22.1.3 Extending the workflow from repair to discovery

22.1.3 Extending the workflow from repair to discovery

The six-stage workflow in Chapter 2 is deliberately conservative. During an ordinary run, the declaration fixes what counts as a failure, which perturbations carry meaning, which variations are null, and which repairs may be admitted. Scientific discovery requires a controlled escape from that fixed frame without abandoning the discipline of independent admission.

Two complementary events can initiate such an extension. In the first, persistent anomaly reveals that no repair in the declared language reconciles theory and observation. The obstruction then points beyond a realization \(D\) toward a proposed change \(\Phi :\mathfrak L\to \mathfrak L'\). In the second, a proposed declaration \(\mathfrak L'\) generates a novel consequence: an observation not used to construct the proposal, but specific enough to discriminate it from available alternatives.

These are not merely two additional boxes appended to the workflow. They form an outer discovery loop around it:

Every candidate \(\mathfrak L'\) begins a new six-stage run. It must declare its own paths and probes, expose rather than erase the obstruction that motivated it, and face evidence not controlled by the mechanism that proposed it.

Planck’s black-body analysis illustrates the anomaly-driven direction. New spectral measurements showed that the Wien distribution law lacked general validity. Planck explicitly concluded that the theory required improvement and sought the replaceable link in its derivation. His revised statistical argument introduced finite energy elements with size proportional to frequency, \(\varepsilon =h\nu \), and yielded a radiation law consistent with the measured spectrum [ Planck , 1901 ] . The historical claim should be kept precise: Planck’s step discretized the energy elements used in the resonator calculation; the stronger ontology of freely propagating light quanta came later. For the present purpose, the important feature is the structural leap. A failed relation among spectrum, entropy, and classical derivation was not repaired by another parameter fit. One of the theory’s constitutive assumptions was replaced.

General relativity illustrates the prediction-driven direction. The completed theory implied that starlight grazing the Sun should be deflected by approximately \(1.75\) arcseconds [ Einstein , 1916 ] . A total solar eclipse made the relevant stars observable close to the Sun, turning a theoretical consequence into an experimental design. Expeditions to Príncipe and Sobral observed the eclipse of 29 May 1919; Dyson, Eddington, and Davidson reported measurements consistent with the general-relativistic value [ Dyson et al. , 1920 ] . Einstein had derived an earlier, incomplete light-deflection value in 1911, so the novelty was not simply the idea that gravity bends light. The mature theory supplied the discriminating magnitude and thereby specified a test capable of separating declarations.

The two episodes expose a symmetry that ordinary optimization often hides. Data can obstruct a theory strongly enough to demand a new primitive, while a new primitive can generate observations that did not previously belong to the experimental agenda. Scientific creativity moves in both directions:

LINCS can represent the first arrow as obstruction-driven proposal and the second as probe generation. Neither arrow is self-certifying.