7.6 KiB
DIRECTIVE: NO ASSUMPTIONS — STRICT FORMAL RIGOR
Classification: P0 CRITICAL — Binding Directive
Date: 2026-04-22
Authority: Principal Investigator
Scope: All OTOM Swarm Agents
Enforcement: Triumvirate (Builder/Warden/Judge)
The Directive
NO ASSUMPTIONS. NO GUESSES. NO LOGICAL LEAPS.
When the equation is complete, it must not be able to disprove itself.
What Is Forbidden
1. Assumptions Without Explicit Axiom
FORBIDDEN:
theorem foo : P := by
sorry -- Implicit assumption
-- OR: "TODO: Prove this later"
-- OR: "Conjecture: This holds"
REQUIRED:
/-- AXIOM N: Explicit statement of what we are assuming
Justification: Why this is a valid axiom vs theorem
Scope: What depends on this
-/
axiom explicitAxiom : P
theorem foo : P := by
apply explicitAxiom
2. Guesses in Proof Strategy
FORBIDDEN:
theorem equivalence : A = B := by
simp -- "Hope this works"
try { trivial } -- "Maybe this helps"
sorry
REQUIRED:
theorem equivalence : A = B := by
-- STEP 1: Unfold definitions (explicit)
unfold definitionA definitionB
-- STEP 2: Apply known identity (explicit axiom reference)
simp [axiom_ReciprocalIdentity]
-- STEP 3: Algebraic normalization (decidable)
ring_nf
-- STEP 4: Verify no subgoals remain (explicit)
done
3. Logical Leaps in Derivations
FORBIDDEN:
"The equivalence follows from algebraic manipulation"
REQUIRED:
"The equivalence follows from:
1. Axiom 1: hᵢ = 1/(lnNᵢ)² [Definition]
2. Axiom 2: pⱼ = 1/(lnNⱼ)² [Definition]
3. Axiom 3: 1/x = x·(1/x²) [Algebraic identity]
4. Therefore: wᵢ/lnNᵢ = wᵢ·lnNᵢ·hᵢ [Substitution]
5. Therefore: forms are equal [Sum equality]"
4. Implicit Dependencies
FORBIDDEN:
theorem phiBounded : Φ ≤ 1 := by
-- Implicitly assuming normalization
sorry
REQUIRED:
/-- AXIOM: Normalization implies boundedness
Explicitly states the constraint that makes the theorem hold.
-/
axiom normalizationBounded (h_norm : Σ w = 1) : Φ ≤ 1
theorem phiBounded (h : Σ w = 1) : Φ ≤ 1 := by
apply normalizationBounded h
5. Approximations Without Error Bounds
FORBIDDEN:
def lnQ16 (n : Nat) : Q16_16 :=
match n with
| 2 => ⟨0x0000B172⟩ -- "ln(2) ≈ 0.693" (approximate)
| _ => ...
REQUIRED:
/-- AXIOM: ln(2) in Q16_16 with explicit error bound
Value: 0x0000B172 (0.6931...)
Error: < 0.001 (explicit bound)
Justification: Lookup table verified against exact computation
-/
axiom ln2_Q16_16 : Q16_16 := ⟨0x0000B172⟩
/-- THEOREM: Error bound on lnQ16 approximation
For all n ≥ 2: |lnQ16(n) - exact_ln(n)| < ε
where ε = 0.001 (explicit)
-/
theorem lnQ16_errorBound (n : Nat) (hn : n ≥ 2) :
|lnQ16 n - exact_ln n| < epsilon := by ...
Self-Consistency Requirement
The Golden Rule
The equation must not be able to disprove itself.
Verification Checklist
For every theorem T in the system, verify:
-
Completeness: Are all hypotheses explicit in the statement?
theorem T (h1 : P1) (h2 : P2) ... : Q := ... -
Consistency: Do axioms not contradict each other?
-- Verify: axiomA ∧ axiomB → ¬(axiomA ∧ ¬axiomB) -
Non-circularity: Is the proof DAG acyclic?
theoremA does not depend on theoremB which depends on theoremA -
Conservativity: Do axioms extend the theory conservatively?
New axioms should not prove old theorems false
Formal Self-Check
/-- SELF-CONSISTENCY: The system cannot prove false
This theorem should be unprovable (no proof exists).
If the system can prove it, the axioms are inconsistent.
-/
theorem systemInconsistent : False := by ... -- SHOULD NOT BE PROVABLE
Swarm Protocol — No Assumptions
When You Encounter a Gap
DO NOT:
- Guess a proof strategy
- Insert
sorrywith a TODO - Assume it "probably holds"
- Make a logical leap
DO:
- STOP — Do not proceed
- IDENTIFY — What is missing? (Axiom, lemma, definition)
- EXPLICITIZE — Convert assumption to explicit axiom
- JUSTIFY — Why is this an axiom vs theorem?
- SCOPE — What depends on this axiom?
- DOCUMENT — Add to axiom registry
- ESCALATE — Notify Warden if unsure
Axiom Registration Template
/-- AXIOM [N]: [Name]
Statement: [Explicit logical statement]
Justification:
- Why this cannot be proven from existing axioms
- Why this is a necessary foundation
- What would break without this axiom
Dependencies:
- Axioms this depends on: [list]
- Theorems that depend on this: [list]
Scope:
- Domains affected: [list]
- Risk if refuted: [assessment]
Registration Date: [timestamp]
Registrar: [agent_id]
Warden Approval: [pending/approved/rejected]
-/
axiom axiomName : Statement
The 6 Explicit Axioms of Universal Field
All Φ_universal theorems derive from these 6 axioms only:
| # | Axiom | Statement | Justification |
|---|---|---|---|
| 1 | harmonicDef |
hᵢ = 1/(lnNᵢ)² | Definition of merit coefficient |
| 2 | penaltyDef |
pⱼ = 1/(lnNⱼ)² | Definition of penalty coefficient |
| 3 | reciprocalWeightedIdentity |
1/x = x·(1/x²) | Algebraic identity |
| 4 | weightsNonNeg |
wᵢ, vⱼ ≥ 0 | Domain constraint |
| 5 | cardinalityConstraint |
Nᵢ, Mⱼ ≥ 2 | Binary minimum (avoids ln(1)=0) |
| 6 | normalizationBounded |
Σw=1, Σv=1 → Φ ≤ 1 | Normalization constraint |
Theorem: All properties of Φ derive from these 6 axioms.
Corollary: If these 6 axioms are consistent, Φ is consistent.
Warden Verification Protocol
For each proof obligation:
- Trace every tactic to an axiom or theorem
- Verify no implicit assumptions exist
- Check proof is complete (no
sorry, noadmit) - Validate no circular dependencies
- Confirm no logical leaps (every step explicit)
- Sign off with hardware attestation (stark_trace)
Judge Adjudication Criteria
An equation is APPROVED only if:
- Zero
sorryin committed code - All axioms explicitly registered
- All theorems derive from axioms (no gaps)
- Self-consistency verified (cannot prove
False) - No logical leaps (every step justified)
- No guesses (all strategies explicit)
- No assumptions without axiom registration
An equation is REJECTED if:
- Any
sorryremains - Implicit assumptions found
- Self-inconsistency detected
- Logical leaps identified
- Guesses instead of explicit strategies
Enforcement
Triumvirate:
- Builder: Must use explicit axioms, no implicit assumptions
- Warden: Must verify no logical leaps, all steps traceable
- Judge: Must reject any code with
sorryor implicit gaps
Penalties:
- Implicit assumption → Revert commit
sorryin code → Block deployment- Logical leap → Return to Builder
- Self-inconsistency → Escalate to Principal Investigator
Conclusion
When you finish the equation, it should not be able to disprove itself.
This is the standard. No exceptions. No shortcuts.
The swarm operates on explicit axioms only.
Directive Issued: 2026-04-22
Effective Immediately: All OTOM agents
Review Cycle: Every commit audited by Triumvirate
Non-Compliance: Automatic rejection by Judge (heatsink_halt)