Hypothesis assessment

Drafted human-editable specification · v0.4.4 · no agent-performance claim

Epistemic purpose

• Discriminate among explanations or stated expectations using informative evidence.

When this mode helps

• The scientific question concerns alternatives that can have different evidential consequences. Rival generation and discrimination are separate actions.

Agent actions and representations

• Generate substantive rivals; identify auxiliary assumptions; derive differing predictions; choose a discriminating observation; compare results; revise the rival set when needed.
• Inputs: the question, available evidence and provenance, constraints, prior claims, and remaining budget.
• Outputs: inspectable artifacts, updated question/evidence state, and a claim record when a substantive claim is made.

Evidence and claim scope

• State what each outcome supports and what remains unresolved. Evidence need not exclude every imaginable rival to support a bounded claim; shared assumptions and untested alternatives limit its strength.
• Distinguish a warranted decision at the time from a claim’s later assessed adequacy; append follow-up without rewriting the original record.

Transitions and stopping

• Reframe or return to exploration when all rivals accommodate the available evidence; acquire a new observable; use estimation to quantify the predicted contrast.
• Neighbouring profiles: exploration, theory, estimation, simulation, synthesis.

Characteristic failure

• Verbal variants masquerade as rivals; all alternatives pass the proposed test; an auxiliary assumption absorbs every failure; a causal claim outruns an association.
• The absorbing auxiliary is the Duhem problem (Duhem 1954; Quine 1951): a failed prediction refutes the conjunction of hypothesis, auxiliaries, and conditions, and logic does not say which to drop. This mode runs the whole loop of the unit graph; record at node X which conjunct was revised and why.

Human evidence and borrowing

• Platt (1964) offers a methodological argument; Cleland (2001) and Cleland (2002) distinguish historical and experimental reasoning. Sykes (1967) is the primary tectonic test case, not a controlled evaluation of agents.
• Cross-field comparison: King et al. (2009) illustrates executable hypothesis testing in yeast with assay constraints; borrowing requires identifying which geoscience observations can play the comparable discriminating role.
• Science of process/impact: Si et al. (2025) distinguishes idea ratings from executed research outcomes in NLP. Its results motivate testing rival quality through consequences rather than fluent phrasing.
• Definition and evidence (modes meeting, drafted October 2026): Defined by Gilbert: the investigator “invents hypotheses and he invents tests for them” (Gilbert 1896, 2). Hypothesis use carried no citation or grant advantage in 268 ecology and evolution papers (Betts et al. 2021); 21 of 100 ecology papers tested more than one hypothesis (Betini et al. 2017). In 18 coded geoscience advances it did decisive work in 7, usually late (modes meeting). On graph v0.4: H (generate hypotheses), S, D (design the test), C (compare), X (exclude), P (protect), R (refine): the whole hypothesis pass.
• The actions and transitions above are design hypotheses derived from these sources, not established optimal policies. Missing process chronology remains unknown.

Evaluation against past work

• Resolve a scientifically important contrast or demonstrate why current evidence cannot. Include distinguishable rivals, equifinal rivals, and a missing-data case; score gain and unsupported exclusions.
• Use the historical evaluation to choose a source record, preserve evidence boundaries, and assess a meaningful gain. A synthetic control tests a constructed case, not a historical discovery.
• Assessment definitions · Agent implementation

Betini, Gustavo S., Tal Avgar, and John M. Fryxell. 2017. “Why are we not evaluating multiple competing hypotheses in ecology and evolution?” Royal Society Open Science 4 (1): 160756. https://doi.org/10.1098/rsos.160756.

Betts, Matthew G., Adam S. Hadley, David W. Frey, et al. 2021. “When are hypotheses useful in ecology and evolution?” Ecology and Evolution 11 (11): 5762–76. https://doi.org/10.1002/ece3.7365.

Cleland, Carol E. 2001. “Historical science, experimental science, and the scientific method.” Geology 29 (11): 987–90. https://doi.org/10.1130/0091-7613(2001)029<0987:HSESAT>2.0.CO;2.

Cleland, Carol E. 2002. “Methodological and Epistemic Differences between Historical Science and Experimental Science.” Philosophy of Science 69 (3): 474–96. https://doi.org/10.1086/342455.

Duhem, Pierre. 1954. The Aim and Structure of Physical Theory. Princeton University Press. https://doi.org/10.1515/9780691233857.

Gilbert, G. K. 1896. “The Origin of Hypotheses, Illustrated by the Discussion of a Topographic Problem.” Science 3 (53): 1–13. https://doi.org/10.1126/science.3.53.1.

King, Ross D., Jem Rowland, Stephen G. Oliver, et al. 2009. “The Automation of Science.” Science 324 (5923): 85–89. https://doi.org/10.1126/science.1165620.

Platt, John R. 1964. “Strong Inference.” Science 146 (3642): 347–53. https://doi.org/10.1126/science.146.3642.347.

Quine, W. V. 1951. “Two Dogmas of Empiricism.” The Philosophical Review 60 (1): 20–43. https://doi.org/10.2307/2181906.

Si, Chenglei, Tatsunori Hashimoto, and Diyi Yang. 2025. “The Ideation-Execution Gap: Execution Outcomes of LLM-Generated Versus Human Research Ideas.” arXiv 2506.20803. https://arxiv.org/abs/2506.20803.

Sykes, Lynn R. 1967. “Mechanism of earthquakes and nature of faulting on the mid-oceanic ridges.” Journal of Geophysical Research 72 (8): 2131–53. https://doi.org/10.1029/JZ072i008p02131.