ifc-0235

17.2 Two directions between theory and evidence

Two famous episodes show why the output of document-grounded construction should be an admission interface, not merely another document. The interface specifies how a claim could meet the world: an observation protocol, a controlled intervention, a simulation, or a manipulable physical model.

General relativity: theory constructs a measurement.

Einstein presented the field equations of general relativity in 1915 and a systematic exposition the following year [ Einstein , 1916 ] . The theory implied a quantitative displacement of apparent stellar positions near the Sun. A total solar eclipse made the otherwise invisible comparison observable by blocking the Sun’s glare. During the eclipse of 29 May 1919, British expeditions collected photographic plates at Príncipe, off the west coast of Africa, and Sobral, Brazil. Eddington and Cottingham observed at Príncipe; a parallel team observed at Sobral. The reported analysis agreed with Einstein’s predicted light deflection and was announced publicly that November [ Dyson et al. , 1920 ] .

The methodological point is independent of later debates about the precision of those plates. A theoretical construction generated a new observable, the conditions under which it could be measured, and outcomes capable of distinguishing it from a competing theory. In the terminology of this chapter, the theory supplied both a consequence and an acquisition policy:

general-relativistic theory \(\longrightarrow \) predicted stellar displacement \(\longrightarrow \) eclipse measurement \(\longrightarrow \) evidential update.

DNA: evidence constructs a manipulable theory.

The route to the double helix ran in the other direction. Chemical knowledge, Chargaff’s base ratios, and X-ray diffraction work by Rosalind Franklin, Raymond Gosling, Maurice Wilkins, and their colleagues constrained the possible geometry. Pauling and Corey proposed a three-chain structure with phosphates near the axis, a model Watson and Crick criticized on chemical and geometric grounds [ Pauling and Corey , 1953 , Watson and Crick , 1953 ] . Watson and Crick then used model building—including movable representations of molecular components—to search configurations satisfying bond angles, distances, helical symmetry, and the empirical constraints. Complementary base pairing made the two-chain structure cohere and immediately suggested a copying mechanism [ Watson , 1968 , Watson and Crick , 1953 ] . The result depended on the experimental contributions of Franklin, Gosling, Wilkins, and others; the physical model was a device for composing those constraints, not a substitute for them [ Franklin and Gosling , 1953 ] .

This model was not a dynamical simulator in the sense of Chapter 16. It was an executable structural theory: components could be moved, rejected when they violated geometry or chemistry, and retained when many constraints closed simultaneously. Here the direction was

measurements and chemical constraints \(\longrightarrow \) manipulable model \(\longrightarrow \) double-helix theory \(\longrightarrow \) new biological consequences.

These examples refine the chapter’s constructive goal. Given a scientific corpus, the system should not merely extract its causal claims. It should construct the smallest source-bearing theory that organizes them and then compile that theory into an external test. Depending on the domain, the test may be an Eddington-style measurement under rare conditions, a controlled intervention, a numerical simulation, or a Watson–Crick-style model whose components realize the proposed structure. When no such interface can be specified, the system has produced an interpretation, not yet a testable scientific theory.