Discovery Attribution in Human–AI Science ↗
Computational discovery, collective-centred discovery, priority, validation and machine novelty.
Intellectual lineage & novelty · living register
Do not reinvent the wheel and call the spokes a revolution.
Agalmic Research treats prior art as infrastructure. The aim is to discover what is already known, reuse it aggressively, and identify the narrower remainder that actually deserves investigation.
Novelty discipline
Novelty is a conclusion of search, not a tone of voice.
An idea can be new to the curator, new to a model and even independently rediscovered without being new to the world. “I had not seen this before” is provenance. It is not prior-art clearance.
Correction
The project changes its claims when prior art changes the picture.
Several ideas initially developed here have close or foundational antecedents.
“Epistemic Economics” is now Economics of Epistemic Abundance. The “Law of Displaced Scarcity” is the Displaced Scarcity Hypothesis. Working Paper 001 is now v0.4 and explicitly grounded in prior work on metacognition, verification gaps, claim accountability, computational/collective discovery, contributorship, epistemic dependence and computational opacity.
The new frontier-management review adds another useful correction: effective adjacent possible already appears explicitly in Taalbi’s innovation research, while idea capture, selection, portfolios and option-like staged commitment all have mature literatures.
Agalmic abundance applies to ideas too: finding that someone already solved part of the problem is a gain, not a loss.
Deep reviews
Four branches now have dedicated second-pass reviews.
These notes record collisions, surviving deltas and search boundaries rather than hiding prior art behind a bibliography.
deep review
Epistemic Handoff, Knowledge Brokering and Uptake ↗
Absorptive capacity, transfer stickiness, brokers, intermediaries, transactive memory and expert-attention bottlenecks.
Displaced Scarcity and Shifting Bottlenecks ↗
Limiting factors, complements, attention, value capture, idea processing, induced innovation and current AI-abundance economics.
Idea Portfolios, Selection and the Active Frontier ↗
Exploration/exploitation, innovation portfolios, real options, idea management, selection bias and the effective adjacent possible.
Living register
What we currently think we are adding.
This is intentionally conservative and corrigible. New prior art should downgrade a claim rather than be squeezed into a footnote.
Agalmics / Agalmic Economics
Antecedents: Robert Levin’s late-1990s “Agalmics: The Marginalization of Scarcity” already develops agalmics around non-scarce goods and their relation to scarce inputs.
Current position: Agalmic Research uses the existing term for a contemporary programme centred on machine cognition, knowledge production and abundance-era institutions. It does not claim to originate agalmics.
Displaced Scarcity Hypothesis
Antecedents: Simon’s attention scarcity, fixed-proportions bottlenecks, Teece’s complementary assets, Weitzman’s idea-processing constraint, strong-complementarity models and recent AI-abundance papers all anticipate substantial parts of the mechanism.
Current position: “Law” is withdrawn. The useful research task is formal: identify when an abundance shock changes binding constraints, shadow values, rents or market power, and distinguish complementarity, substitutability and induced-demand regimes.
Economics of Epistemic Abundance
Antecedents: Anna Grandori already published “Epistemic Economics and Organization”; Arrow and later knowledge-economics traditions address invention and information.
Current position: The programme uses “Economics of Epistemic Abundance” for its narrower question: allocation when candidate knowledge becomes cheap while verification, judgment, attention, trust and realization remain scarce.
Adjacent possible / effective adjacent possible
Antecedents: Kauffman introduced the adjacent possible, and recent innovation work by Taalbi explicitly distinguishes an effective adjacent possible constrained by search complexity, absorptive/search capacity and resources.
Current position: Agalmic Research does not claim the theoretical/effective distinction. The meta-level question is how cognitive tools, representation, validation, provenance and realization infrastructure change the effective frontier available to a particular human-machine system.
Björneborn 2020 · Adjacent Possible ↗ · Taalbi 2026 · effective adjacent possible ↗
Innovation Search Under AI
Antecedents: Search is central to evolutionary economics; recombinant growth and adjacent-possible frameworks are established. Recent evidence also suggests AI can alter the direction of scientific search, not merely its speed.
Current position: The prospective contribution is the effect of machine cognition on search cost, branching, recombination, search direction, selection pressure and realization.
Idea preservation and Active Frontier
Antecedents: March’s exploration/exploitation framework, innovation portfolio management, real-options approaches to R&D, mature idea-management research and creativity-selection work already cover the main mechanisms of capturing, staging, selecting and abandoning ideas.
Current position: Possibility Portfolio, Possibility Garden and Active Frontier are local operating metaphors. The system’s practical purpose is to preserve a large option set while limiting scarce active commitments and evaluating projects by the capacity they create, not to claim a new theory of idea management.
March 1991 ↗ · Cooper et al. 2001 ↗ · Lint & Pennings 2001 ↗ · Baraboshkin et al. 2026 ↗
Authorial Capacity Constraint
Antecedents: Fernandes et al., Maier, O’Keefe and van Zoonen et al. directly address gaps between AI-assisted artefact performance, human understanding, metacognitive calibration and claim defensibility.
Current position: The label is retained as an organizing construct while explicitly withdrawing any implication that Agalmic Research first discovered the underlying phenomenon.
Fernandes et al. 2026 ↗ · Maier 2026 ↗ · O’Keefe 2026 ↗ · van Zoonen et al. 2026 ↗
Defensible stewardship
Antecedents: Claim accountability already argues that a named human should be able to reconstruct and defend scholarly claims. BMJ contributorship and guarantorship predate the current AI debate.
Current position: “Epistemic authority should not exceed defensibility” is an Agalmic Research operating maxim, not a claim of first discovery. The research problem is operationalizing authority evidence without creating theatre.
Contribution roles and epistemic authority
Antecedents: CRediT standardizes fourteen contributor roles, and contributorship models have separated “who did what” from the byline for decades.
Current position: Use CRediT instead of inventing another contribution taxonomy. The experimental extension is a separate, claim-sensitive record of who or what can warrant which claims and why.
Epistemic handoff and uptake
Antecedents: Absorptive capacity, transfer stickiness, knowledge brokers, innovation intermediaries, boundary spanning, transactive memory and expert-jury bottlenecks all cover important parts of routing and uptake.
Current position: “Epistemic handoff” is retained only for a narrow, claim-sensitive transfer: preserve origin and contribution, state the authority gap, route toward a specified missing epistemic function, and record what warrant the recipient actually adds.
Cohen & Levinthal 1990 ↗ · Howells 2006 ↗ · Meyer 2010 ↗ · Gimpel et al. 2025 ↗
Distributed discovery provenance
Antecedents: Computational discovery predates LLMs. Langley studied human–computer discovery cooperation; Clark & Khosrowi explicitly argue for a collective-centred account of AI-assisted discovery and discuss human recognition of significance. Priority scholarship separately studies disclosure, validation and credit.
Current position: The project does not present “does AI discover or facilitate discovery?” as a new question. The prospective contribution is operational: represent novelty, priority, contribution and epistemic authority separately across human–AI discovery episodes.
Langley 2000 ↗ · Clark & Khosrowi 2022 ↗ · Vale & Hyman 2016 ↗ · AlphaDev 2023 ↗
Human-Scale Epistemic Horizon
Antecedents: Computational opacity and computer-assisted mathematics already examine results that may be reliable without being surveyable or intuitively possessed by a human.
Current position: The useful question is the institutional and economic consequence of increasing opacity, not a claim that Agalmic Research discovered epistemic opacity.
Memory Palace research graph
Antecedents: W3C PROV, RO-Crate, CRediT, MIRA and decision-provenance work already cover provenance, research-object packaging, contribution roles, claim/evidence graphs and contemporaneous decision records.
Current position: The Memory Palace is an application profile over those systems. Local extensions are reserved for authority state, search boundaries, branch state, reversal conditions and explicit epistemic handoff.
Frontier expansion
The meta aim inherits an existing vocabulary.
The goal is not more ideas. It is greater capacity to reach worthwhile futures.
This is a programme-level objective, not a claim that Agalmic Research discovered the effective-adjacent-possible concept. The useful question is what machine cognition and complementary human/institutional capacities change about the frontier that can actually be searched, recognized, validated and realized.
Distributed discovery
Generation, discovery credit and novelty are separate questions.
The literature already rejects a simple one-actor picture of scientific discovery.
Machine discovery research predates generative AI, and Clark & Khosrowi explicitly develop a collective-centred account of AI-assisted discovery. Our useful remaining task is operational rather than titular.
Framing / directionDefine the problem and constrain the search.
GenerationProduce or surface a candidate result.
RecognitionSelect or notice that a candidate is worth pursuing.
ValidationSupply proof, evidence, replication or expert scrutiny.
IntegrationLocate the result in the existing body of knowledge.
RealizationTurn the warranted result into a usable object.
These are proposed provenance fields for study, not a claim to have discovered the constituent acts.
Novelty is a relation between a result and prior knowledge. Discovery attribution is a relation between a discovery episode and its contributors.
Standards-first provenance
Reuse the machinery that already exists.
Agalmic Research is migrating ordinary provenance to W3C PROV, research-object packaging to RO-Crate, and scholarly contribution roles to CRediT. The local graph is an extension layer rather than a rival standard.
Search boundary
This register does not certify novelty.
The review is deepening but remains non-exhaustive. Scopus, Web of Science, EconLit, PhilPapers, ProQuest, patent databases, archival books and non-English literatures have not all been systematically exhausted. Technical Nemosyne/Moneta mechanisms remain outside scope.
“Not found” means novelty remains unassessed, not that the project owns an intellectual island no one has visited.