A record whose update rule is uniform says exactly what was logged
ProvedTranscendenceTower.UniformityPlatform.record_auditA record whose update rule treats all propositions alike says exactly what was logged. Every fact in it is attributable to a revision, and nothing in it is lost silently.
Read the contexts of a lifting schema as the revisions of a record. At revision a set of entries is logged; entries lie in the base language, since they are claims about the world rather than about the record. Write for the entries logged at or before , let be a fixed update schema with consequents , and let be an extension of .
Four parts.
-
Audit. If is proposition-uniform and is consistent, then a base sentence lies in the extension exactly when it follows from the entries. The update rule contributes nothing about the world.
-
Attribution. If a base sentence is present at revision and absent at revision , then it follows from the entries logged up to and not from those logged up to . Some entry between the two revisions is therefore necessary for it: no fact appears that no revision accounts for.
-
No silent loss. For , every base sentence of lies in . The record's content about the world only grows.
-
A monotone schema's extension is not . With the tautological base and the single rule , the extension is , which does not contain , while does. A monotone rule still carries a prerequisite, and a rule whose prerequisite is never derived never fires.
Part 4 is included because the identity it refutes is the natural bridge between the two halves of this subject, and stating the audit property without it invites the reader to prove it a shorter way that does not work.
What is not assumed. The schema need not be monotone: the practically interesting update rules — carry the previous value forward unless something supersedes it — are covered. The base theory need not be symmetric, which matters, because a log is a specific claim about the world and could never satisfy a symmetry hypothesis. Non-uniqueness of the extension is not an obstacle either: the statement holds of every extension, so what multiple extensions cost is reproducibility rather than auditability.
Limits. Consistency of is a hypothesis, and a necessary one: a log that contradicts what the update rule could ever say is a record that proves anything. Context-freeness of the entries is likewise necessary, and so is proposition-uniformity — a schema naming a particular sentence is an assertion in disguise, and it is the assertion rather than the record that would be doing the work.
import Definitions.Def_TranscendenceTowerContextUniformity set_option autoImplicit false open TranscendenceTower.UniformityPlatform
theorem TranscendenceTower.UniformityPlatform.record_audit (A C : Type) :
(∀ (T : Set (CForm A C)) (S : Set (Rule A C)) (E : Set (CForm A C)),
(∀ p ∈ T, IsBase p) → PropUniformRules S →
Consistent (T ∪ {γ | ∃ r ∈ S, r.conseq = γ}) →
IsExtension T S E →
∀ φ : CForm A C, IsBase φ → (φ ∈ E ↔ φ ∈ Cn T))
∧ (∀ (Log : ℕ → Set (CForm A C)) (S : Set (Rule A C))
(Ext : ℕ → Set (CForm A C)),
(∀ n, ∀ p ∈ Log n, IsBase p) → PropUniformRules S →
(∀ n, Consistent ({p | ∃ i ≤ n, p ∈ Log i} ∪ {γ | ∃ r ∈ S, r.conseq = γ})) →
(∀ n, IsExtension {p | ∃ i ≤ n, p ∈ Log i} S (Ext n)) →
∀ (m n : ℕ) (φ : CForm A C), IsBase φ → φ ∈ Ext n → φ ∉ Ext m →
φ ∈ Cn {p | ∃ i ≤ n, p ∈ Log i} ∧ φ ∉ Cn {p | ∃ i ≤ m, p ∈ Log i})
∧ (∀ (Log : ℕ → Set (CForm A C)) (S : Set (Rule A C))
(Ext : ℕ → Set (CForm A C)),
(∀ n, ∀ p ∈ Log n, IsBase p) → PropUniformRules S →
(∀ n, Consistent ({p | ∃ i ≤ n, p ∈ Log i} ∪ {γ | ∃ r ∈ S, r.conseq = γ})) →
(∀ n, IsExtension {p | ∃ i ≤ n, p ∈ Log i} S (Ext n)) →
∀ m n : ℕ, m ≤ n → ∀ φ : CForm A C, IsBase φ → φ ∈ Ext m → φ ∈ Ext n)
∧ (∀ a : A, ∃ (S : Set (Rule A C)) (E : Set (CForm A C)),
Monotone S ∧ IsExtension (∅ : Set (CForm A C)) S E ∧
E ≠ Cn ((∅ : Set (CForm A C)) ∪ {γ | ∃ r ∈ S, r.conseq = γ})) := by sorry