Novelty Engines

Novelty Engines

The novelty engines are the core differentiator of Sisyphus Academica. Each thinks from a completely different cognitive frame. Most papers use one "gap analysis" approach. This system uses 6.

The Contrarian

File: novelty-engines/contrarian.md

Method: Read the literature, extract every "well-established" claim, invert it.

Original Claim	Inverted Hypothesis
"X causes Y"	"Y causes X" or "Z causes both"
"Method A is optimal"	"Method A works for unrelated reasons"
"Dataset D supports generalization"	"Dataset D has systematic biases"

Output: 10 counter-hypotheses ranked by impact_if_true × plausibility.

The Cross-Pollinator

File: novelty-engines/cross-pollinator.md

Method: Import solutions from 15 completely unrelated fields:

1. Astrodynamics — orbital mechanics, trajectory optimization
2. Epidemiology — disease spread modeling
3. Portfolio theory — risk optimization
4. Thermodynamics — entropy, free energy
5. Linguistics — grammar formalisms
6. Ecology — predator-prey models
7. Economics — game theory
8. Neuroscience — synaptic plasticity
9. Materials science — phase diagrams
10. 15th century shipbuilding — empirical design optimization
11. Music theory — harmonic progression
12. Urban planning — traffic flow
13. Immunology — adaptive systems
14. Geology — deep time processes
15. Philosophy of science — paradigm shifts

Output: Top 5 cross-domain analogies with novelty and transferability scores.

The Assumption Excavator

File: novelty-engines/assumption-excavator.md

Method: Read 20+ papers and extract assumptions at 3 levels:

1. Explicit — stated in the paper ("We assume i.i.d. data")
2. Implicit — never stated but clearly present ("The benchmark rewards the right thing")
3. Foundational — so basic the field doesn't state them ("Mathematical modeling can capture this phenomenon")

Output: Assumption tree diagram + top 5 assumptions that could plausibly be false, with proposed tests.

The Counterfactual Generator

File: novelty-engines/counterfactual-generator.md

Method: Build the field's history as a DAG, identify branching points, rewrite history.

Example counterfactual: "What if the Transformer had FAILED on WMT 2014?"

• 2018: No BERT paper
• 2019: LSTM research continues
• 2020: Sparse gated architectures become dominant
• 2022: Much less attention to attention mechanisms

Output: 5 counterfactual histories with traced consequences and testable hypotheses.

The Paradox Sifter

File: novelty-engines/paradox-sifter.md

Method: Extract every "Limitations" and "Future Work" sentence from the entire literature corpus. Cross-reference them.

Finds:

• Contradictions: Paper A says X is solved, Paper C says X is unsolved — what happened?
• Elephants in the room: Problems that appear in >50% of limitations sections but nobody tackles
• Paradoxes: Simultaneously held contradictory beliefs

Output: Paradoxes found, elephants identified, contradictions resolved.

The Heretic (Crown Jewel)

File: novelty-engines/heretic.md

Method: Read ONLY the title and abstract. Generate 50 wild hypotheses. Then read the full paper. Find the gap between what the paper IS and what it COULD HAVE BEEN.

50 hypotheses in 5 categories:

#	Category
1-10	Methodological — what if they used a different approach?
11-20	Theoretical — what if the theory is wrong?
21-30	Empirical — what if results show the opposite?
31-40	Foundational — what if a key assumption is false?
41-50	Wild cards — ideas that would never appear in a review

The Haunting Idea: One hypothesis the authors should have explored but didn't. Novelty gap ≥ 8, tractability ≥ 5, impact ≥ 7, surprise ≥ 8.

Output: 50 scored hypotheses + the haunting idea.