Trits: The Distinction Alphabet
From ⊥, meaning is not allowed to be decorative. A "difference" is admissible only if a test can separate it. Otherwise it collapses back into Nothingness.
Meaning = testable difference.
This immediately forces a shape of evidence:
- Evidence can support a claim
- Evidence can oppose a claim
- Evidence can fail to distinguish at this budget
That is already a three-state logic.
The Trit Alphabet: {−1, 0, +1}
Trits are balanced ternary: {−1, 0, +1}.
Not as a compression trick. As an alphabet.
| Value | Meaning | Interpretation |
|---|---|---|
| +1 | Positive evidence | Aligned, verified, supports |
| 0 | No distinction | Irrelevant at this resolution |
| −1 | Negative evidence | Opposed, refuted, contradicts |
Unknown, Nothing, and Irrelevant become a first-class state — not a faint buzz in the mantissa.
Why Trits Are Natural
Trits are not a random alphabet. They reflect the structure of meaningful distinctions:
| Concept | Trit Manifestation |
|---|---|
| Irrelevance | 0 = differences that don't matter are removed |
| Cost | Energy only spent when flipping between 0 and ±1 |
| Direction | Sign distinguishes support (+1) from opposition (−1) |
The semantically relevant distinctions are:
- Positive / negative influence → ±1
- Neutral / irrelevant vs relevant → 0 vs non-zero
That's exactly what the three trit values encode.
Universal Expressivity
As an alphabet, {−1, 0, +1} is universal.
Any finite description — binary, floats, tokens — can be encoded in trits:
- Bits ↔ integers in base 2
- Trits ↔ integers in base 3 (balanced ternary)
- Bijection between finite sequences over {0,1} and finite sequences over {−1, 0, +1}
There is no loss of generality in using {−1, 0, +1} as your primitive alphabet. It's as complete as {0,1} for encoding any finite object.
For any finite dataset and any tolerance ε > 0, there exists a high-dimensional trit embedding that preserves all the geometry you care about to within ε. This follows from Johnson-Lindenstrauss + quantization.
Trits vs Floats
The substrate decides what becomes cheap to represent.
| Floats | Trits |
|---|---|
| Store magnitude | Store distinction |
| Treat silence with same weight as signal | Silence is truly zero |
| Tiny minted differences are cheap | Clean distinctions are cheap |
| Language of statistics (probable noise) | Language of logic (structural invariant) |
The "Almost Zero" Problem
Take a neuron that would fire for "nuclear physics" while the sentence is "The cat sat on the mat."
In float land: It's rarely exactly zero. It becomes 0.000000042. That dust is still fetched, multiplied, accumulated, carried forward. Energy is spent computing precisely how much "nuclear physics" is not happening.
In trit land: Dust snaps to 0. In sparse storage, 0 is not even written down. No index, no fetch, no multiply. Silence stays silent.
Nothingness should be free. Irrelevant things should not leak into actual meaning.
The Semantic Claim
We are not claiming:
"You can compress 32-bit float precision into 1.58 bits (a trit) without loss at fixed dimension."
That's obviously impossible.
We are claiming:
- Alphabet universality — Any finite object can be coded in trits
- Geometric preservation — For finite datasets, trit embeddings can preserve all relevant geometry with enough dimensions
- Semantic naturality — {−1, 0, +1} match the structure of distinctions that matter for truth-seeking
For semantics at the resolution that matters to a truth-seeking agent, −1, 0, +1 is not only sufficient but natural.
Everything more (continuous magnitude) is either:
- Gauge (can be rescaled/rotated without changing truth)
- Finer detail that can be layered as multiple trit planes or extra dimensions
Trits Respect Structural Truth
Why Trits Work Better Than Floats
| Float Problem | Trit Solution |
|---|---|
| Tiny differences are noise | 0 snaps irrelevant distinctions to Nothingness |
| Magnitude encodes non-structural info | Direction (+1/-1) encodes real evidence |
| Dust leaks into downstream computation | Silence (0) has zero computational cost |
Validity Rule
A trit value is valid only if it survives representation changes. If relabeling would change the trit, snap to 0 instead.
Trits encode structural truth. Everything else is Nothingness.
Summary
Floats store magnitude. Trits store distinction.
Trits make "directional evidence" native, they make Nothingness native, and they align representation with the fundamental rule:
Truth only exists where separations exist.
Prerequisites: Gauge-Invariant Truth Machine (GITM)