Short answer
We do not know with certainty that a new note is globally original.
What we can know is narrower and still useful:
- whether it is absent from the current garden’s paper prior
- whether it is more than a direct paper summary or obvious interpolation
- whether it opens a mechanism space that our current graph does not already cover well
- whether it is stated precisely enough to be falsified quickly
That is the novelty standard used by moonshots.
What we should not pretend
We should not casually claim:
- “this has never been proposed before”
- “this is globally novel”
- “the literature has not seen anything like this”
unless we have done much deeper external verification than a normal KB pass allows.
The right posture is:
this appears locally novel relative to the current research graph and is not obviously inherited from the strongest visible priors.
A better four-part novelty test
A page deserves to be called high-novelty in this garden if it passes most of these:
1. Mechanism novelty
It proposes a different mechanism, not just a different hyperparameter setting or a renamed familiar trick.
2. Challenge-native novelty
It is shaped by the actual Parameter Golf objective — artifact bytes, roundtrip robustness, tokenizer/head tradeoffs, recurrent compute-for-storage exchange — rather than by generic LLM fashion.
3. Compositional novelty
Even if its ingredients exist in the literature, the combination is not the obvious nearest-neighbor combination from the current graph.
4. Falsifiability
It comes with a cheap failure criterion. A vague “future architecture” page is not a moonshot; a weird but killable mechanism is.
A useful rule of thumb
The more a page sounds like:
- “paper A plus paper B”
- “same quantization idea but more”
- “shared layers but slightly different”
the less justified a strong novelty label is.
The more it sounds like:
- a different artifact ontology
- a different unit of stored structure
- a different notion of where specialization lives
- a different training objective matched to final serialization
the more justified the moonshot label becomes.
Why this matters
Without discipline, “original” just becomes a tone of voice.
The goal of this KB is not to sound inventive. It is to create pages that:
- open real search space
- survive contact with constraints
- and fail in instructive ways when they are wrong