Lakatos's methodology of research programmes

Lakatos’s methodology of scientific research programmes is an account, from the philosophy of science, of how a body of scientific work is judged over time rather than one hypothesis at a time. A programme keeps a hard core of assumptions fixed and surrounds it with auxiliary hypotheses that are revised when predictions fail. It counts as progressive while its revisions lead to new predictions, some of which are then confirmed, and as degenerating while its revisions only absorb failures after the fact. The Broken-Born program describes its own status in these terms, “a Lakatosian research program, not a theory”, and checkpoint #11 closes with an assessment sorted into hard core, progressive signs, degeneration risks and falsification exposure.

Status: Established — [T], explained here from general knowledge; no source document cites Lakatos’s own work, so the references below are marked †. The program’s use of it is the self-assessment of #11 §15, a record rather than a result (Assessment: a live research program).

The problem it answers#

Karl Popper proposed that what makes a theory scientific is that it can be refuted: it rules out some observable outcomes, and a theory whose prediction fails should be given up [T]. Two difficulties with the simple form of this rule motivated Lakatos [T]:

  • A prediction never follows from one hypothesis alone. It follows from the hypothesis together with auxiliary assumptions, initial conditions and a theory of the instruments (the point is usually credited to Pierre Duhem). A failed prediction shows that something in that whole set is false, not which member. Scientists therefore have a choice about where to place the blame, and they exercise it.
  • Every live theory has anomalies. Taken literally, the simple rule would have discarded most successful theories early in their development.

Imre Lakatos’s answer, in “Falsification and the Methodology of Scientific Research Programmes”†, was to change the unit of appraisal from a single theory to a programme: a sequence of theories T₁, T₂, T₃, … that share a fixed core and differ in their auxiliary assumptions. The question is no longer whether one theory has been refuted, but whether the sequence is getting better or worse at anticipating the world [T].

Hard core, protective belt and the two heuristics#

A programme has four parts in Lakatos’s account [T]:

  • The hard core: the assumptions that define the programme. Its adherents decide, as a matter of method, not to revise them; a programme that gives up its core has become a different programme.
  • The negative heuristic: that decision itself. Refutations are not directed at the core.
  • The protective belt: the auxiliary hypotheses around the core. When a prediction fails, the belt is adjusted: an assumption about initial conditions is changed, a new body or field is posited, a parameter is reinterpreted.
  • The positive heuristic: the programme’s plan for developing the belt, a sequence of increasingly realistic models to be built, with some anomalies anticipated in advance and set aside until the plan reaches them.

The standard illustration is the Newtonian programme [T]. When the orbit of Uranus departed from Newtonian prediction, the core (the laws of motion and gravitation) was kept and the belt was changed: John Couch Adams and Urbain Le Verrier posited an unseen planet, computed where it had to be, and Neptune was found close to that position.

Progressive and degenerating problem shifts#

Each change to the belt produces a new theory in the sequence, and Lakatos called the step a problem shift. He graded it by what it adds [T]:

  • Theoretically progressive: the new theory predicts some novel fact, one that its predecessor did not predict or ruled out.
  • Empirically progressive: some of that new content is then confirmed. The Neptune step was both.
  • Degenerating: the change accommodates the anomaly and predicts nothing new. Such changes are ad hoc: they save the core at the cost of the programme’s content.

What counts as a novel fact was refined later. Elie Zahar and John Worrall counted a fact as novel for a theory if it was not used in building that theory, even if it was already known†. On either reading, a result the programme was constructed to reproduce does not count in its favour.

A programme is appraised by its record over many steps, and no single experiment ends it at once. On Lakatos’s account a degenerating programme is displaced when a rival accounts for its successes and predicts more; continuing to work on a degenerating programme can be rational, provided its record is stated honestly [T].

How the program uses the framework#

A research program, not a theory#

The program’s status line comes from this framework. The root CLAUDE.md and HANDOFF.md §1 both give it as “a Lakatosian research program, not a theory”. The plain-language passage of #11 §15 ends: “That is a live research program; the word “theory” still waits on items 1–3 above”, which are the first three open problems of #11 §14: the wedge reduction, the partially-massless comparison and the Lorentzian dressing. The v1 paper had put it as “The word “theory” awaits items 1–2 of §10; the phrase “research program,” used precisely, is earned now” (#9 §11).

The hard core#

#11 §15 names it: “hard core (spontaneously broken reciprocity, ω-only rigidity) intact, twice sharpened by forced corrections (P2′; the w-dictionary)”. It is the program’s hard core, stated as postulates P1–P5. The two corrections are P2′, which replaced the first rigidity postulate after the program’s own geometry showed it reduced the theory to general relativity, and the corrected duality dictionary w_eff = −w̃. In Lakatos’s scheme the core is held fixed by decision [T]. #11 presents its two changes as forced by the program’s own derivations, a more exact statement of the core, not a response to a failed observation.

The assessment’s four headings#

The v1 paper stated the criterion in one line: “Programs that keep predicting novel facts are progressive; programs that only redescribe are degenerating” (#9 §11, superseded as reference by #11 §15). #11 §15 sorts the record under headings that follow Lakatos’s grades (the full lists and their evidence are on Assessment: a live research program):

  • Progressive signs: “forced convergence onto consistency islands; predictions not previously posed (rows 3, 11); an unsolicited identity (§7.3); one over-constraint (DE sector); the §11.3 unification generating new questions”, DE being dark energy. Rows 3 and 11 of the predictions table, the island discriminant and dark-sector microstructure, are the program’s candidates for novel content in Lakatos’s sense.
  • Degeneration risks: “tunable wave-island collective sector (quarantined); gate tail not yet forced; the novelty demotion of §7.4 as a standing caution, joined in v2.1 by the §12 soliton obituary and the §5 re-attribution of the compatibility theorem; DESI DR2’s phantom crossing against §10’s strict w ≥ −1”, DESI DR2 being Data Release 2 of the Dark Energy Spectroscopic Instrument.
  • Falsification exposure: “genuine and near-term (rows 1, 4, 7)”.

In Lakatos’s terms (the wiki’s reading [T]), every item #11 lists as progressive is a structural result or a prediction not yet tested, so the progress it claims is theoretical. Empirical progress would need one of the exposed rows to be confirmed, and for row 4, the dark-energy drift, the current data lean the other way (#11 §10, §15).

Novelty and its demotion#

Novelty carries the weight in Lakatos’s grading, and the program’s record shows two claims losing it. A literature search placed the vacuum-energy identity of §7.3 next to known work, and #11 downgraded it “from “likely standalone result” to “requires detailed comparison with the PM and exchange-symmetric literature,”” (#11 §7.4, PM being partially massless; the vacuum-energy identity). Review round 3 identified the compatibility theorem as a result of Freidel, Leigh and Minic, and v2.1 re-tagged it [T][V]. #11 §15 lists both under degeneration risks.

Degeneration flags in the calc notes#

The calc notes used the vocabulary while the work was under way:

  • After the tuning test, notes/debt1a-iii-tuning-test.md §5: “three structural tests in a row (multi-graviton no-go, signature obstruction, tuning test) could each have killed the geometric sector outright; each instead forced it into the unique known consistent structure at that junction. Repeated forced convergence onto consistency islands is the opposite of degeneration.”
  • On the wave island of Debt 3, where the coupling g of the gated self-interaction is left free (Scenario B): “In Lakatos terms from the earlier discussion: on this branch the granularity prediction is no longer a novel requirement of the program, only an option — a degeneration flag.” (notes/debt3b-wave-island.md, ‘What pins g?’; Scenario B: tunable streams).
  • The closing note of Debt 3 calls velocity-space clumping “merely optional elsewhere (Scenario B, ten tunable orders of magnitude — a degeneration flag, not a prediction)” (notes/debt3c-ufd-endpoint.md; the net theorem). #11 §15 still carries it as the “tunable wave-island collective sector (quarantined)”.

A coupling that can be set to make an effect appear or vanish across ten orders of magnitude lets the program accommodate either outcome, which in Lakatos’s sense is the mark of an ad hoc adjustment (the wiki’s reading [T]).

The declined rescue of the soliton candidate#

After review round 3 found the soliton candidate’s abundance misnormalized, one way to keep the candidate was a free dilution factor “of at least 10³–10⁴ (ruling 2), probably far more (ruling 3)”. Decision 2 rejected it, because the mass “would then no longer be a prediction”. The other route, a symmetry-breaking scale v ≲ 100 PeV, works but was noted as prior art and not claimed: “claiming it would add no testable content and would misattribute it” (review/round3-adjudication.md, ‘Decisions’; Round-3 decisions and lessons; Obituary: superheavy conjugate solitons). In Lakatos’s vocabulary (the wiki’s reading), the first route was an ad hoc belt adjustment and the second would have counted borrowed content as the program’s own; both were declined.

The referees’ recurring objection that “the framework does no work” asks, in the same vocabulary, whether the hard core does heuristic work: whether the program’s results follow from it or are imported from established physics (the wiki’s reading). Round-3 Decision 4 made the toy Lagrangian the next task because “The toy action is the only item that can change that”.

Used by#

References#

Added by the wiki from model memory (†); no program document cites them.

  • I. Lakatos, “Falsification and the Methodology of Scientific Research Programmes,” in I. Lakatos and A. Musgrave (eds.), Criticism and the Growth of Knowledge, Cambridge University Press (1970)†.
  • I. Lakatos, The Methodology of Scientific Research Programmes: Philosophical Papers, vol. 1, ed. J. Worrall and G. Currie, Cambridge University Press (1978)†.
  • K. R. Popper, The Logic of Scientific Discovery, Hutchinson (1959)†.
  • E. Zahar, “Why did Einstein’s programme supersede Lorentz’s?”, British Journal for the Philosophy of Science 24 (1973)†.

Sources#

  • papers/broken-born-checkpoint-v2.md §15 — the current assessment in Lakatos’s terms; §14 items 1–3 — what the word “theory” waits on; §7.4 — the novelty downgrade
  • review/round3-adjudication.md ‘Decisions’ — Decision 2 (the free dilution factor, misattribution) and Decision 4
  • notes/debt1a-iii-tuning-test.md §5 — forced convergence as “the opposite of degeneration”
  • notes/debt3b-wave-island.md ‘What pins g?’ and notes/debt3c-ufd-endpoint.md ‘Verdict and closure of Debt 3’ — the degeneration flag on the wave island
  • notes/broken-born-program-v1.md §11 — the v1 assessment (superseded by #11 §15)
  • HANDOFF.md §1 and root CLAUDE.md (preamble) — “a Lakatosian research program, not a theory”
Backward Method The collaboration: roles, attribution and conventions Forward