the tone of certainty when the actual warrant is something weaker — to present convergent evidence as proof, or personal experience as convergent evidence. The audience, lacking meta-information about what kind of warrant is in play, cannot correctly calibrate their own confidence. Explicit labeling is not hedging — it is the honest account of what is actually known and how.
The tiers need not be formal or elaborate. “I know this,” “I think this is likely,” and “I’m working on whether this is true” are already three meaningfully different registers. The discipline is to maintain them honestly, especially when social pressure runs toward collapsing them into a single confident register.
In practice:
A friend asks you for advice about a medical decision. You find yourself saying: “Definitely do X — I’ve heard it’s much better.”
Slow down. Definitely is the register of logical necessity. I’ve heard is the register of hearsay. You used one to prop up the other.
The honest version: “I’ve heard good things about X — I don’t know the research well, but my sense from what I’ve read is that it’s worth asking your doctor about.” That’s an accurate account of what you actually know. It lets your friend calibrate appropriately. It doesn’t make you sound less confident in any meaningful way — it makes what you say more trustworthy, because you’re telling them which register you’re speaking from. Before you speak, notice which tier you’re in. Then speak from that tier.
Formal Statement (Concordius Framework)
Typed warrant. A claim’s warrant is a pair (type, magnitude), not a scalar, and the type fixes the update law and is not discardable:
- deductive — entailment from the primitives, P = 1 conditional on the axioms, closed under modus ponens, brittle (one false premise voids it);
- convergent — a posterior assembled from independent likelihoods (Face C1), soft and accumulating;
- testimonial — a single likelihood increment (Face C0), soft and easily overturned.
A deductive 0.9 and a testimonial 0.9 are different objects: they respond differently to new evidence. Collapsing the pair to its magnitude — presenting everything at one confidence — is a type error, and it is exactly TruthChecker’s OVERCLAIM (a concordance dressed as a derivation, a granted premise dressed as established). The types map onto the constraint hierarchy: Tier 1 derivations from the cogito + Gelfand triple (τ → 1, the anchor nodes of the dependency graph, Appendix F §2); Tier 2 concordance from independent witnesses; Tier 3 the practitioner’s not-yet-confirmed working-out.
Tier: derivation (as a discipline) — keeping a warrant’s type (its update operator), not merely its number; the corpus’s own confidence-tier rule made explicit.
Face B1: Logic Where It Reaches; Concordance Where It Doesn’t →