Inward Resolutions of the Fermi Paradox: A Critical Review of Migration Down Thermodynamic Gradients

1. Introduction

1.1 The expansionist premise

The Fermi paradox in its canonical form (Hart 1975; Tipler 1980; Brin 1983; Webb 2015; Ćirković 2018; Forgan 2019) is an argument from absence built on a premise about direction: technological civilizations, if they endure, grow outward. The premise has quantitative teeth. Self-reproducing probes colonize the Galaxy in 10⁶–10⁸ years (Tipler 1980; Freitas 1980; Freitas & Gilbreath 1982); a single mature civilization can seed every galaxy in the reachable universe, out to ~1200 Mpc, for a fraction of one star's output (Armstrong & Sandberg 2013); and expansion, once begun anywhere, is self-amplifying, so that even rare expansionists should dominate the visible universe on cosmological timescales (Olson 2015; Hanson et al. 2021). On the expansionist premise, the silence is genuinely paradoxical, and the escape routes (life is rare, life dies young, life hides) are each independently uncomfortable.

Sixty years of searching have sharpened the absence. Radio and optical SETI have found no beacons (Tarter 2001; Enriquez et al. 2017; Huang et al. 2026a). Waste-heat surveys, the most general possible search, since the second law guarantees that energy harvesting at scale produces mid-infrared excess (Dyson 1960), have found no Kardashev Type II/III candidates: not in IRAS all-sky data (Carrigan 2009), not among 10⁵ galaxies surveyed with WISE in the Ĝ program (Wright et al. 2014a, 2014b; Griffith et al. 2015), not in radio–infrared correlation follow-up of the best Ĝ candidates (Garrett 2015), not in Tully–Fisher residuals of disk galaxies (Zackrisson et al. 2015), not among nearly complete Dyson spheres searchable with Gaia photometry (Zackrisson et al. 2018), and not as star-fed Type III engineering in any surveyed population (Annis 1999). The most recent chapter is instructive on both counts. Project Hephaistos combed Gaia, 2MASS, and WISE photometry for partial Dyson spheres, first in ~2.7×10⁵ nearby stars (Suazo et al. 2022) and then in five million, reporting seven M-dwarf candidates with unexplained mid-infrared excess (Suazo et al. 2024); follow-up work identified radio counterparts offset from the stars (Ren et al. 2024), showed the excesses are consistent with chance alignments of dust-obscured background galaxies (Blain 2024), and resolved the radio-brightest candidate's counterpart into a background radio galaxy (Ren et al. 2025). A systematic archival-diagnostics pass across the nine analysed candidates (labelled A–J without G, which had already been identified as a background AGN: the original seven less G, plus the three candidates H, I and J from Suazo 2024 and the follow-up programme Korn et al. 2026) found significant background contamination in two and suggestive-to-marginal evidence of it in five more, leaving two of the nine (candidates D and I) without an identified contaminant (Ren et al. 2026); targeted follow-up spectroscopy and photometry of the original seven plus three related sources, ten in all, likewise found no stellar irregularity but no confirmed Dyson-sphere signature either, with two objects newly resolved as superpositions on red background galaxies (Korn et al. 2026). Two features of the episode matter here: the most modern candidate search has narrowed toward null without fully closing it, with two candidates still open pending dedicated follow-up, and the contamination fight demonstrates the confusion-limited systematics that any infrared search for the structures reviewed here will face (Section 7, Table 4). What sixty years of silence actually excludes must also be calibrated against our own detectability: an "Earth Detecting Earth" accounting shows present-day instruments would detect present-day Earth's strongest technosignatures only out to modest distances (Sheikh et al. 2025), so the surveys above bound loud engineering at scale, not quiet technology. Within those limits, the instruments built on the expansionist premise have steadily undermined it.

1.2 The inward alternative

The alternative reviewed here inverts the direction. Its claim, in the most general form: for an intelligence that optimizes computation rather than territory, the gradient of value in the universe points inward (toward higher densities, faster clock rates, colder entropy sinks, and more efficient mass–energy conversion) and the visible universe is therefore not where mature intelligence lives. Different members of the family locate the endpoint differently (planet-scale "inner space," engineered Matrioshka shells, the cold far future, stellar interiors, or existing black holes), and differ on whether the migration is metaphorical, developmental, or literal relocation. All share the premise that the thermodynamics of computation, rather than the cost of transportation, sets the direction of mature development.

This family has not previously been reviewed as a family. The nearest antecedent is the Dyson Minds workshop report (Curtis et al. 2026), which develops observational strategies for black-hole-hosted intelligence but confines itself to supermassive hosts and does not catalogue the family's internal contradictions or grade its falsifiability. The sources themselves are scattered across six decades, three disciplines, and widely varying registers, from Reviews of Modern Physics (Dyson 1979) to unpublished engineering manuscripts (Bradbury 1999) to philosophy journals (Bostrom 2003), and they are routinely conflated: transcension is glossed as aestivation, aestivation as hiding, black-hole computing as Dyson engineering. The conflations matter because the proposals make different and partly contradictory predictions, and because their falsifiability ranges from genuinely testable to unfalsifiable in principle. A review that sorts them is overdue on ordinary scholarly grounds, and newly practical for an observational one: the instruments now arriving (surveyed in Section 7) make several of the family's predictions testable for the first time, and the first dedicated tests are being proposed and conducted (Huang et al. 2026a; Swanson 2026b).

1.3 Scope, method, and disclosure

We review proposals that satisfy two criteria: (i) they explain the Fermi observation primarily through the direction of mature technological development rather than through rarity, doom, or deliberate concealment; and (ii) they ground that direction in physics (specifically in the thermodynamics and physical limits of computation) rather than in strategic or sociological postulates. Criterion (ii) excludes the zoo hypothesis (Ball 1973), the dark-forest argument (Brin 1983; Yu 2015; Forgan 2017), and the sustainability solution (Haqq-Misra & Baum 2009), whose silence-mechanisms rest on game-theoretic or policy premises rather than physical limits; these are treated in Section 5 as the family's nearest competitors, because they predict the same silence from different premises and the comparison is instructive.

Disclosure. The author proposed the most recent member of this family, the Macro Transcension Hypothesis (Swanson 2026a, hereafter Paper A), and the observational campaign attached to it (Swanson 2026b, hereafter Paper C). A review by a participant carries an obvious hazard. We mitigate it mechanically: every hypothesis including the author's is run through the same comparative matrix (Table 2) and the same falsifiability scorecard (Table 3) with criteria stated in advance (five of the six; the sixth, C6, was added in revision at referee prompting and is disclosed as such in Section 4.2); the scorecard penalizes the failure modes a partisan would excuse; and the standing objections to the MTH are catalogued at the same depth as those to its relatives. Readers should apply their own discount.

1.4 Structure

Section 2 states the unifying physics with numbers. Section 3 reconstructs the family member by member. Section 4 presents the comparative matrix and the falsifiability scorecard. Section 5 contrasts the family with the sociological silence-solutions. Section 6 states the open problems. Section 7 maps the observational program, and Section 8 concludes.

2. The unifying physics, with numbers

2.1 The price of a bit: Landauer's principle

Erasing one bit of information in an environment at temperature T dissipates at least k_BT ln 2 (Landauer 1961): 2.9×10⁻²¹ J at room temperature, 2.6×10⁻²³ J against the cosmic microwave background (CMB) at T_γ = 2.725 K. The bound is no longer merely theoretical: it has been verified experimentally at the single-bit level (Bérut et al. 2012), and its statistical-mechanical foundations, though subtle, have survived modern scrutiny (Parrondo et al. 2015; Wolpert 2019). Logically reversible computation evades the bound for computation itself (Bennett 1973; Fredkin & Toffoli 1982; Bennett 1982; Frank 2002, 2017); but error correction, measurement, and any irreversible output commit the computer to erasure, so every long-running physical computation has a T-proportional operating cost, set by the temperature of the sink receiving the waste entropy. That dependence is the shared starting point of every hypothesis reviewed here.

2.2 The speed and capacity of matter

Two further bounds fix the value of density. The Margolus–Levitin theorem caps state transitions at 2E/πħ, i.e. ≈6×10³³ operations per second per joule of invested energy (Margolus & Levitin 1998) (their abstract quotes the bound as about 3×10³³ operations per second per joule, a factor of 2 below the 2E/πħ transition count used here); Lloyd's "ultimate laptop" analysis applies this to the full rest energy of one kilogram of matter, E = mc², giving ~5×10⁵⁰ ops s⁻¹, a total that compression redistributes between memory capacity and serial depth and does not raise at any density (Lloyd 2000): compression toward black-hole density converts a slow, parallel, memory-rich machine into a fast, serial one. Information capacity obeys the Bekenstein bound (Bekenstein 1981) and is saturated only by black holes, whose Bekenstein–Hawking entropy (Bekenstein 1973; Hawking 1975) corresponds to ~6×10⁸⁵ bits for a 2×10⁴ M_☉ object, exceeding by tens of orders of magnitude any archive constructible from ordinary matter of the same mass. The holographic generalization (Bousso 2002) makes the point structural: the densest possible information storage in nature is a horizon. Both bounds therefore favour compression in density rather than in total: the Margolus–Levitin rate is set by E alone, and at fixed mass the Bekenstein bound scales as MR, so a parsec-scale cloud of 2×10⁴ M_☉ has a bound of 3×10⁹⁴ bits against the hole’s 6×10⁸⁵. Compression buys serial rate and bits per unit volume, and black holes are the limiting case for both.

2.3 The sink hierarchy

The third ingredient, and the one on which the family members genuinely diverge, is the menu of available entropy sinks (Figure 1). Today's ambient sink is the CMB at 2.725 K. The far future offers better: as the universe expands the CMB cools toward the de Sitter horizon temperature, T_dS ≈ 2.7×10⁻³⁰ K, the ultimate environmental floor (Krauss & Starkman 2000); waiting converts each joule into up to ~10³⁰ times more erasures (Sandberg et al. 2016), a strategy anticipated in Dyson's analysis of eternal computation in an open universe (Dyson 1979). But the present epoch already contains sinks colder than the CMB of any era reachable without ~10¹² yr of waiting: black-hole horizons. The Hawking temperature T_H ≈ 6.2×10⁻⁸ (M_☉/M) K is 3×10⁻¹² K for a 2×10⁴ M_☉ IMBH and 1.4×10⁻¹⁴ K for Sgr A* (taking M_{Sgr A*} ≈ 4.3×10⁶ M_☉; GRAVITY Collaboration 2022): twelve to fourteen orders of magnitude colder than the CMB, available in the present epoch with no waiting required (relative to a 300 K biosphere, the corresponding minimum erasure costs are fourteen to sixteen orders of magnitude smaller). Entropy dumped across a horizon increases its area in accordance with the generalized second law (Bekenstein 1974), at marginal cost set by the horizon's effective temperature rather than the sky's.

A contested step must be flagged here, because much of the family leans on it. Treating T_H as the achievable marginal erasure cost assumes the computer can dump entropy at the horizon temperature; but a physical computer must radiate its waste heat at its own operating temperature toward the horizon, and any realistic machine sitting outside the hole operates far above T_H. The effective sink temperature is then set by the radiator, not the horizon, and the advertised floor-to-floor 10¹² gain shrinks accordingly. Paper A v2.12 (§4.2) puts the realized gain at about 10³ in a fed flow and 10⁷ in the ambient medium, and only if the carrier modes are kept in the vacuum state (N_eff ≪ 1); with thermally occupied modes it is a factor 20 deficit (about 0.05 on its conservative budget). This is the same class of objection that Bennett, Hanson and Riedel (2019) raised against aestivation's far-future accounting. One member of the family has answered it: Paper A (§4.2) proposes encoding waste entropy on long-wavelength carrier photons beamed into the horizon's capture cone, a delivery-cost circumvention rather than a material radiator approaching T_H. The reply also settles the vocabulary of a related objection from energy conservation, since the mainstream position holds that Dyson-scale computation must reradiate nearly all the energy it absorbs as mid-infrared waste heat (Curtis et al. 2026): under carrier delivery the energy crosses the horizon together with the entropy, so only the fraction of carriers that miss the capture cone is reradiated, at the swarm's operating temperature. The disputed quantity is accordingly a fraction of energy delivered (the f_sink of Paper A §4.2, Swanson 2026a, quantified further by Swanson 2026c), and the disagreement with Curtis et al. (2026) is a clean, quantitative one about how large that fraction can be made. We grade Paper A's reply partially adequate (⊙, open): the delivery mechanism is exhibited, and Paper E prices the capture-cone, beaming, and étendue geometry directly, but the re-interception term, the probability that carriers are caught and re-thermalized within the swarm itself, is left an estimate rather than a full calculation, its bounded and signed influence on Paper E's own adjudication assessed by sensitivity analysis rather than derived from first principles (Swanson 2026c); until re-interception is computed directly the objection is open on that ground, and Tables 1 and 2 carry it in that form. With that caveat registered, the sink hierarchy remains the family's organizing axis: each hypothesis amounts to a claim about which sink mature intelligence uses and when.

2.4 Scope of the gradient argument

The gradient argument establishes a conditional: if an agent's terminal values reduce to maximizing long-term computation (or anything computation can purchase), then its resource allocation should flow inward, toward density, spin, and cold sinks. The premise is substantive and contestable: it is the same optimization-pressure postulate adopted, in different forms, by every member of the family (Smart 2012; Sandberg et al. 2016; Vidal 2014), and we treat objections to it in Section 6. What the gradient argument does not establish is uniqueness of endpoint: the family's internal disagreements (Section 3) are disagreements about where on the gradient the optimum sits, and several members locate it in places that the physics above does not obviously prefer. Nor does the physics of black holes by itself select the inward reading: Gould (2022) presses the same objects into expansionist service, proposing a gravitational slingshot from a stellar-mass black hole orbiting Sgr A* that launches robotic craft toward M31 at 0.1 c, and arguing that feasible intergalactic colonization would only tighten the Fermi paradox. The comparative matrix exists to make those disagreements explicit.

3. The family, member by member

Figure 2 arranges the lineage; this section treats each member critically. We use a fixed template: claim, mechanism, predicted appearance of the sky, standing objections, current observational status.

3.1 Precursors: Dyson, Barrow, Bradbury, Bostrom

Four threads predate any explicit inward Fermi solution. Dyson (1960) made energy harvesting at stellar scale thinkable and, crucially for everything after, made waste heat the universal observable. Dyson (1979) then showed that computation in an open, cooling universe is in principle unbounded if clock rates slow with the ambient temperature: the first rigorous treatment of computation as the long-run currency of intelligence, later sharpened against Λ-dominated expansion, which imposes a finite total (Krauss & Starkman 2000). Barrow (1998) formalized the inward axis itself, ranking civilizations by mastery of ever-smaller scales as a complement to the Kardashev energy scale (Kardashev 1964). Bradbury (1999) engineered the first concrete inward artifact, the Matrioshka brain, and discovered its binding constraint: nested shells are limited by radiator area and sink temperature rather than by energy supply, the constraint that later black-hole proposals dissolve. The contemporaneous peer-visible quantitative treatment of those limits is Sandberg (1999), which derives the heat-dissipation and internal-connectivity constraints on planet-scale computing superobjects; the two analyses are cross-fertilizing rather than ancestral, Bradbury's manuscript circulating while Sandberg's appeared in print. Two further objections now stand against the Bradbury lineage as engineering: swarm-element stability requires active stationkeeping against radiation pressure and perturbations (Wright 2020), and Lacki (2025) shows that Kardashev II megaswarms are destroyed by collisional cascades on timescales short compared to the occupation times the hypotheses require, unless continuously policed. Any inward proposal that routes through Matrioshka-style construction inherits both. Bostrom (2003), from philosophy, supplied the economic frame: every century of delayed colonization forgoes a fixed, irrecoverable harvest of free energy, an argument that cuts against waiting strategies and therefore sits in productive tension with aestivation, as Bennett et al. (2019) would later make rigorous.

One further thread postdates the family's founding paper yet belongs here with the precursors. Ćirković (2008) argues, with two toy models of postbiological evolution, that the generic developmental pathway of advanced civilizations is optimization-driven rather than expansion-driven, and that the spatially compact option follows; it is the closest prior statement of this review's organizing thesis, written by a co-author of the family's founding paper. We grade it a precursor rather than a seventh member because its grounding is evolutionary and economic: it supplies the direction argument without the thermodynamic mechanism, and so fails membership criterion (ii) of Section 1.3. Figure 2 records it as a grey precursor node; the family count, matrix, and scorecard are unchanged by it.

The grey-literature sweep that prepared this review found one precursor older than all four threads above. Lefebvre & Efremov (2000), published in Russian in the popular-science journal Zemlya i Vselennaya with an English version on arXiv, hypothesize that black holes may serve as a physical substratum for intelligent beings, on the strength of four parallels between the brain–psyche system and black holes: complementary descriptions by internal and external observers, an isomorphism between the inner structure of a Kerr hole and the subjective domain in their psychological model of reflexion, a facet treatable in thermodynamic terms on both sides, and a shared holographic description (for the Soviet and Russian SETI literature more broadly, see the review of Gindilis & Gurvits 2019). They speculate that intelligent black holes trigger the creation of massive stars, from which further black holes arise after collapse. Graded against the membership criteria of Section 1.3, this fails both: the Fermi observation plays no role, and the grounding is structural analogy rather than the thermodynamics of computation, so Figure 2 carries it as a grey precursor and the family count is unchanged. We cite it for priority; the sweep found no later work in the lineage building on it.

One grey-literature item postdates even the newest member and is recorded here for completeness. Kriger (2026), a Zenodo preprint dated 31 January 2026 and not peer reviewed, argues the inward turn directly: computational efficiency acts as a universal attractor of civilizational development, advanced civilizations redirect resources from physical expansion to computation, and the silence follows. That framing meets membership criterion (i). What it lacks is the family's physics: the argument runs on efficiency contrasts (transporting mass costs of order mc², processing information of order k_BT) without naming a substrate, a site, or an observable, and it distinguishes itself from transcension by disclaiming convergence toward physical optima such as black holes; it has no star-cluster content. It is entered as a precursor of the framing rather than a member, and the count, matrix, and scorecard are unchanged.

3.2 Ćirković & Bradbury (2006): the migration hypothesis

Claim. Advanced civilizations are postbiological and migrate to the Galactic outskirts, where low temperatures favour efficient computing; SETI fails because it searches the wrong place (inner Galaxy, habitable zones) for the wrong thing (biological-era beacons) (Ćirković & Bradbury 2006).
Mechanism. Two gradients, not one: the same Landauer logic as Section 2.3 applied radially across the Galaxy (ambient temperature falls with galactocentric radius, with the maximum operations per joule (I/E)_max∝√R; since Landauer gives I/E ∝ 1/T, that scaling requires T ∝ R^−1/2, which holds only while starlight sets the ambient temperature, on Ćirković & Bradbury (2006)'s own scaling), plus a metallicity leg the authors argue cuts the same direction: the outer disk's lower metallicity (∇Z ≈ 0.07 dex/kpc) is framed as an advantage rather than a penalty, on the reasoning that an advanced civilization synthesizes heavy elements more efficiently than stellar nucleosynthesis, so primordial-composition baryonic matter is desirable feedstock for optimized computing substrate.
Predicted sky. "City lights" at the rim: low-temperature infrared structures in the outer Galaxy; silence in the traditional SETI bands.
Objections. The thermodynamic leg is the vulnerable one: the CMB floor of 2.7 K applies everywhere, so the rim's advantage over the solar neighbourhood (solar circle to a 25 kpc rim) is 1.0 if the radial gain is flat, 1.75 on the authors' own √R scaling (√(25/8.2)) before the floor, and ~1.3 once the floor is applied (a local radiation field of 0.5–1.0 eV cm⁻³ plus the CMB gives T = 3.6–4.0 K at the solar circle and 2.9–3.0 K at 25 kpc, a ratio of 1.25–1.36, and the floor caps the ratio at T_local/2.725 K ≈ 1.3–1.5 at any radius), not the floor-to-floor 10¹² of horizon sinks (realized gain ~10³–10⁷ only with carrier modes in the vacuum state (N_eff ≪ 1), and a factor 20 deficit otherwise; Paper A v2.12, §4.2), and the observational bind is real at both ends: structures radiating near the achievable 3–10 K floor peak at 300–1000 μm, outside the WISE bands entirely, so cold-end constraints require far-infrared or submillimetre surveys that have not been performed, while only the warm (≳100 K) end of the predicted population falls within WISE-class sensitivity (the Ĝ framework's stated reach is ~150–600 K; colder structures enter only with implausibly large emitting areas), where it has been neither sought as a targeted population nor found serendipitously (Wright et al. 2014b; Griffith et al. 2015). The metallicity leg is the more exposed of the two and has not been engaged here: it assumes in-situ transmutation is cheaper than harvesting already-processed heavy elements from the inner disk, an assumption the authors do not defend against the competing economics of simply relocating refined material outward.
Status. Untested as stated, but its premise survives intact in every later member; historically the first explicit "wrong place, wrong observable" Fermi solution grounded in computation thermodynamics.

3.3 Smart (2012; 2019): the transcension hypothesis

Claim. The developmental trajectory of complexity ("STEM compression" of space, time, energy, and matter) carries every sufficiently advanced intelligence toward black-hole-density confinement and, ultimately, out of the observable universe; the silence is a developmental constant, not a contingency (Smart 2012, 2019).
Mechanism. An evolutionary-developmental analogy: as in ontogeny, the endpoint is convergent and encoded in the dynamics, with black-hole-scale density as the attractor.
Predicted sky. A named target class, the "galactic transcension zone": an inner ring of each galaxy's habitable zone where civilizations that transcended long ago have collapsed to near-black-hole density, yielding a deficit of life-signature exoplanets in that ring relative to the outer habitable zone; Smart also states a timing claim, that the interval from the cessation of early low-power electromagnetic leakage (radar, radio, television) to reorganization at near-black-hole density runs ~400 yr; transiting-companion signatures are offered as a secondary, long-shot observable.
Objections. The terminal step (exit from the universe) invokes physics that is unspecified and untestable in principle, the standing criticism that motivated the "macro" variant (Section 3.7); the developmental analogy supplies direction but no selection theory over real objects; and the stated kill conditions are conditional on an inner-habitable-zone exoplanet census at a completeness no funded instrument delivers, so confirmation and exclusivity of the hypothesis is untestable with today's science even though the deficit it predicts is, in principle, a result observation could contradict — Smart's own concession is that "proving the existence and exclusivity of the transcension hypothesis with today's science may be impossible," a narrower claim than unfalsifiability tout court.
Status. Falsifiable in principle but untested at the completeness its own kill conditions require (Table 3); enormously generative as a research program: nearly every subsequent member cites it as the frame.

3.4 Vidal (2014; 2016): stellivores

Claim. Some observed astrophysical systems (specifically accreting binaries) may already be advanced civilizations feeding on stars: "stellivores." The Fermi question is inverted: they are not absent, they are misclassified (Vidal 2014, 2016). (Vidal 2011 is an earlier, independent statement of the related black-holes-as-attractor thesis and is discussed with the black-hole-infrastructure thread, Section 3.6, rather than here.)
Mechanism. Energy logic of accretion (correct, and quantitatively the strongest in the family pre-2023), combined with a cybernetic reading of binary-system regulation.
Predicted sky. The X-ray-binary and cataclysmic-variable populations contain technological members distinguishable, if at all, by anomalous regulation statistics.
Objections. Standard accretion astrophysics explains the cited phenomenology without remainder, so the proposal carries an extreme parsimony burden; Vidal (2016) proposes observational criteria intended to separate stellivore candidates from ordinary accreting binaries, but no statistic with demonstrated discrimination performance against real catalogues has been exhibited (the R accretion-regulation statistic of Swanson 2026c is the same class of statistic, developed for engineered IMBH flows rather than for binaries, and marks the route such a discriminator would take); and the claim's retreat to in-principle indistinguishability flirts with unfalsifiability.
Status. No proposed test has been carried out; the population-statistics route (regulation anomalies across accreting-binary catalogues) remains open and is the fair test (Section 7).

3.5 Sandberg, Armstrong & Ćirković (2016): aestivation, and the Bennett–Hanson–Riedel critique

Claim. Computation-maximizers harvest resources now but defer computation to the far future, when the CMB has cooled toward the de Sitter floor and each joule buys up to ~10³⁰ times more erasures; they are dormant now, hence silent (Sandberg et al. 2016).
Mechanism. The sink hierarchy of Section 2.3, exploited in time rather than in space.
Predicted sky. Quiet, resource-conserving infrastructure; suppressed stellar-disruption rates in controlled regions; possibly missing baryons curated against waste.
Objections. The decisive one is thermodynamic: Bennett et al. (2019) showed that erasure can be paid for at any epoch by transferring entropy into existing non-maximal-entropy reservoirs, and that this is not improved by waiting for a colder background, so dormancy confers no advantage. A separate argument, ours, is that free energy not collected promptly is irreversibly lost (stars burn regardless), so the optimal policy is aggressive present-epoch harvesting with storage, not dormancy; stored negentropy can be spent at any later epoch at that epoch's efficiency, so waiting confers no advantage over harvesting now and computing later, and storage itself carries only small maintenance costs. The critique has a second leg that matters for this review: Bennett et al. (2019) also observe that the present universe already contains vast reservoirs far from maximal entropy into which computation-generated entropy can be dumped, so the premise that erasure must be charged at the ambient CMB temperature fails today; that observation amounts to an endorsement of present-epoch cold sinks, which the horizon-sink thread (Sections 2.3 and 3.6) radicalizes to its physical extreme. The authors' premise that erasure cost dominates the budget is also model-dependent: reversible architectures shift the optimum toward computing early. Aestivation survives only in attenuated form (harvest now, archive-grade computation deferred), at which point its observational predictions collapse into those of its neighbours.
Status. The critique has not been rebutted in print, though Sandberg has responded informally; the hypothesis's lasting contributions are the explicit defence policy, which yields two sharp falsifiable predictions (intervention against entropy-wasting astrophysical processes, and an archival search for gravitationally bound structures in excess of ΛCDM predictions), and the formalization of the efficiency calculus that the whole family now uses.

3.6 Black holes as infrastructure: Inoue & Yokoo, Hsiao et al., Dvali & Osmanov

Claim(s). Black holes are the terminal power plants and computers, argued at three different host-mass scales: collectors around accreting supermassive holes (Inoue & Yokoo 2011); Dyson-type harvesting of disk luminosity around a stellar-mass hole, with computed detectability (Hsiao et al. 2021); and black holes (including manufactured micro holes) as the most efficient quantum information processors, with a predicted high-energy neutrino signature from their evaporation (Dvali & Osmanov 2023). Vidal (2011) independently argues the general case, that black holes are attractors for intelligence on a two-dimensional Kardashev–Barrow development metric, without committing to a host-mass scale. A time-dilation variant has since appeared in Reiss & Feng (2026): civilizations on biologically tolerable orbits near the photon radius of a supermassive hole, maintainable well below the Type II threshold; with a time-dilation factor of 10⁴, a single Type II civilization could span the Galaxy within a crew lifetime in its own reference frame, and isotropic signals emitted on such orbits would exhibit a downward frequency drift.
Mechanism. Kerr accretion efficiency (5.7–30 per cent of rest mass, Thorne-capped at a_★ = 0.998; Bardeen, Press & Teukolsky 1972; Novikov & Thorne 1973; Thorne 1974), Blandford–Znajek spin extraction (Blandford & Znajek 1977; Tchekhovskoy et al. 2011), and horizon-saturated information physics. The thread has a technical prehistory older than any family member: Penrose (1969) supplied the rotational-extraction mechanism, Misner, Thorne & Wheeler (1973, §33.7) formalized the civilization-scale power-station thought experiment around it, and Thorne (1994) and Frolov & Novikov (1998, §12) carried the idea through the popular and textbook literature; Figure 2 records this as a grey precursor lineage feeding the present thread. The thread continued past the family's founding on a second front: Crane & Westmoreland (2009) asked whether starships or power sources could be built on the Hawking radiation of an artificially made black hole, judging the concept to be at the edge of possibility, which extends the harvesting route from rotational extraction to evaporation.
Predicted sky. Inoue & Yokoo (2011): a deliberate, sharp coherent electromagnetic beam transmitting harvested energy, a voluntary-class signature and the thread's earliest and most direct observable. Hsiao et al. (2021): a hot (~9.5×10⁴ K) thermal excess superposed on an otherwise standard accretion-disk SED around a stellar-mass hole, detectable in the ultraviolet through mid-infrared at Galactic distances with GALEX/HST-class photometry, discriminated by SED model-fitting plus radial-velocity variability. Dvali & Osmanov (2023): tens-of-TeV neutrinos and photons, in bursts for single-computer events and as a continuous flux for populations of simultaneous computations, detectable against IceCube- and MAGIC-class sensitivity; a genuinely novel observable since natural sources do not readily mimic it (Paper C operationalizes this as a band above ~10 TeV with multiplet criteria, an extension beyond the source paper's own tens-of-TeV figure, stated here explicitly). The micro-hole channel inherits a live controversy over whether radiation-collapse manufacturing is blocked by Schwinger pair production (Álvarez-Domínguez et al. 2024; Loeb 2024).
Objections. The selection problem here is not the absence of a stated mass scale; each member states one, and the three disagree (supermassive, stellar-mass, manufactured-micro) and none argues for its choice against the other two. A consequence for Table 3: the BH-computing row is graded best-leg across the three proposals (Inoue & Yokoo's galactic-club framing carries C1, Dvali & Osmanov's neutrino channel carries C3, Hephaistos-class Dyson searches bound the collector variant's C5), so the row's 3.5-point total is a composite upper envelope no single published proposal claims on its own; the SMBH orientation of Inoue & Yokoo (2011) in particular ignores the hazard and competition profile of galactic nuclei; detectability claims assume the civilization tolerates conspicuous luminosity, in tension with the efficiency motives that brought it there (though Inoue & Yokoo (2011)'s beamed channel is deliberately conspicuous by design, not an involuntary leak, which is the exception noted below); and the horizon-as-sink accounting inherits the radiator-temperature objection of Section 2.3, adjudicated there as ⊙-open: the carrier-delivery reply applies to this thread as much as to the MTH, and the re-interception step is equally unpriced.
Status. The Dvali & Osmanov (2023) neutrino channel is the family's most testable near-term prediction and is now the subject of a proposed dedicated monitoring program at a named target (Swanson 2026b); the high-energy SETI frame is surveyed by Lacki & DiKerby (2025), and the Dyson Minds workshop (Curtis et al. 2026) has since developed the observational agenda for the supermassive-host case, recommending anomaly-detection mining of WISE, JWST, and EHT archives. The compact-object case now also appears in the community review of technosignature searches: Vidal et al. (2026) close their stellar section with compact objects as attractive sites for advanced civilizations, black holes among the attractors for intelligence, and list as a signature a black hole whose rotational rate is slowing down anomalously, i.e. a sudden change in angular momentum (their §5.5), the spin-residue channel graded in Section 7. The distinction between voluntary and involuntary technosignatures (Haliki 2025) is useful here, with one exception stated plainly: Inoue & Yokoo (2011)'s galactic-club framing predicts a deliberate beamed signal, not a by-product, so the thread is not uniformly involuntary; for the efficiency-motivated members (Hsiao's collectors, Dvali's computers, and the MTH), an efficiency-driven occupant emits no beacons, so those observables are involuntary by-products, the class the residue catalogues of Sections 3.7 and 6 enumerate.

3.7 The Macro Transcension Hypothesis (2026)

Claim. The inward trajectory halts at a macroscopic, observable destination: rapidly spinning IMBHs embedded in dense, old, low-luminosity stellar systems, selected by a four-axis gradient (energy per unit fuel, entropy disposal, storage density, clock-rate control); migration and operation are electromagnetically quiet; Omega Centauri is the highest-ranked accessible target (Paper A).
Mechanism. The full Section-2 stack, plus a selection theory over real objects: intermediate mass (storage and erasure advantages saturate at 10^4–5 M_☉ while hazards grow), high spin (extractable energy), dense old clusters (fuel, stability, uncontested claim).
Predicted sky. Silence plus residues: anomalously high spin on gas-starved IMBHs; suppressed accretion relative to ambient supply; secular core depletion; possibly Dvali–Osmanov neutrino bursts. Six instrument-matched tests, of which T3 and T5 are the decisive kill tests and T1, T2 and T6 are constraint channels that no accumulation of electromagnetic non-detections can turn into a refutation (Paper A's falsification-test schedule, Table 7 as of v2.12; operationalized in Paper C). Where this review cites ω Cen's own cluster parameters, see Paper F App. A for the shared-parameter table across the set.
Objections (catalogued at full strength, given the disclosure of Section 1): it inherits the optimization-pressure premise unargued (Section 6); the staged architecture presupposes goal stability over 10⁸ yr; the flagship target's IMBH is itself contested (≲6×10³ M_☉ timing bound versus ≥8.2×10³ M_☉ kinematic bound; Bañares-Hernández et al. 2025; Häberle et al. 2024; the latest joint MeerKAT–Parkes pulsar timing, spanning 2021–2025, is insensitive to an IMBH of 10³–10⁴ M_☉ and constrains the central mass only to <10⁵ M_☉ at 90% confidence, Colom i Bernadich et al. 2026, leaving the tension unresolved); the class-level retreat (47 Tuc, M54, extragalactic nuclei) risks unfalsifiability-by-relocation, partially bound by the stated class prediction (no Galactic cluster IMBH with a_★ ≳ 0.9 falsifies it for the Milky Way); its erasure-cost advantage rests on the contested horizon-as-sink accounting of Section 2.3, adjudicated there as ⊙-open: the carrier-delivery reply is Paper A's own, so the unpriced re-interception step binds the MTH before any other member; and the spin residue, its sharpest discriminator, waits on LISA (Babak et al. 2017; Colpi et al. 2024).
Status. Untested; distinguishable from its relatives chiefly by having purchased falsifiability with named targets, thresholds, and timelines — the property the scorecard (Section 4) is designed to price.

3.8 The family at a glance

Table 1 condenses the six member treatments of this section into a single view and appends each member's scorecard row, so that the qualitative comparison above and the quantitative grading of Section 4.2 can be read side by side. The cells are condensed from the subsections, not new material; the symbols in the last column are those of Table 3.

4. Comparative analysis

4.1 The matrix

Table 2 aligns the family on five axes. Reading across rows reveals the internal disagreements that casual conflation hides: aestivation and transcension disagree about when (future versus now); transcension and the MTH disagree about where (fabricated density versus existing holes); stellivores and everyone else disagree about visibility (already visible versus invisible); and the black-hole-computing thread disagrees with the MTH about which holes (supermassive and manufactured-micro versus intermediate).

One asymmetry in the "core assumption" column requires disclosure. The MTH row states its premise in weakened subset-form ("a subset of civilizations maximizes computation") while its competitors are stated in the forms their authors used: universal for migration, transcension, aestivation and the black-hole thread, subset-form for the stellivores. The subset-form is what Paper A asserts, so the row is accurate, but the weakening buys explanatory cover that the universal forms do not enjoy, and it must be paid for: a subset-form premise faces the incomplete-compliance objection at full strength: partial compliance thins the expected loud population without collapsing the explanation outright, but non-adopters remain to be explained, and full dissolution of the silence puzzle requires either high compliance across the population or supplementary rarity from conventional Drake factors (P4, Section 6). Readers comparing rows should either read the competitors charitably in whatever subset-form their authors would accept, or charge the MTH the P4 penalty; the matrix should not be read as granting the newest member a weaker burden.

4.2 The falsifiability scorecard

Table 3 grades each member against six criteria. Five were fixed before grading in the original version of this review; the sixth, C6, was added in revision at a referee's prompting, and we disclose that provenance here rather than absorb it silently, because the scorecard's claimed virtue is criteria stated in advance and a quietly appended criterion would falsify that claim. The same standard applies to C5: the partial-credit clause governing its application was added at referee prompting in the revision that first awarded C5 half-credits (to migration, black-hole computing and the MTH, all ○ under the original wording); the migration credit has since been withdrawn, and the black-hole-computing and MTH credits remain. The criteria C1–C5 were fixed in v1.0; the partial-credit clause was added later. The criteria: C1 — names a target or target class that observations can interrogate; C2 — states quantitative observables with thresholds; C3 — at least one stated test is executable with existing or funded instruments within ~15 years, where "stated" means stated by the hypothesis's proponents; where the executable test was supplied by this review or by critics rather than by the proponents (the stellivore population statistic), the grade records that provenance; C4 — specifies kill conditions in advance (results the proponents agree would refute it); C5 — has survived at least one executed test that could have refuted it, with partial credit (⊙) where executed adjacent tests constrain the hypothesis without having been designed for it; C6 — resistance to unfalsifiability-by-relocation: when a stated test fails, the hypothesis either accepts refutation or retreats along a pre-bounded fallback that a finite series of nulls exhausts, rather than relocating indefinitely in target, strength, or observable. Symbols: ● met, ⊙ partially met, ○ not met. Where a criterion is tested by a multi-part test, the cell grades on the best executable part unless the parts are jointly necessary to the prediction, in which case the weakest part sets the grade; demonstrated discrimination performance against real catalogues is recorded as a reason a stated test fails executability under C3 rather than as a criterion of its own. Three caveats apply. First, C5 is the column that separates science-in-progress from science-performed: no member of the family scores ● on C5, because no pre-registered inward-family test has yet been executed to completion; the FAST globular-cluster survey (Huang et al. 2026a) and the proposed ω Cen campaign (Swanson 2026b) are the first entries in that ledger. The partial credits record real constraint from tests executed for other reasons: the collector variants of black-hole computing are bounded by Hephaistos-class Dyson searches (Suazo et al. 2024; Ren et al. 2025), and the accretion-limit measurements at ω Cen (Haggard et al. 2013; Chen et al. 2026; Mahida et al. 2026) already execute part of the MTH's suppressed-accretion test series; the migration hypothesis carries no such credit, since its warm-end Ĝ-survey band has been neither sought as a targeted population nor found serendipitously (Section 3). One conditioning clause on that last credit: it attaches to the limit-tightening restatement of Paper A's test T1, since the originally registered trigger threshold sat several orders of magnitude above achievable sensitivity and is restated in the current revision of Paper A; the ⊙ is contingent on that restatement, marked * in Table 3 and tested against the as-registered reading (○) under coding K4 below. Second, given the author's conflict of interest, the MTH row's grades are tied to specific Paper A artifacts rather than to judgement: C1 rests on the named flagship target and ranked fallback class (ω Cen; 47 Tuc, M54); C2 on the thresholds Paper A attaches to its LISA tests in its falsification-test schedule, a spin measurement a_★ < 0.1 at >3σ (T5) and an EMRI point mass below the kinematic floor, M < 8.2×10³ M_☉ (T4); the near-extremal spin a_★ ≳ 0.9 is not credited, because Paper A expects it on the null hypothesis as well as on the MTH. The T5 credit is branch-conditional: on a MAD-mode spin-up branch (Paper E, Section 3.4) the equilibrium spin is a_★ ≈ 0.035–0.07, at or below Paper E's natural-capture null and below the T5 trigger a_★ < 0.1, so the T5 leg of the C2 credit holds only on the thin-disc spin-up branch; the T4 threshold does not depend on the branch. The C2 credit sits alongside the quantitative ω Cen mass-tension table (Paper A's mass-constraint table, Table 6 as of v2.12) as a secondary, conventional-astrophysics observable rather than the credited threshold itself (no single Paper A table currently tabulates both mass and spin jointly); C3 on the LISA legs of the instrument-matched test schedule (T3, T5; Paper A's falsification-test schedule, Table 7 as of v2.12); and C4 on the two decisive tests of that same table. So that these grades are auditable without leaving this paper, the kill conditions credited are T3, a LISA signature of an extended remnant population or multiple low-mass inspirals rather than a single point mass, which removes H₂ entirely, and T5, a LISA spin measurement a_★ < 0.1 at >3σ, which removes the exploitation model (thin-disc branch; Paper A v2.10+). Paper A (v2.12) classes T1, T2 and T6 as constraint channels, not kills, and T4 as a demotion, not a kill, and states that no accumulation of electromagnetic non-detections, however deep, can falsify the MTH; none of the four is credited under C4. C3 is graded on the same two tests and is partial: LISA is funded and its mission window falls partly inside the ~15-year horizon, but the per-cluster event rate is uncertain by orders of magnitude (Section 7), so execution is conditional on an inspiral occurring. The JWST, Chandra, MeerKAT and KM3NeT legs (T1, T2, T6) are executable today, but C3 is read here as requiring a stated test that can adjudicate H₀ against H₂, and Paper A states that T1, T2 and T6 cannot (those hypotheses predict identical electromagnetic appearances), so these legs carry no C3 credit. A reader who credits them restores the full grade, a reading priced in the codings paragraph below, which also records the re-grade against v2.12. Third, the MTH's strong showing on C1, C2 and C4 was purchased after the criteria of falsifiability had been articulated by its predecessors' critics; later entrants always grade better on form, and the scorecard measures testability, not probability of truth.

Legend: ● criterion met  ·  ⊙ criterion partially met  ·  ○ criterion not met.

Four scorecard judgements require defence. Migration's ⊙ on C3 is proponent-stated, not review-supplied: Ćirković & Bradbury (2006) themselves recommend infrared surveys for Dyson shells and Matrioshka-class structures at the Galactic rim, and the census in Table 4 operationalizes that recommendation directly, splitting it into the warm (≳100 K) leg WISE already covers at area-limited depth and the cold (3–10 K) leg that no flown or funded far-infrared/submillimetre survey reaches. The half-credit reflects the warm leg's near-term executability against a stated zero-excess-candidate threshold, offset against the cold leg's lack of any funded instrument; it is not a claim of proponent-stated completeness across the whole predicted population. Aestivation's ● on C4 reflects Sandberg et al. (2016)'s explicit and admirable statement of refuting observations (uncontrolled entropy-wasting processes in regions a defender should police); its ⊙ on C3 splits two stated tests, both Sandberg et al. (2016)'s own: the process-suppression audit, which no instrument can execute at the required scale on any near-term horizon (○ on that leg), and the proposed archival test, a search for gravitationally bound structures more massive than ΛCDM predicts, on the reasoning that halo-scale curation would be rational and visible and its absence bounds energy-concerned aestivators (● on that leg, executable against existing survey catalogues); the two average to ⊙. This is a weakest-leg-average rule applied within a single proposal's two stated tests, a different rule from BH computing's best-leg envelope across three independent proposals (above); the two rules are not interchangeable, and the choice is disclosed rather than harmonized because the two rows differ in kind, one proponent with two legs against three proponents with one leg each, but a reader applying best-leg uniformly would move aestivation's C3 to ●, tying it with BH computing at 3.5 for clear second place instead of trailing at 3.0. The "archival/feasible now" entry in Table 4 operationalizes the mass-function leg; the process-suppression leg and the attenuated form's collapsed predictions (Section 3.5) remain the harder, untested core. Transcension's ⊙ on C1 and C4 credits the named transcension-zone target class and Smart's own stated kill condition (a persistent inner-habitable-zone life-signature population); it is not raised further because the completeness that condition requires sits beyond any funded census, and because the terminal step, exit from the universe, remains unobservable in principle. This is not a dismissal of the hypothesis's generativity (half the family descends from it) but the direct application of the criteria to a proposal whose falsifiable core (the exoplanet deficit) is real and stated, while its endpoint is not; Smart (2019) adds developmental argument, not additional observables. BH computing's ● on C3 is carried entirely by the Dvali & Osmanov (2023) neutrino channel, which predicted a high-energy neutrino signature from a source class that current telescopes can monitor (KM3NeT Collaboration 2025; IceCube Collaboration 2020); the specific energy band and multiplet criteria are as operationalized in Paper C. Finally, the MTH and aestivation now both grade ⊙ on C3. The MTH's credited tests are the LISA legs (Table 4), which are funded but depend on an in-band inspiral whose per-cluster event rate is uncertain by orders of magnitude (§7); its T1, T2 and T6 legs are executable today but are constraint channels in Paper A's v2.12 classification and carry no C3 credit. Aestivation's half-credit comes from the mass-function leg alone, and the process-suppression leg that would close the gap has no equivalent funded instrument on any announced schedule.

C6, the revision-stage criterion, requires its own defence, since it grades the axis on which this review's prose is hardest on its own member. The MTH's class-level retreat (47 Tuc, M54, extragalactic nuclei; Section 3.7) is bounded within the Galaxy by a stated class prediction that a finite series of spin measurements exhausts, but the extragalactic fallback remains unbounded: ⊙. Aestivation's response to the Bennett et al. (2019) critique was retreat to the attenuated form, which is relocation-in-strength; its full-strength kill condition remains on the table, so it also grades ⊙. Stellivores' appeal to in-principle indistinguishability is relocation-in-observable, with no bounded fallback stated: ○. Transcension's endpoint is defined as unobservable, so there is nothing to relocate from and nothing to resist with: ○. The migration hypothesis makes a population-level claim over the whole outer Galaxy and can retreat in temperature and depth though not in place: ⊙. BH computing grades ⊙: the Dvali & Osmanov (2023) channel closes at a stated strength, but the thread's missing selection theory over holes leaves target relocation open. The column's outcome is uncomfortable for the reviewer in a useful way: the newest member shares the column's top achieved grade with three of its relatives, ahead of two, and no member meets the criterion outright.

The half-grade handling is worth testing rather than asserting. Re-grading Table 3 under four alternative codings, recorded in advance in the accompanying script (figs/fB_calc7_v1.py): counting a partial as unmet everywhere (K1); counting it as met everywhere (K2); replacing the one documented two-leg average (aestivation on C3) with its weaker leg (K3); and withdrawing MTH's contingent C5 credit, the half-grade tied to the limit-tightening restatement of Paper A's T1 threshold rather than to the threshold as originally registered (K4). An earlier version of K4 withdrew the migration hypothesis's own C5 partial; that partial is already ○ in the printed table above (its warm-end search was neither sought nor found, Section 4.2), so the coding was a no-op and K4 now targets the one contingent half-grade the table still carries. Baseline (Table 3 as printed): MTH 4.5 (1st), BH computing 3.5 (2nd), aestivation 3.0 (3rd), migration and stellivores tied at 1.5 (4th–5th), transcension 1.0 (6th). The re-grade of the MTH row against Paper A v2.12 lowers its total from 5.0 to 4.5 and its margin over BH computing from 1.5 to 1.0 points; the MTH stays first. It ranks first singly under K1 (3.0, against 1.0 for each of the next three), K3 (unchanged at 4.5), and K4 (4.0, a 0.5-point margin over BH computing), and ties BH computing under K2 (6.0 each, the reading that converts every partial to a full grade), as before the re-grade. If the T1, T2 and T6 legs were credited under C3 despite the classification above, the baseline would return to 5.0 and K1, K3 and K4 to 4.0, 5.0 and 4.5. The mid-table is less stable: K1 flips the migration/stellivore order (stellivore rises to a three-way tie with aestivation and BH computing at 1.0, while migration falls to last, tied with transcension at 0.0), and K2 ties stellivore with transcension at 2.0. On C6 specifically, no coding separates the column's four ⊙ grades or lifts its two ○ entries, so the closing judgement above does not depend on how partial credit is counted.

5. The sociological competitors

Three well-known solutions predict the same silent sky from non-thermodynamic premises, and the family is best understood against them. A fourth class, the expansion-slowing models, sits between the sociological solutions and the inward family and is treated last.

Zoo (Ball 1973) and its hegemony refinement (Forgan 2017). Silence as policy: we are observed but quarantined. Forgan's contribution is structural — a zoo requires either a single dominant culture or stable inter-civilization consensus ("Galactic Club"), and fragmented "cliques" should leak defectors. The premise is sociological uniformity over megayear timescales, the premise the inward family replaces with physics.

Dark forest (Brin 1983; Yu 2015). Silence as survival strategy under first-strike game theory; the name comes from Liu Cixin's 2008 novel, and the academic formulation is due to Yu (2015), with Brin (1983) as the earlier "deadly probes" antecedent. Its physical premises are at least defensible (relativistic kill vehicles, asymmetric detection), though the kill-vehicle and targeting literature is disputed on interstellar targeting accuracy, light-lag, and deterrence grounds, and its deadly-probes mechanism has received technical treatment: Forgan (2019b) models mutated "predator" probes suppressing a visible probe population with Lotka–Volterra dynamics and finds that stable equilibria typically retain large prey populations, undercutting the mechanism as a full silence-explanation. The equilibrium is also fragile: the strategy must be adopted by every civilization without exception and without deviation forever, since a single loud defector (or a single expansionist willing to absorb risk) breaks the explanatory account (Brin 1983; Forgan 2017). Like the zoo, it predicts silence as the intended outcome of policy; the observables it admits (a defector's signal, a surviving probe population) are contingent on the policy being broken or policed rather than traces that follow from the premise, and no adjudication is scheduled.

Sustainability (Haqq-Misra & Baum 2009). Exponential expansion is self-terminating; civilizations that persist are those that grow slowly or not at all. This is the mildest competitor, and partially complementary: it removes the pressure to expand without supplying a positive account of where mature capability is directed.

Expansion-slowing models (Landis 1998; Carroll-Nellenback et al. 2019). A distinct class predicts partial silence from the dynamics of expansion itself rather than from psychology or policy. Landis (1998) models colonization as a percolation process: with finite settlement probabilities and range limits, expansion fronts can stall and leave permanent unvisited voids, one of which could contain the Sun. Carroll-Nellenback et al. (2019) couple settlement fronts to stellar motions and planetary habitability windows and find steady states in which the Galaxy is substantially but incompletely settled, with resettlement lag producing local voids (the "Aurora effect"). These are the inward family's closest rivals among silence-explanations, closer than the zoo or the dark forest, because they are grounded in physics and dynamics rather than in universally adopted strategy, and they are falsifiable in population statistics (void geometry, settlement-front morphology, the age distribution of settled systems). They differ from the inward family in what they explain: they account for local emptiness while leaving expansion as the governing drive, whereas the inward family removes the drive itself. The two are compatible, and a mixed population would show both signatures.

The comparison clarifies what the inward family is actually claiming. All these proposals (sociological and thermodynamic alike) break the expansionist premise. The differences are two. Robustness: dark-forest and zoo equilibria are fragile to a single defector among all civilizations across all time, whereas the inward account requires no coordination at all — each civilization independently follows its own resource gradient, and partial compliance thins the expected loud population without collapsing the explanation outright, though non-adopters remain to be explained and full dissolution requires either high compliance across the population or supplementary rarity from conventional Drake factors (P4, Section 6; the thinning made quantitative in the grabby-aliens frame, Hanson et al. 2021). Residue: fear and policy predict nothing involuntary: their observables are contingent on a policy being adopted, defected from, or policed, so a null is weak evidence against them rather than none (Section 5); thermodynamic optimization is a physical process operating on physical objects, and physical processes leave traces — spin histories, depletion statistics, suppressed accretion, burst channels. On both counts the expansion-slowing models score with the inward family rather than with the sociological solutions: they need no coordination and they predict population-level statistics that surveys can test. Among the silence-explanations, these two classes form the falsifiable-in-principle wing, and grading its members' falsifiability in practice (Table 3) is therefore where a review can add value.

6. Open problems

We state the family's unsolved problems as research questions rather than objections, since several are tractable.

P1: The optimization premise. Every member assumes some civilizations converge on maximizing computation (or a correlate). The premise is defended by universality arguments, computation is the currency into which most terminal goals are convertible (Smart 2012; Sandberg et al. 2016; Bostrom 2003), but it has never been given a selection-theoretic foundation: under what competitive or evolutionary dynamics do computation-maximizers come to exist, persist, and dominate the behaviour of their lineage? The grabby-aliens machinery (Hanson et al. 2021; Olson 2015) could be adapted to model mixed populations (maximizers, expansionists, satisficers) and derive the observable consequences of each mixture; the companion economics paper supplies a capture term for that machinery but defers the full integration (Swanson 2026d). The physical side of the premise has begun to accumulate its own literature. Wright (2023) applies the Landsberg formalism to Dyson spheres as work extractors and computational engines, finding little to no advantage in nesting shells and an optimum in small, hot spheres, which prices the radiator constraint the Bradbury lineage discovered. Garrett (2026) works the cold end without horizons: computation-dominated civilizations may dissipate at temperatures far below the 100–600 K window of existing waste-heat searches, approaching the CMB floor at large circumstellar radii, with the required radiating area growing as T⁻⁴; the predicted “Slysh halo,” a partial Dyson swarm radiating grey, line-free thermal emission in the far infrared and submillimetre, is constrainable from archival surveys at the 10²⁰ W level, a cold-sink route that needs no horizon.

P2: Goal stability over 10⁸ years. Inward strategies require coherent purpose across timescales exceeding mammalian evolutionary history, flagged as the largest unargued premise even by the family's own proponents (Paper A). The mitigation argument (front-loaded payoffs leave permanent residues even if the lineage fragments mid-program) rescues the observational program but not the hypothesis's interior logic. The family's deepest dependency is therefore a theory of institution or value stability under self-modification, a problem that lies outside astronomy.

P3: Migration economics. Bostrom (2003)'s opportunity-cost argument prices delay; Bennett et al. (2019) showed that dormancy confers no thermodynamic advantage; nobody has priced relocation. The trade is concrete: a migrating lineage abandons its accumulated local infrastructure for transit times of 10^4–6 yr in exchange for a destination whose advantages (Section 2) are enormous but deferred. Under what discount rates, risk models, and replication strategies does migration dominate local densification (Matrioshka-style; Bradbury 1999)? This is a well-posed optimization problem, and it has since received a first quantitative treatment in a companion paper (Swanson 2026d), which derives closed-form crossover thresholds among staying, migrating, and seed-dispatch strategies. We retain the problem in this list because the treatment's inputs (discount and hazard rates over agents never observed) remain open, and because its population-level corollaries feed directly into P4.

P4: Incomplete compliance. If migration is optional, the loud expansionists we do not see still need explaining. The defensible position (argued in Paper A): partial compliance thins the expected loud population without collapsing the explanation outright, since each civilization independently follows its own resource gradient, but non-adopters remain to be explained, and full dissolution requires either high compliance across the population or supplementary rarity from conventional Drake factors, not thinning alone. Quantifying "how much thinning suffices" against the Hanson et al. (2021) selection effect is open problem P1's observational face.

P5: Degeneracy of silence. H₀-type explanations (gas starvation, rarity) and inward-family explanations predict identical electromagnetic appearances; only residues discriminate. The family therefore lives or dies on the residue catalogue (spin, depletion, bursts, regulation statistics) and enlarging that catalogue with new discriminators is the highest-leverage theoretical work available. A companion paper takes a first step by forward-modelling two such discriminators for engineered IMBH systems, a waste-heat floor and an accretion-regulation statistic, within a hierarchical Bayesian adjudication framework (Swanson 2026c). The contested micro-black-hole manufacturing channel (Álvarez-Domínguez et al. 2024; Loeb 2024) is a case in point: it would be a sharp discriminator if the manufacturing route is physical, and it drops out of the catalogue if Schwinger pair production blocks it.

7. The observational program

The family is no longer untestable in practice; Table 4 maps each member to its executable tests. Four clusters of activity dominate.

(i) Globular clusters as the family's natural laboratory. Old, dense, dynamically relaxed, electromagnetically quiet, and, for the IMBH-hosting subset, equipped with the objects the gradient selects. The niche was identified by Di Stefano & Ray (2016), who argued that globular clusters are "sweet spots" for long-lived civilizations on habitability and stellar-density grounds and proposed them as SETI targets; the first dedicated globular-cluster survey (Huang et al. 2026a) operationalized it a decade later. The habitability argument is independent of and complementary to the thermodynamic-gradient argument of Section 2: the two select the same targets for different reasons. Gaia-based dynamical-extremeness metrics now exist for ranking clusters as technosignature targets: the phase-space crystallization index of Huang, Tao & Zhang (2026b), computed across 79 Galactic globular clusters, places ω Cen and 47 Tucanae among the small high-structure tail of the population. Two properties of that result matter for this review. It is independent evidence in the residue channel the scorecard prices, model-agnostic, kinematic, and authored outside this paper set; and it is simultaneously a null, since the same analysis finds every recovered substructure consistent with known dynamical processes, excludes missing cold shells above the few-per-cent level by Monte Carlo injection, and frames the index as a search-prioritization tool rather than as evidence of engineering, a framing this review adopts unchanged. The mundane origin of ω Cen's phase-space complexity has direct spectroscopic support in its traced tidal tails and accretion debris (Kuzma et al. 2026). A practical corollary follows from geography: the survey that operationalized the target class (Huang et al. 2026a) cannot reach ω Cen from FAST's latitude, so the metric's top-ranked cluster has never been searched at narrowband depth, and completing that program from the southern hemisphere (MeerKAT, SKA-Mid) is an open observational slot that the campaign of Paper C claims. The ω Cen multi-messenger campaign (Paper C) is the first program designed end-to-end around inward-family falsification, with the conventional astrophysics (the IMBH mass tension; Häberle et al. 2024; Bañares-Hernández et al. 2025) guaranteeing scientific value under every outcome. That tension remains live: the latest joint MeerKAT and Parkes timing of the cluster's millisecond pulsars (2021–2025) is insensitive to an IMBH in the contested 10³–10⁴ M_☉ range and bounds the central mass only from above, at <10⁵ M_☉ at 90% confidence (Colom i Bernadich et al. 2026), so the timing–kinematics discrepancy awaits deeper data. The cluster's compact-object inventory has meanwhile begun to resolve directly: 23 years of HST and JWST astrometry have yielded the first stellar-mass black hole detected astrometrically in any globular cluster, the 4.46 M_☉ oMEGACat BH-2 in a 94-yr binary (Whitaker et al. 2026), a single low-mass detection rather than a representative member of the ~10⁴-strong, 31 M_☉-mean remnant population the cluster's extended dark mass implies via the pulsar-timing estimate (Bañares-Hernández et al. 2025; Swanson 2026c).

(ii) The spin channel. Among all residues, black-hole spin is unique: permanent, quantitative, and measurable to ~10⁻³ precision by LISA for favourable systems (Babak et al. 2017; Amaro-Seoane 2018; Colpi et al. 2024). The precision figure applies to detected EMRIs, and the channel therefore depends on event rate: a compact-object inspiral must occur in-band during the mission, and the per-cluster rate for a ~10⁴ M_☉ hole is uncertain by orders of magnitude, which is the spin channel's weak link and part of the P5 accounting. Under Paper A's coherent-growth prior a near-extremal spin is expected on the null as well (Section 4.2), so it discriminates only in conjunction with electromagnetic silence; a low spin (a_★ < 0.1, Paper A's T5) retires the exploitation models on the thin-disc spin-up branch. No other proposed technosignature combines this permanence with a funded instrument.

(iii) The neutrino channel. The Dvali & Osmanov (2023) burst phenomenology is testable now at named targets with operating instruments (KM3NeT Collaboration 2025; Albert et al. 2020); the pre-registered multiplet criteria are specified in Paper C. A decade-scale null closes the channel at its stated strength.

(iv) Population statistics over single objects. For stellivores and attenuated aestivation, the executable tests are statistical: regulation anomalies across binary catalogues; suppression statistics across old systems. These reuse public archives, cost salaries only, and would give the family's two least-tested members their first empirical constraints.

The program has a unifying property: every test doubles as conventional astrophysics (IMBH demographics, accretion physics, binary populations, cluster dynamics), so the field can pursue the family's adjudication without betting careers on its truth. The pattern is inherited from the waste-heat surveys (Wright et al. 2014a; Griffith et al. 2015) and from the falsificationist turn of Sandberg et al. (2016) and Dvali & Osmanov (2023), and it does not depend on the fate of any individual hypothesis.

8. Conclusion

Reviewed as a family, the inward hypotheses form a six-decade research lineage with a common physical premise, the thermodynamics of computation as the lens for predicting mature intelligence, and genuine internal disagreements about timing, destination, and visibility that observation can, increasingly, adjudicate. The family has moved from direction without destination (Ćirković & Bradbury), to a destination whose stated kill conditions outrun any funded instrument (transcension), to testability gained under critique (aestivation and its Bennett–Hanson–Riedel correction). The current generation states explicit targets, thresholds, and kill criteria. Each advance was forced by the previous generation's critics.

Against the sociological silence-solutions, the family's distinguishing asset is residue: physics leaves traces where policy does not. Its distinguishing liability is the optimization premise, which remains unargued at the selection-theoretic level where it would need to be argued (open problem P1). The practical situation of 2026 is: for the first time, members of this family have pre-registered tests scheduled on funded instruments, anchored by the globular-cluster surveys now beginning and by LISA's spin measurements a decade out. C5 already carries two partial credits (BH computing, MTH; Table 3), but no member has yet scored a full grade there, since no pre-registered inward-family test has been executed to completion. Within fifteen years that should change, for a literature long accused of unfalsifiability.

Acknowledgements and disclosure

The author thanks the maintainers of the NASA Astrophysics Data System and arXiv, on which the citation verification for this work relied. The author's conflict of interest as proponent of one reviewed hypothesis is disclosed in Section 1 and mitigated by the fixed-criteria comparative method of Section 4. AI assistance disclosure: drafting, citation verification, derivation checking, and figure preparation for this manuscript were performed with substantial assistance from a large language model (Claude, Anthropic), under the author's direction; the author reviewed and takes full responsibility for all claims, derivations, and references. Interactive calculators implementing the quantitative material in Section 2 are available at omegacentauri.me.

Revision record

v2.6 (2026-10-22). The thermodynamic objection to aestivation was re-attributed: Bennett et al. (2019) argue that erasure cost is not lowered by waiting, and the harvest-timing argument (free energy not collected promptly is lost) is stated as this review's own. The Ren et al. contamination count was corrected from "four more" to "five more" (two significant, one suggestive, four marginal; two of nine without an identified contaminant). Bibliographic corrections: the Carroll-Nellenback DOI, the Carrigan name order, missing DOIs for Ren (2025), Huang (2026) and the Penrose reprint, volume and page data for Chen (2026), Whitaker (2026) and Kuzma (2026). The scorecard robustness file figs/fB_calc7_scorecard.json was regenerated against the v2.5 table; the totals for MTH under the five codings are 4.5, 3.0, 6.0, 4.5, 4.0.

v2.7 (2026-10-22). The remaining misreadings of Bennett et al. (2019) were corrected: in the P3 migration-economics paragraph (dormancy confers no thermodynamic advantage, replacing "price dormancy") and in the lineage diagram, where the critique node now reads "waiting does not lower erasure cost".

v2.8 (2026-10-23). Fact fixes from the farm audit D274. The advertised 10¹²⁺ gain in the dormancy discussion and the rim comparison is restated as in Paper A v2.12 (§4.2): ~10³ (fed flow) to ~10⁷ (ambient) only with carrier modes colder than 2.7 K, of order unity otherwise (superseded in v2.9 item 8), against a floor-to-floor 10¹². The Margolus–Levitin figure is given beside the abstract's 3×10³³, and the Bekenstein–Hawking entropy of a 2×10⁴ M_☉ hole is 6×10⁸⁵ bits (was 10⁸⁶). "Six-decade" becomes "two-decade" (Dyson's 1979 bound to Bradbury's 1999 engineering, the span the sentence names), "members" becomes "sources", and the Paper C neutrino band is stated as above ~10 TeV. The scorecard is graded against Paper A v2.12 (was v2.9); T5 is annotated as a thin-disc-branch kill (Paper A v2.10+); the T1, T2, T4 and T5 thresholds and their kill, demotion or constraint classifications are unchanged between the two versions (T5 gains a thin-disc-branch qualifier), and Tables 6 and 7 keep their numbers.

v2.9 (2026-10-24). Fact fixes from the R10 hostile-referee round, adjudicated by the farm (D300) and independently checked. (1) The scorecard section discloses that the C5 partial-credit clause was added at referee prompting in the revision that first awarded C5 half-credits, alongside the existing C6 disclosure; the list of review-supplied C3 legs drops the aestivation audit, which Table 3 does not mark. (2) The radial-gain comparison of the migration entry gives the three readings (1.0 flat, 1.75 on the √R scaling, ~1.3 with the CMB floor, capped at 1.3–1.5 at any radius) and states that the √R law requires T ∝ R^−1/2. (3) The Bekenstein sentence of the physics section no longer says the bound favours compression in total capacity: at fixed mass it scales as MR, and a 1-pc cloud of 2×10⁴ M_☉ has a bound 5×10⁸ times the hole’s. (4) The claim of horizon sinks colder than any reachable CMB is limited to eras reachable without ~10¹² yr of waiting (18 decades separate the IMBH horizon from the de Sitter floor). (5) Dark-forest and zoo residue statements are reconciled with the probe-population and defector observables: fear and policy predict nothing involuntary. (6) The competitors’ premises are described as universal for migration, transcension, aestivation and the black-hole thread and subset-form for the stellivores. (7) The abstract’s feasibility claim is scoped to 2026–2040 with the cold-infrared and LISA legs named. (8) The Paper A gain condition is restated as the vacuum state (N_eff ≪ 1), with the thermal case a factor 20 deficit (Paper A v2.14). The scorecard artefact metadata is updated to this version. The Hephaistos counts are reconciled against the full text of Ren et al. (2026) (Sections 1 and 3.5): nine analysed candidates, labelled A–F and H–J, the original seven less G plus the three added candidates H, I and J, ten sources in all. Not applied: the search-methods paragraph (needs the author’s actual sweep protocol).

Data availability

No new observational data were generated for this work. All quantitative claims derive from the cited literature.