Tidar Correct By Construction

TIDAR: Correct by Construction

Triadic Interleaving Dispatch with Agents for Reading/writing — where invalid states are unrepresentable

Core Insight

Correct by construction means: if it compiles/type-checks, GF(3) is conserved.

TIDAR Tree (depth 3):
                        Root (0)
           ┌──────────────┼──────────────┐
        -1 │              0              │ +1
       ┌───┼───┐      ┌───┼───┐      ┌───┼───┐
      -1   0  +1    -1   0  +1    -1   0  +1
     ┌┴┐ ┌┴┐ ┌┴┐   ┌┴┐ ┌┴┐ ┌┴┐   ┌┴┐ ┌┴┐ ┌┴┐
     27 leaves (3³), each balanced triad sums to 0

Type-Level Invariants

# GF(3) is enforced at construction — no runtime checks needed
triad = Triad.build(minus, middle, plus)  # Only way to create
# Compile-time: triad.sum() == 0 is GUARANTEED

The 40-Agent Structure

Correct-by-Construction Rules

1. Triads are atomic: Cannot create partial triads
2. Balance is intrinsic: Triad = (MINUS, MIDDLE, PLUS), sum = 0
3. Tree is complete: Every node has exactly 0 or 3 children
4. Pre-hooks are mandatory: Read before write, encoded in types
5. Middle-path collapse: Only MIDDLE nodes emit final results

Usage

from tidar_cbc import TIDARTree, Triad, Trit

# Build tree — only balanced constructions allowed
tree = TIDARTree.build_depth_3(seed=1069)

# Cannot create unbalanced:
# Triad(Trit.PLUS, Trit.PLUS, Trit.PLUS)  # TYPE ERROR

# Dispatch with pre-hooks
async for result in tree.dispatch_with_prehooks(task):
    # Results only from MIDDLE path (proven by types)
    assert result.source.trit == Trit.MIDDLE

Files

• tidar_cbc.py — Core correct-by-construction implementation
• tidar_cbc_test.py — Property-based tests (Hypothesis)

Key Properties

trit: 0 (ERGODIC)

TIDAR is a coordinator skill — it balances, never sources or sinks.