Possibility portfolio · current commitments

Put scarce effort where it can change capability.

The Active Frontier now reflects the work actually consuming scarce attention: one flagship empirical programme, one owned capability practice, and one independent related testbed. Earlier conceptual directions remain preserved without masquerading as current projects.

Operating rule

Preserve possibilities cheaply. Commit where the next action can change what you are able to do.

The portfolio remains an operating aid rather than a validated measure of value. Its role is to stop every interesting branch from silently becoming another obligation.

Work-in-progress limit

At most three substantial bets.

A slot is a ceiling, not a quota. Two good projects are better than three invented obligations.

The default Active Frontier limit is 3. A new candidate enters when a slot is free, an incumbent leaves, capacity demonstrably increases, or a time-sensitive exception is made explicitly.

Finding prior art, reusing someone else's tool, handing work off or deciding not to pursue a branch can all be successful outcomes.

Meta-work budget

The scaffolding gets one slot, not the laboratory.

At most one of the three active bets should primarily concern the machinery, governance, taxonomy or provenance of Agalmic Research itself.

The other slots should terminate in object-level work: a measured capability change, a working technology under test, an empirical result, a formal result, a reusable intervention, or another outcome constrained by something outside the programme's own vocabulary.

This is an operating ceiling rather than a metaphysical distinction. If infrastructure becomes the binding constraint on real work, it can earn a slot. It does not earn one merely because the framework can be made more elegant.

Research machinery is justified by the research it makes cheaper, clearer or more falsifiable.

Promotion gate

Every active bet needs a way to lose.

Before a direction consumes an Active Frontier slot, it should identify the external constraint that can change the decision, the minimum useful test, and the condition under which the claim, mechanism or special terminology would be narrowed, parked or retired.

“Produce a better explanation” is not enough. A discriminating action should alter belief, capability, architecture, resource allocation or continuation.

See results, corrections and retired claims →

Active frontier

Where substantial effort is going now.

Each active item must have a concrete next action tied to evidence or capability. Ownership boundaries are part of the state: Nemosyne appears here as an independent testbed, not an Agalmic-owned roadmap.

A01

Attention Allocation Under Cognitive Abundance

Empirically test how scarce evaluation should be allocated when candidate generation is cheap, while preserving temporal integrity, selection-aware observability, lineage controls and the right to abstain before outcomes are opened.

Intended capability outcome: A defensible empirical account of which allocation policies preserve valuable and unconventional candidates under scarce attention, or a defensible demonstration that the available evidence cannot answer the question.

Capacity created: allocation evidence · temporal provenance · selection-aware evaluation · pre-outcome research firewalls

Next discriminating action: Recover a selection-neutral additional historical cohort, preferably ICLR 2021, then rerun the v0.2 source and prospective-precision gates before any further bulk text or citation acquisition.

Lineage: active learning, selective labels, constrained ranking, peer-review allocation and exploration/exploitation antecedents + outcome-blind empirical synthesis

active
A02

Personal capability practice and Epistemic Gym

Use abundant machine cognition and inherited human knowledge to identify and relax real constraints on scholarliness, judgment and durable personal capability, with delayed transfer rather than assisted output as the key test.

Intended capability outcome: A practitioner who becomes progressively more capable of understanding, judging, creating and defending work even when AI scaffolding is reduced.

Capacity created: scholarliness · calibration · source judgment · transfer · authority-boundary recognition

Next discriminating action: Run the Personal Epistemic Gym on live research and professional artefacts, record assistance level and confidence before critique, then test delayed unaided transfer on unfamiliar material.

Lineage: learning science, metacognition, intelligent tutoring and human-capital antecedents + local synthesis/application

active
A03

Nemosyne as an independent discovery and reconstruction testbed

Use the independent Nemosyne project as a bounded testbed for representation, verification attention and discovery/reconstruction methods without treating its product roadmap as owned by Agalmic Research.

Intended capability outcome: A real external testbed where bounded Agalmic methods can be challenged by difficult engineering and representation problems without blurring project ownership.

Capacity created: discovery breadth · representation insight · systems understanding · epistemic reconstruction · responsible ownership

Next discriminating action: Run only bounded Agalmic experiments whose intervention and outcome are observable; keep Nemosyne product implementation and operational ownership outside the Agalmic Research project map.

Lineage: prototype/learning, exploratory programming, software comprehension and human-AI coding antecedents; local operating method

active

Frontier filter

Capability before novelty.

A candidate is considered across separate dimensions rather than collapsed into one score: current constraint, capability leverage, information gain, realization or handoff path, attention burden, inheritance already available, urgency and uncertainty.

The promotion question is: if scarce effort is spent here, what becomes meaningfully easier to understand, judge, create, lead or operate?

High uncertainty is often a reason to buy information cheaply. High polish is not evidence of learning. Novelty is not required if reuse or implementation produces the needed capability.

Preserved options

Worth keeping. Not active obligations.

Earlier theory-building directions remain visible and reversible without consuming an Active Frontier slot.

O01

Software engineering and engineering leadership capability growth

Treat software engineering and engineering leadership as live domains inside the Personal Capability Practice rather than as a separate frontier project.

Next discriminating action: Instrument engineering and leadership tasks inside the Personal Capability Practice and promote a separate project only if a distinct intervention or research question emerges.

Lineage: deliberate practice, expertise, software engineering education and leadership-development antecedents; application synthesis

option
O02

Cognitive tools for expanding the effective adjacent possible

Develop the programme's earlier meta aim as a lineage-grounded synthesis connecting cognitive tools, search systems, recognition, validation, assimilation and realization.

Next discriminating action: Keep as a conceptual option; return only if the personal capability practice exposes a concrete mechanism that benefits from formal frontier modelling.

Lineage: established adjacent-possible/effective-adjacent-possible antecedents + synthesis/extension

option
O03

Distributed discovery provenance

Operationalize provenance for human-AI discovery episodes by separating framing, generation, recognition, validation, integration, realization, priority and epistemic authority using standards-first provenance.

Next discriminating action: Use provenance where it supports live capability work; avoid expanding the provenance theory unless concrete gaps appear in practice.

Lineage: established computational-discovery/philosophy antecedents + operationalization opportunity

option
O04

Economics of epistemic abundance: formalize shifting constraints

Preserve the question of when abundance shocks change binding constraints, shadow values, rents or market power, while recognizing that current theory/prediction novelty has not survived the formal null test.

Next discriminating action: Use economics primarily as a personal capability-development domain; revisit distinct research claims only if empirical practice exposes a real gap.

Lineage: mature antecedents; no distinct explanatory or predictive Agalmic primitive established

option
O05

Epistemic handoff pilot

Test whether claim-sensitive, provenance-preserving routing to missing expertise improves evaluation of low-authority contributions without erasing originators or overwhelming expert attention.

Next discriminating action: Exercise handoff on real authority boundaries encountered in the personal capability practice before proposing platform machinery.

Lineage: knowledge brokerage/absorptive-capacity antecedents + narrow operational extension

option
O06

Innovation is search, but search is not enough

A research note connecting established innovation-as-search work to the machine-cognition regime, where larger search breadth can worsen the selection, testing and realization bottlenecks.

Next discriminating action: Publish only if the note reduces future search or clarifies a concrete practice that is not already captured elsewhere.

Lineage: established search literature + explanatory synthesis

option
O07

What should we make abundant next?

Examine which upstream forms of abundance, such as knowledge, algorithms, energy, compute, standards or access, most strongly expand feasible capability sets.

Next discriminating action: Apply the question inwardly first: identify which abundant resource most effectively attacks a demonstrated personal bottleneck.

Lineage: novelty unassessed; requires growth/capability/technology lineage review

option
O08

Allocation of scarce expert attention

Study mechanisms for allocating scarce specialist evaluation when candidate production outgrows expert review capacity, while avoiding credential capture, pay-to-be-heard dynamics and filter monopoly.

Next discriminating action: Keep as an option until the inward practice produces repeated real expert-attention bottlenecks.

Lineage: existing expert-evaluation/attention/intermediary antecedents; narrower delta unassessed

option
O09

Representation as discovery through Nemosyne/Moneta

Use Nemosyne/Moneta as an applied testbed for whether alternative representations can increase what analysts can notice and act upon while controlling cognitive and computational cost.

Next discriminating action: Keep the scientific claim as an option while the active Nemosyne work concentrates on discovery followed by epistemic reconstruction.

Lineage: broad representation/visual analytics antecedents; concrete technical novelty unassessed

option
O10

Epistemic artefacts, assimilation and frontier formation

Distinguish outputs of directed human-machine search from assimilated knowledge by modeling epistemic artefacts, warrant, agent-relative understanding, assimilation, capability and frontier effects separately.

Next discriminating action: Use the distinction operationally in the Personal Epistemic Gym and Nemosyne before expanding it into a larger theory.

Lineage: established epistemic-artifact and epistemic-opacity antecedents + synthesis/application; novelty unassessed

option
O11

Constitutional attention for continuous scarcity-displacement search

Explore a governance/search architecture in which scarce human intervention shapes objectives, constraints, escalation and constitutional updates while autonomous systems search for strategies that displace scarcity.

Next discriminating action: Keep as an option until the personal capability practice creates a concrete need for an automated attention-allocation mechanism.

Lineage: interactive evolutionary computation + selective human oversight + autonomous discovery antecedents; terminology/synthesis novelty unassessed

option
Parked

Preserved, but not presently competitive for attention.

Parking retains provenance and reversal conditions without pretending every interesting branch deserves current effort.

P01

Human-Scale Epistemic Horizon

Explore institutional consequences if reliable frontier knowledge increasingly exceeds what any individual human can personally survey or defend.

Revisit: Keep parked until nearer-term capability and verification work yields concrete evidence or mechanisms.

parked

Public boundary

This is only the public-safe portfolio.

Confidential, personally sensitive, security-sensitive or potential-IP-sensitive ideas can use the same operating discipline outside the public projection.

The public JSON projection is versioned at src/data/possibilityPortfolio.json ↗.

See project ownership and current state →