fix(L3): unbreak UnifiedCovariant build + kill stealth-True opaque predicates

Two defects in the pushed Tier-1 Layer-3 work (8c5ee2aa/ba8ee111), both
invisible to single-file LSP checks but caught by a clean `lake build`:

1. BUILD-BLOCKER: `import Mathlib.LinearAlgebra.Matrix.Adjoints` — no such
   mathlib module (the real one is `Matrix.Adjugate`), and nothing in the file
   used it. Dead + wrong import; `lake build SilverSight.PIST.UnifiedCovariant`
   failed with "bad import" until removed. The file did NOT compile on main.

2. HONESTY: `admits_Cartan_connection` / `has_SO_1_6_holonomy` were
   `def ... := Nonempty (M -> M)`, trivially inhabited by `id` -- a stealth-True.
   The docstrings falsely claimed "neither True nor provable." Reverted to
   opaque `axiom ... : Type -> Prop` signatures (assert nothing; tagged
   HONESTY CLASS: OPAQUE). Companion conjectures can now ONLY be closed by
   their CONJECTURE sorry, not a silent `<id>`. Scanner blind spot for this
   pattern (Nonempty (M->M), not literal := True) noted for a follow-up.

Verified: `lake build` green (3299 jobs). Axiom footprint matches the §7
whitelist -- goldenCP7_is_Kaehler = {propext, Classical.choice, Quot.sound,
sorryAx}; Cartan_connection_on_J1_exists adds only the OPAQUE
admits_Cartan_connection. No stray custom axioms.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
allaun 2026-07-03 07:39:13 -05:00
parent 526357e391
commit 35bb1274e0

View file

@ -65,7 +65,6 @@
import Mathlib.Data.Real.Basic
import Mathlib.Data.Nat.Fib.Basic
import Mathlib.Data.Matrix.Basic
import Mathlib.LinearAlgebra.Matrix.Adjoints
import Mathlib.Tactic
import SilverSight.PIST.CartanConnection
@ -319,35 +318,29 @@ theorem cp_FS_Kaehler (n : ) : is_Kaehler (Fin (n+1) → ) := by
A Cartan connection is a principal bundle connection infinitesimally
modeled on a homogeneous space G/P.
HONESTY CLASS: OPAQUE (predicate signature — NOT True, NOT axiom)
BLOCKED ON: Mathlib Cartan geometry API
This is a named Prop with mathematical content (existence of a
connection with specific structure group properties). It is
deliberately NOT `True` — that would make the companion conjecture
vacuously proven via `trivial`, the exact smuggle pattern the
anti-smuggle scanner catches. The companion theorem carries a
CONJECTURE sorry. -/
def admits_Cartan_connection (M : Type) : Prop :=
-- When Cartan geometry API lands, replace with:
-- ∃ (G H : Type) [LieGroup G] [ClosedSubgroup H G],
-- Nonempty (CartanConnection M G H)
-- For now: an opaque Prop that is neither True nor provable.
-- It references a hypothetical structure, making it impossible to
-- close with `trivial` or `rfl`.
Nonempty (M → M) -- placeholder: real content needs Cartan API
HONESTY CLASS: OPAQUE (axiom predicate signature — asserts nothing).
BLOCKED ON: Mathlib Cartan geometry API.
Declared as an opaque `axiom` predicate, NOT a `def`. Every concrete
`def` body writable today (e.g. `Nonempty (M → M)`) is *trivially
inhabited* — closable by `⟨id⟩` — which would let the companion
conjecture be discharged vacuously, a silent smuggle the scanner
cannot see (it is not literally `:= True`). An opaque axiom symbol
carries no content, so the ONLY proof of the companion conjecture is
its CONJECTURE `sorry`. When Mathlib ships Cartan geometry, replace
with `∃ (G H : Type) [LieGroup G] [ClosedSubgroup H G],
Nonempty (CartanConnection M G H)`. -/
axiom admits_Cartan_connection (M : Type) : Prop
/-- OPAQUE: M has holonomy contained in SO⁰(1,6).
The pseudo-orthogonal group SO⁰(1,6) preserves a quadratic form of
signature (1,6).
HONESTY CLASS: OPAQUE (predicate signature — NOT True, NOT axiom)
BLOCKED ON: Mathlib holonomy + indefinite orthogonal group API
Same treatment as admits_Cartan_connection. NOT True. -/
def has_SO_1_6_holonomy (M : Type) : Prop :=
-- When holonomy API lands, replace with:
-- HolonomyGroup M ⊆ SO(1,6)
-- For now: an opaque Prop referencing M's structure.
Nonempty (M → M) -- placeholder: real content needs holonomy API
HONESTY CLASS: OPAQUE (axiom predicate signature — asserts nothing).
BLOCKED ON: Mathlib holonomy + indefinite orthogonal group API.
Same treatment as `admits_Cartan_connection`: an opaque axiom, not a
secretly-inhabited `def`. Replace with `HolonomyGroup M ⊆ SO(1,6)`
when the API lands. -/
axiom has_SO_1_6_holonomy (M : Type) : Prop
-- ============================================================
-- §6 THE GEOMETRIC CONJECTURES (honestly labeled)
@ -398,9 +391,15 @@ theorem holonomy_is_SO_1_6 : has_SO_1_6_holonomy openSimplex7 := by
-- ============================================================
-- §7 AXIOM INVENTORY (anti-smuggle witness)
-- ============================================================
-- After the Tier 1 conversion, the custom axiom set should be EMPTY.
-- All former axioms are now either:
-- - def (is_Kaehler, admits_Cartan_connection, has_SO_1_6_holonomy)
-- Custom axiom set after the Tier-1 conversion:
-- - is_Kaehler → def (Nonempty (KählerManifold V), real content)
-- - admits_Cartan_connection → OPAQUE axiom (predicate signature, asserts nothing)
-- - has_SO_1_6_holonomy → OPAQUE axiom (predicate signature, asserts nothing)
-- The two opaque axioms are honest: a predicate SYMBOL with no body asserts
-- nothing, so the companion conjectures can only be closed by their sorry.
-- (Earlier these were `def … := Nonempty (M → M)`, which is trivially
-- inhabited by `id` — a stealth-`True`; reverted to opaque axioms.)
-- Proven / conjectured statements:
-- - theorem with sorry (cp_FS_Kaehler, Cartan_connection_on_J1_exists,
-- holonomy_is_SO_1_6)
--
@ -410,7 +409,8 @@ theorem holonomy_is_SO_1_6 : has_SO_1_6_holonomy openSimplex7 := by
-- - holonomy_is_SO_1_6 (CONJECTURE: 1+6 observer split)
--
-- Whitelist of acceptable axioms after lake build:
-- propext, Classical.choice, Quot.sound, sorryAx
-- propext, Classical.choice, Quot.sound, sorryAx,
-- admits_Cartan_connection, has_SO_1_6_holonomy (OPAQUE predicate signatures)
-- Any axiom outside this whitelist is a smuggle.
-- #print axioms -- uncomment after lake build to verify