CN-10 — Bats: active inference you can measure

What you are building. The wing's keystone chapter for the method. Everywhere else in this wing, active inference is an abstraction laid over a system that was not built to expose it. A bat exposes it. A bat emits energy at a metered cost in order to receive an informative return, changes the emission to resolve a specific uncertainty, and does all of it in Hz, dB, ms and metres — quantities you can put a microphone on. This is the one animal where NA-02's expected-free-energy epistemic term stops being a metaphor and becomes an instrument reading. Every number below either carries a source you can check or is written NOT-MEASURED / NOT-SOURCED. Nothing here raises any UNI rung — a citation to sensory biology is never a UNI gate.

---

Why this chapter is the keystone, stated as a claim that can fail

NA-02 decomposes expected free energy as G(pi) = risk + ambiguity = −(epistemic) − (pragmatic). The epistemic term says: an agent will pay to reduce uncertainty, before that reduction pays out in goal terms. It is the whole reason active inference is not control theory with a prior. It is also the hardest term to see in a real animal, because information-seeking normally hides inside behaviour doing three other things at once.

A bat's call does not hide it. The call costs (measurable in joules and dB). The call returns information and nothing else — the echo does not feed the bat, warm it, or move it. And the call is adjusted in flight, on a millisecond timescale, in the direction that reduces uncertainty about the hypothesis currently in question. Emission cost paid for information return, both sides on a meter, is the epistemic term with wires attached.

The active-inference reading is a lens, fenced HYPOTHESIZED throughout. The measurements are not the lens; they stand whether or not the lens is any good.

The one calculation: lambda = c/f, and why a bat cannot be a baritone

Take the speed of sound in air, c ≈ 343 m/s at 20 °C, 1 atm (standard acoustics; c ≈ 331.3·sqrt(1 + T/273.15) m/s — not primary-sourced in this pass). Wavelength is lambda = c/f. Read the ladder:

f	lambda = c/f	What that size is
10 kHz	34 mm	bigger than most moths
20 kHz	17 mm	a large moth's wing
50 kHz	6.9 mm	a midge
100 kHz	3.4 mm	a small fly
212 kHz	1.6 mm	a gnat

At 50 kHz, lambda = 343 / 50000 = 6.86 mm ≈ 7 mm. That single line is why bats are ultrasonic. A target much smaller than the wavelength does not reflect a sound wave so much as get walked around by it — the returned energy collapses (the Rayleigh regime). An echolocator wanting an echo off a chironomid with a 9–12 mm wingspan cannot work at 10 kHz, where lambda is three times the animal. It has to shorten the ruler, and shortening the ruler is raising the frequency. There is no third option.

The measured range across species is consistent with exactly this pressure. Bat calls span roughly 9–11 kHz to 212 kHz in dominant (peak) frequency, with aerial-feeding assemblages on four continents dominated by 20–60 kHz (Fenton, Portfors, Rautenbach & Waterman 1998, Can J Zool 76(6):1174–1182, DOI 10.1139/z98-043 — that 20–60 kHz assemblage result is what Fenton 1998 measures; the full span is NOT-CONFIRMED to that paper in this pass). At the top, Cloeotis percivali is reported at a 212 kHz carrier — lambda = 1.6 mm. At the bottom, Euderma maculatum calls at ~9–12 kHz (Fullard & Dawson 1997, J Exp Biol 200:129–137) and eats eared moths — that low frequency is not about target size at all, but about being inaudible to the prey. One end of the range is set by physics, the other by an arms race.

Two scope notes the range needs, or it is not a number. First, the endpoints are not the same measurand: the 212 kHz figure is a CF carrier, the 9–12 kHz figure a dominant/peak frequency. For a CF bat the carrier is the dominant component, so the span is readable — but only if the metric is named, and it is named here rather than left to the reader. Second, the floor is the same animal as the low endpoint. The literature's canonical phrasing puts the floor at 11 kHz and names E. maculatum as the species setting it — while this chapter's own f_Euderma row reports 9–12 kHz across sources. An 11 kHz floor and a 9 kHz call by the species that sets that floor are one measurement reported at two precisions, not two findings. The floor is therefore printed ~9–11 kHz, species-dependent, instead of being contradicted two rows later by its own source species.

Carry the citation chain honestly. The 212 kHz figure is stated by Thiagavel, Santana & Ratcliffe (2017), Sci Rep 7:828, DOI 10.1038/s41598-017-00959-2, citing Bell & Fenton (1984), Behav Ecol Sociobiol 15:109–114 — a paper about Hipposideros ruber. The primary Cloeotis measurement was not confirmed in this pass. Per M22, an upstream statement is not fresh evidence. The number is printed with its chain visible, not laundered.

And now the part it would be dishonest to omit. The wavelength argument is a derivation, it has been tested against real insects, and it did not fully survive. Waters, Rydell & Jones (1995), Behav Ecol Sociobiol 37(5):321–328, DOI 10.1007/BF00174136, measured echo target strength of actual prey and found it virtually independent of emitted frequency across 20–100 kHz — contrary to models built on spheres and disks. Real insects are not spheres. So the wavelength argument explains why bats are not at 5 kHz; it does not cleanly explain the choice within the band bats actually use. OBSERVED-CONTESTED, not smoothed over: the chapter's best derivation and its most instructive partial defeat.

What the frequency costs: the compromise the title of Fenton 1998 names

Frequency is not free. Atmospheric absorption rises steeply with it. Jakobsen, Brinkløv & Surlykke (2013), Front Physiol 4:89, DOI 10.3389/fphys.2013.00089, tabulate (from Lawrence & Simmons 1982 and ANSI 1995):

• 25 → 50 kHz: 0.7 → 1.7 dB/m (20 °C, 50% RH)
• 45 → 90 kHz: 1.4 → 4 dB/m (25 °C, 80% RH)

Put a target at 5 m — a 10 m round trip. At 45 kHz that is 10 × 1.4 = 14 dB of absorption; at 90 kHz, 10 × 4 = 40 dB. Doubling the frequency costs 26 dB of round-trip absorption alone, before spreading loss, before target strength. A bat buying resolution with frequency pays for it in range at a brutal exchange rate. Every echolocator sits somewhere on that curve; 20–60 kHz is where most of them sit. Cloeotis at 212 kHz has bought a 1.6 mm ruler and, by the trend, a very short reach — its absorption coefficient at 212 kHz is NOT-MEASURED here, and I will not extrapolate a curve past its sourced points to manufacture the number.

Range is time: 2R = ct

The echo carries range as delay. Sound goes out and comes back, so 2R = ct, i.e. t = 2R/c:

Target range R	Echo delay t = 2R/c
0.1 m	0.58 ms
0.5 m	2.9 ms
1 m	5.8 ms
2 m	11.7 ms
5 m	29 ms
10 m	58 ms

The conversion constant is 5.83 ms per metre of range. Everything downstream is bookkeeping against that number.

CF vs FM falls straight out of it — the two call architectures are two ends of the time–bandwidth trade:

• FM (frequency-modulated sweep) is for ranging and resolution. Matched-filter range resolution is Δr = c/(2B) for bandwidth B (standard sonar/radar theory; not primary-sourced in this pass). For a single-frequency pulse of duration tau, B ≈ 1/tau, so Δr ≈ c·tau/2 — resolution is chained to duration, and a 0.5 ms tone gives Δr = 343 × 0.0005 / 2 = 8.6 cm. Sweep the same 0.5 ms across B = 60 kHz (an illustrative input, not a sourced species value) and Δr = 343/(2 × 60000) = 2.9 mm — a ~30× gain at identical duration. FM buys resolution without paying duration. That is what a sweep is for.
• CF (constant-frequency tone) is for Doppler and flutter. Frequency resolution runs the other way: Δf ≈ 1/tau. Hipposideros armiger — a hipposiderid, not a rhinolophid horseshoe bat; the splice this chapter fences below starts here if the families are merged — holds echo frequency to a standard deviation of 110 Hz (Schoeppler, Schnitzler & Denzinger 2018, Sci Rep 8:4598, DOI 10.1038/s41598-018-22880-y). Now the premise, stated out loud, because the arithmetic below is worthless without it: 110 Hz is a measure of how tightly the bat stabilises its own emission — a motor-control statistic. It is not a measurement of the frequency structure the bat's receiver must resolve, and Schoeppler et al. do not claim it is. If the bat stabilises the echo to 110 Hz because 110 Hz is the scale it must resolve — an inference, not a measurement — then resolving structure at that scale needs tau ≥ 1/110 s ≈ 9 ms, and the CF component cannot be short. The arithmetic bounds it, conditional on that premise; it forbids nothing until the premise is argued. (Fourier resolution is a soft bound besides: coherent frequency estimation can beat 1/tau at high SNR. Measured CF durations are NOT-SOURCED here; the bound is derived, not quoted — and deriving it from real CF durations, rather than from a control statistic standing in for a resolution requirement, is the better move once they are sourced.)

FM is a ruler. CF is a tachometer. A bat needing both carries a CF-FM call and pays for both.

The pulse–echo overlap problem, derived rather than asserted

A call of duration tau is not a point in time; it is c·tau metres of air. The echo from range R starts arriving at 2R/c after the call starts. The call is still going until tau. Overlap-free ranging therefore requires 2R/c ≥ tau, i.e. R ≥ c·tau/2. Now put in the measured call durations from Moss & Surlykke (2010), Front Behav Neurosci 4:33, DOI 10.3389/fnbeh.2010.00033:

Phase	Duration tau	Pulse length c·tau	Overlap-free only beyond c·tau/2
Search	15–20 ms	5.1–6.9 m	2.6–3.4 m
Approach	2–5 ms	0.69–1.7 m	0.34–0.86 m
Terminal buzz	0.5–1 ms	0.17–0.34 m	8.6–17 cm

Read the right-hand column against the left. The bat's call duration collapses by a factor of ~30 across the attack, and the overlap-free floor collapses with it, staying just under the closing target. A search call that could not cleanly range anything nearer than 2.6 m would be blind at the moment of capture. The bat does not solve this with better processing; it changes its own emission so the question stays answerable. That is action taken to keep an inference well-posed, and it is measured in milliseconds.

The terminal buzz — and the refutation of its obvious explanation

Closing on prey, a bat drives its call rate up to beyond 160 calls/s (Elemans, Mead, Jakobsen & Ratcliffe 2011, Science 333(6051):1885–1888, DOI 10.1126/science.1207309), with Moss & Surlykke reporting up to ~170/s. This is the terminal buzz: the measurable signature of an agent throwing its information rate to the ceiling at exactly the moment uncertainty is most expensive. In NA-02's terms, the epistemic term is being maximised precisely where the pragmatic term is about to be settled.

Now the part worth the whole section. The obvious explanation of the buzz ceiling is pulse–echo overlap: at pulse interval PI the unambiguous range is c·PI/2, so 160 calls/s (PI = 6.25 ms) implies 343 × 0.00625 / 2 = 1.07 m — a tidy story in which the bat stops speeding up because it would start confusing echoes.

That story is refuted. Elemans et al. showed laryngeal motor performance, not echo overlap, sets the ceiling: bats have a previously unknown class of superfast muscle (the anterior cricothyroid, in Myotis daubentonii) producing positive work in cyclic contraction up to 160 Hz, and in one case 200 Hz. The constraint is the meat, not the mathematics. A clean, plausible, arithmetically correct hypothesis lost to a direct measurement of the actuator. Record which one won.

Doppler-shift compensation: the sharpest measured instance of action-to-optimize-inference

This is the strongest thing in the chapter.

A horseshoe bat's cochlea has an auditory fovea: a grossly expanded frequency representation centred on a narrow reference frequency, with sharply tuned neurons overrepresented at that frequency throughout the auditory pathway (Schnitzler & Denzinger 2011, J Comp Physiol A 197(5):541–559, DOI 10.1007/s00359-010-0569-6). In Rhinolophus ferrumequinum the inferior colliculus overrepresents best frequencies of 83.0–84.5 kHz (Schuller & Pollak 1979, J Comp Physiol 132:47–54, DOI 10.1007/BF00617731). The bat's model of the world is sharp in a ~1.5 kHz-wide slot and blunt outside it.

Flying at the target ruins this. For a bat closing at speed v on a stationary reflector, the echo returns at f_r = f_e·(c+v)/(c−v). At v = 5 m/s that factor is 348/338 = 1.0296 — an echo from an 83 kHz emission comes back at 85.4 kHz: ~2.4 kHz above the 83 kHz reference frequency, and roughly 1 kHz clear of the fovea's 84.5 kHz upper edge. (Both numbers are correct and they are not the same number; "2.4 kHz above the fovea" would collapse a 1.5 kHz-wide band to its lower edge. The ~1 kHz is the one that carries the argument.) The bat has flown its own signal out of the only band where its model is precise.

So the bat lowers its voice. To land the echo back on 83.0 kHz it must emit 83000 / 1.0296 = 80.6 kHz — dropping emitted frequency by ~2.4 kHz (computed here). The Rhinolophus control literature established that the animal does exactly this. Two quantities are involved and they are kept separate, because collapsing them manufactures a third that nobody measured: rhinolophids and P. parnellii hold F_echo to a precision of 0.1–0.2% around F_ref — the band, ≈83–166 Hz at F_ref = 83 kHz — and hold F_ref at approximately 150–200 Hz above F_rest — the offset. Both are stated in Schoeppler et al. (2018), who also measured in-flight echo frequency held to SD 110 Hz — 0.17% of the reference in H. armiger.

Sit with what that is. The bat cannot move its fovea. So it moves the world's input into the fovea, by changing its own motor output, continuously, in closed loop, at 0.17% precision. It is not adapting its model to its sensations. It is adapting its sensations to its model. Nothing else in this wing is this literal.

Type the connection to NA-02 correctly, because the obvious phrasing is a category error. It is tempting to call this "precision, gamma". It is not gamma. NA-02's gamma is policy precision — the softmax inverse-temperature over policies, sigma(−gamma·G), units nats⁻¹. The fovea is sensory precision: sharpness of the likelihood p(o|s). The correct statement is narrower and better: NA-02's G = risk + ambiguity, where ambiguity is the expected entropy of the likelihood, E_q(s|pi)[H[p(o|s)]]. An action moving the echo into the band where p(o|s) is sharpest lowers the ambiguity term of G directly. DSC is an ambiguity-minimising action. That is the exact hook, and it is worth more than the loose one.

Fence the splice. The fovea width (83.0–84.5 kHz) is R. ferrumequinum; the 110 Hz DSC precision is H. armiger, a hipposiderid whose auditory fovea is reported as less developed. The comparison "the action is ~14× finer than the fovea is wide" is illustrative across two species, not a within-animal measurement. The within-species pairing is NOT-MEASURED here, and its falsifier is in the table.

Reafference: predicting your own action's sensory consequence (cross-ref NA-04)

A bat emitting 140 dB SPL at 10 cm and then listening for an echo dozens of dB quieter has an obvious problem: it is about to deafen itself with its own voice.

It solves this in the shape NA-04's forward-model account describes — but by efference copy, not by temporal precedence, and the distinction is the whole of what Suga & Jen actually measured. Middle-ear muscle activity — chiefly the stapedius — is synchronous with vocalisation, not before it and not a reflex after it. In Suga & Jen's words (1975, J Exp Biol 62(2):277–311, in Myotis lucifugus), the muscles "received a message from the vocalization system when the bat vocalized, and contracted synchronously with vocalization," and "the duration of the contraction-relaxation was so short that the self-stimulation was attenuated, but the echoes were not." The muscles are driven by a copy of the motor command, not by the sound that command produces. That is the forward-model point, and it stands without any pre-vocal lead: the bat's ear is being told what the bat is about to do by the system doing it.

The timings that paper reports are the timings of the mechanism it was ruling out. Suga & Jen's latencies — 3–4 ms (electromyogram) and 4–8 ms (attenuation of the cochlear microphonic) — are acoustic middle-ear-muscle reflex latencies, and they are reported precisely in order to reject the reflex: "these muscles failed to attenuate orientation signals by the reflex." Reading those figures as a pre-vocal lead inverts the paper — it converts a refutation into its opposite. Any lead time of muscle onset before call onset is NOT-SOURCED here; see the table, where the claim is carried as an open row with its falsifier rather than as a number. Reported attenuation is 17–25 dB (the spread is real and printed). Destroying the stapedius reportedly abolishes the effect (Henson 1965 — cited by the above, not read in this pass).

Then there is a second, separately sourced measurement. Jakobsen et al. (2013) report that during approach the bat's receiver sensitivity falls by ~6 dB for each halving of target distance, attributed to the same muscles — an automatic gain control that tracks range, measured in dB. That much is observed.

What is not observed is the anticipation. Reading this schedule as open-loop against an echo amplitude the bat predicts from its own closing geometry — rather than as a loop driven by echo amplitudes already arriving — is HYPOTHESIZED: this pass sources the 6 dB/halving relationship but no experiment discriminating a predicted-geometry drive from an echo-driven one. The gain schedule is the measurement; the forward model is the reading laid over it. The efference-copy result above is the chapter's actual evidence for prediction-not-reflex, and it is enough — this row does not need to be conscripted into carrying more than it weighs.

Intensity: what the epistemic term actually costs, and the contest over it

The correct convention matters, and getting it wrong is a common defect. Source level is dB SPL re 20 µPa at 0.1 m (10 cm from the mouth) — Jakobsen et al. (2013).

• Open-space aerial hawkers: ~130 dB SPL, up to and beyond 140 dB (Surlykke & Kalko 2008, PLoS ONE 3:e2036, DOI 10.1371/journal.pone.0002036) — reported as the highest airborne levels for any animal.
• "Whispering" bats, long assumed to be ~70 dB, actually reach 110 dB SPL (Jakobsen et al. 2013).

At 140 dB SPL the acoustic pressure amplitude at 10 cm is 20e-6 × 10^(140/20) = 200 Pa (computed here) — about 0.2% of one atmosphere, from a mammal weighing a few grams.

Does that cost anything? This is the chapter's live scientific dispute, and it resolves along an axis, which is the best kind:

• Speakman & Racey (1991), Nature 350:421–423, DOI 10.1038/350421a0: mass-adjusted flight energy expenditure in echolocating bats was not significantly different from non-echolocating bats and birds. Title: No cost of echolocation for bats in flight. The mechanism offered is coupling call emission to the wingbeat — the call rides an exhalation the flight muscles are producing anyway. Holderied & von Helversen (2003), Proc R Soc B 270(1530):2293–2299, DOI 10.1098/rspb.2003.2487, found small and medium bats match their maximum detection range to their wingbeat period — the same clock serving both.
• Voigt & Lewanzik (2012), J Comp Physiol B 182:831–840, DOI 10.1007/s00360-012-0663-x, re-tested it in the 5 g Rhogeessa io: cost of transport was not related to pulse emission rate, and flight power was lower than predicted if call cost were additive. Consistent with Speakman & Racey.
• Currie, Boonman, Troxell, Yovel & Voigt (2020), Nat Ecol Evol 4(9):1174–1177, DOI 10.1038/s41559-020-1249-8, bounded it: costs are negligible only for low call intensities; above 130 dB SPL (re 10 cm) sound production becomes, in their word, exorbitant for small bats.

So "echolocation is free during flight" is not wrong — it is scope-limited, and 2020 printed the scope. Below ~130 dB the epistemic term is nearly free because it rides an action the animal is taking anyway; above it, the bat buys information with metabolism at a rate that caps how loud it can be. The price of information is not zero and not constant. It is a function with a knee, and the knee has been measured. For a method claiming agents trade energy for uncertainty reduction, that is about as good as nature gets.

Flight: the membrane wing

Bat wings are not bird wings, and the difference is not decorative. A bat wing is a thin skin membrane stretched over elongated arm and hand bones, with the digits themselves forming the leading and trailing edges — compliant, anisotropically stiff (stiffest along the bones), carrying membrane-tensioning muscles found in no bird, and offering a far wider range of in-stroke morphological adjustment (Hedenström & Johansson 2015, J Exp Biol 218(5):653–663, DOI 10.1242/jeb.031203). A bird changes wing shape by moving feathers over each other. A bat changes wing shape by changing the shape of the wing.

Strouhal (cross-ref NA-06). St = fA/U. Taylor, Nudds & Thomas (2003), Nature 425:707–711, DOI 10.1038/nature02000, found flying and swimming animals — bats included — cruise in 0.2 < St < 0.4, the band of high propulsive efficiency. The bat-specific measurement sharpens it: Lindhe Norberg & Winter (2006), J Exp Biol 209(19):3887–3897, DOI 10.1242/jeb.02446, filmed Glossophaga soricina across 1.23–7.52 m/s and found St = 0.17–0.22 near minimum-power speed (4–6 m/s), St = 0.25–0.40 at 3.4–4 m/s, and St = 0.5–0.68 below 3 m/s, where unsteady effects take over and lift/thrust production degrades. The honest wrinkle: the efficient cruise sits slightly below the canonical 0.2 bound. The band describes where animals cruise; it is not a law they obey.

Wing loading and aspect ratio. Norberg & Rayner (1987), Phil Trans R Soc B 316:335–427, DOI 10.1098/rstb.1987.0030, ran a PCA over 200+ species: high aspect ratio → open air; low aspect ratio → clutter; low wing loading → slower flight. The pattern lines up with the sonar — open-space bats are the loud, long-range ones. The numerical ranges for wing loading in N/m² and aspect ratio are NOT-SOURCED in this pass and are therefore not printed here, and bat aspect ratio is reported under both full-span and half-span conventions, so quoting one against the other is an error waiting to happen.

Jamming, and the arms race

Bertholdia trigona is a palatable tiger moth that answers an attacking Eptesicus fuscus with ultrasonic clicks. Corcoran, Barber & Conner (2009), Science 325(5938):325–327, DOI 10.1126/science.1174096, used ultrasonic recording plus high-speed infrared videography of live bat–moth interactions to show the clicks jam the bat's sonar — a controlled demonstration, not an inference from correlation. Corcoran, Barber, Hristov & Conner (2011), J Exp Biol 214:2416–2425, DOI 10.1242/jeb.054783, then discriminated among three candidate mechanisms — phantom echo, ranging interference, masking — and found bats missed by ~15–20 cm, the distance predicted by ranging interference. In the field, Corcoran & Conner (2012), J Exp Biol 215(24):4278–4287, DOI 10.1242/jeb.076943: clicking moths captured 6.8% of the time vs 71% for silenced conspecifics — a defence ratio of 10.4. Fernández, Dowdy & Conner (2022), J Exp Biol 225(18):jeb244187, DOI 10.1242/jeb.244187, added a dose–response: 77% capture at 0% duty cycle, odds of capture falling ~4% per 1% increase in moth duty cycle, and a jamming window of about 2 ms before echo arrival.

The other direction is on the same meter. An average noctuoid moth first hears Eptesicus fuscus at 20–25 m, but Euderma maculatum at less than 1 m (Fullard & Dawson 1997). The spotted bat pays the wavelength penalty — a 34 mm ruler, hopeless for small targets — to buy 20+ metres of acoustic invisibility against eared prey. That is why the bottom of the frequency range exists. (Reviewed in ter Hofstede & Ratcliffe 2016, J Exp Biol 219(11):1589–1602, DOI 10.1242/jeb.086686.)

The two-agent reading — an arms race as two agents each minimising their own surprise at the other's expense — is HYPOTHESIZED. It is a story about the measurements, not a measurement. No coupled-agent free-energy model has been fitted to these data in this pass, and none is claimed.

The numbers

Symbol	Value	Units	Scope	Class	Source	Falsifier
c	≈343	m/s	air, 20 °C, 1 atm; c ≈ 331.3·sqrt(1+T/273.15)	OBSERVED-REPLICATED (standard reference; not primary-sourced in this pass)	standard acoustics	Fetch a primary reference; a measured c outside 330–350 at stated conditions
f_range	~9–11 to 212	kHz	dominant (peak) frequency of the strongest call component, across species. Floor is species-dependent, set by E. maculatum (see f_Euderma) — printed ~9–11, not 11, so it does not contradict its own source species. The 212 kHz endpoint is a CF carrier (the dominant component for a CF bat, but not the same measurement as a peak-frequency estimate)	OBSERVED-REPLICATED (as a range) / span attribution NOT-CONFIRMED	Thiagavel et al. 2017, Sci Rep 7:828. Fenton et al. 1998's confirmed subject is the 20–60 kHz assemblage result, not the full span — the span is NOT-CONFIRMED to that paper in this pass and is no longer attributed to it	A species outside the range on the stated metric under stated recording conditions; and, separately: locate the 11–212 span in a primary or review source (Fenton 1998? ter Hofstede & Ratcliffe 2016?) and re-attribute this row to whichever carries it, or strike the span
f_mode	20–60	kHz	aerial-feeding assemblages: Canada, Mexico, Brazil, Zimbabwe	OBSERVED-REPLICATED	Fenton et al. 1998	A comparable assemblage not dominated by 20–60 kHz
f_Cloeotis	212	kHz	Cloeotis percivali carrier — chain-flagged (M22)	OBSERVED-CONTESTED / NOT-CONFIRMED	Thiagavel et al. 2017 citing Bell & Fenton 1984, Behav Ecol Sociobiol 15:109–114 (primary not read in this pass)	Read Bell & Fenton 1984; if the measurement is not there, the chain breaks
f_Euderma	9–12 (also reported ~10.5, ~12.7)	kHz	Euderma maculatum dominant/peak frequency — real spread across sources. This species is what sets the f_range floor, which is why that floor is printed ~9–11 rather than 11: an 11 kHz floor and a 9 kHz call by the floor's own source species are one measurement at two precisions, not a contradiction to leave standing	OBSERVED-CONTESTED	Fullard & Dawson 1997, J Exp Biol 200:129–137; ~10.5 in Thiagavel et al. 2017	Recording outside 9–13 kHz; or a study reconciling the reported values
lambda@50kHz	6.9	mm	lambda = c/f, c = 343 m/s	MODELED	Computed here	Arithmetic error, or c refuted
lambda@212kHz	1.6	mm	as above	MODELED	Computed here	as above
lambda@10kHz	34	mm	as above	MODELED	Computed here	as above
TS(f) insects	~independent of f across 20–100 kHz	dB	real prey items — contradicts sphere/disk Rayleigh models	OBSERVED-CONTESTED	Waters, Rydell & Jones 1995, Behav Ecol Sociobiol 37(5):321–8, DOI 10.1007/BF00174136	Measure TS of real insects across 20–100 kHz and recover a strong f dependence
alpha_atm	0.7 → 1.7	dB/m	25 → 50 kHz, 20 °C, 50% RH	OBSERVED-REPLICATED	Jakobsen, Brinkløv & Surlykke 2013, Front Physiol 4:89, DOI 10.3389/fphys.2013.00089, from Lawrence & Simmons 1982; ANSI 1995	Re-measure at stated T/RH; values >2× off
alpha_atm	1.4 → 4	dB/m	45 → 90 kHz, 25 °C, 80% RH	OBSERVED-REPLICATED	as above	as above
ΔL_2f	26	dB	round-trip absorption penalty, 45→90 kHz, R = 5 m (10 m path)	MODELED	Computed here from alpha_atm	Arithmetic error, or alpha_atm refuted
alpha_atm@212kHz	—	dB/m	absorption at Cloeotis's carrier	NOT-MEASURED	not sourced in this pass	Fetch ANSI 1995 / Lawrence & Simmons 1982 and evaluate at 212 kHz
t(R)	5.83	ms per metre of range	t = 2R/c, two-way	MODELED	Computed here	Arithmetic error, or c refuted
tau_search	15–20	ms	search-phase call duration	OBSERVED-REPLICATED	Moss & Surlykke 2010, Front Behav Neurosci 4:33, DOI 10.3389/fnbeh.2010.00033	Measured durations outside range for a search-phase FM bat
tau_approach	2–5	ms	approach phase	OBSERVED-REPLICATED	Moss & Surlykke 2010	as above
tau_buzz	0.5–1	ms	terminal buzz	OBSERVED-REPLICATED	Moss & Surlykke 2010	as above
R_min	2.6–3.4 / 0.34–0.86 / 0.086–0.17	m (search / approach / buzz)	overlap-free floor R ≥ c·tau/2	MODELED	Computed here from tau + c	Arithmetic error; or a bat ranging cleanly inside c·tau/2
rate_buzz	>160 (up to ~170)	calls/s	terminal buzz repetition rate	OBSERVED-REPLICATED	Elemans et al. 2011, Science 333(6051):1885–8, DOI 10.1126/science.1207309; Moss & Surlykke 2010	Recording of a terminal buzz capped well below 160/s
f_muscle	up to 160; 200 in one case	Hz	anterior cricothyroid, Myotis daubentonii, positive work in cyclic contraction	OBSERVED-REPLICATED	Elemans et al. 2011	Repeat the work-loop assay; a ceiling well below 160 Hz
buzz ceiling cause	laryngeal motor performance, NOT pulse–echo overlap	—	M. daubentonii	OBSERVED-REPLICATED	Elemans et al. 2011	A bat exceeding its measured muscle cycling limit; or overlap shown to bind first
R_unamb@160/s	1.07	m	c·PI/2, PI = 6.25 ms	MODELED	Computed here	Arithmetic error
f_fovea	83.0–84.5	kHz	R. ferrumequinum inferior colliculus, overrepresented best frequencies	OBSERVED-REPLICATED	Schuller & Pollak 1979, J Comp Physiol 132:47–54, DOI 10.1007/BF00617731	Map IC best frequencies and find no overrepresentation
SD_echo	110 (= 0.17% of F_ref)	Hz	Hipposideros armiger (a hipposiderid — NOT a rhinolophid/horseshoe bat), in-flight DSC precision. An emission-control statistic: how tightly the bat stabilises F_echo. It is NOT a measured resolution requirement of the bat's receiver, and the source does not claim it is — see tau_CF,min, which depends on reading it as one	OBSERVED-REPLICATED	Schoeppler, Schnitzler & Denzinger 2018, Sci Rep 8:4598, DOI 10.1038/s41598-018-22880-y	Onboard-mic replication with SD >5× larger
drift_Frest/Fref	up to 230 / 250	Hz	H. armiger, within a session	OBSERVED-REPLICATED	Schoeppler et al. 2018	as above
band_DSC	0.1–0.2% of F_ref (≈83–166 Hz at F_ref = 83 kHz)	% of F_ref	rhinolophids and P. parnellii — the band: precision within which F_echo is held around F_ref. A different quantity from offset_DSC below; an earlier version of this row printed "~200 Hz band centred ~150 Hz above resting", which split the single 150–200 Hz offset figure into a width and a centre and invented a band nobody measured	OBSERVED-REPLICATED	Schoeppler et al. 2018, Sci Rep 8:4598, DOI 10.1038/s41598-018-22880-y: "Rhinolophids and P. parnellii maintained Fecho with a high precision of only 0.1–0.2% around Fref"	Re-measure; a band >2× wider under the same paradigm
offset_DSC	~150–200	Hz (F_ref above F_rest)	R. ferrumequinum, R. euryale, P. parnellii, in flight — the offset, not the band	OBSERVED-REPLICATED	Schoeppler et al. 2018: "In flight, Rhinolophus ferrumequinum, Rhinolophus euryale, and P. parnellii accurately maintain Fref at approximately 150–200 Hz above Frest"	Re-measure; an offset outside 100–250 Hz under the same paradigm
Δf_emit	~2.4	kHz (lowering)	bat at v = 5 m/s, f_r = f_e(c+v)/(c−v), F_ref = 83 kHz	MODELED	Computed here	Arithmetic error; or a DSC bat that does not lower emission when closing
tau_CF,min	≳9	ms	derived bound: Δf ≈ 1/tau ≤ 110 Hz — soft (coherent estimation can beat 1/tau) and conditional on an unmeasured premise: that the 110 Hz emission-control SD equals the flutter-resolution scale the receiver must resolve. The premise is an inference, not a measurement, and the bound is worth exactly what the premise is worth	MODELED (premise HYPOTHESIZED)	Computed here from SD_echo	Show CF-FM flutter discrimination at that precision with tau ≪ 9 ms; or show that the required flutter-resolution scale differs from the DSC control SD — which collapses the premise and with it this row; best closure is to source real CF durations and compare rather than derive
fovea width vs DSC precision, same animal	—	—	within-species pairing	NOT-MEASURED	two species spliced here (R. ferrumequinum fovea, H. armiger DSC)	Measure fovea width and DSC precision in one species
Δr = c/(2B)	2.9 (at B = 60 kHz) / 8.6 (at B ≈ 1/tau, tau = 0.5 ms)	mm / cm	matched-filter range resolution — B = 60 kHz is an illustrative input, not a species value	MODELED (standard sonar theory; not primary-sourced)	Computed here	Cite a primary text; or a species FM bandwidth that moves the number
bat range-discrimination threshold	—	—	measured psychophysics (the "jitter" literature)	NOT-SOURCED in this pass	—	Read the jitter experiments and their replication attempts before quoting any threshold
SL_open	~130, up to and beyond 140	dB SPL re 20 µPa @ 0.1 m	open-space aerial-hawking bats	OBSERVED-REPLICATED	Surlykke & Kalko 2008, PLoS ONE 3:e2036, DOI 10.1371/journal.pone.0002036; Jakobsen et al. 2013	Calibrated on-axis recording well below 130 dB for an open-space hawker
SL_whisper	up to 110 (not ~70)	dB SPL re 20 µPa @ 0.1 m	"whispering" bats	OBSERVED-REPLICATED	Jakobsen et al. 2013	Calibrated recording capping at ~70 dB
p@140dB	200	Pa (≈0.2% of 1 atm)	p = 20e-6 × 10^(SL/20) at 0.1 m	MODELED	Computed here	Arithmetic error
t_reflex	3–4 (EMG) / 4–8 (cochlear microphonic)	ms	acoustic middle-ear-muscle reflex latency, M. lucifugus — reported by the authors in order to REJECT the reflex as the mechanism: too slow to attenuate the outgoing call. "its shortest latency in terms of electromyograms and of the attenuation of the cochlear microphonic was 3-4 and 4-8 msec, respectively, so that these muscles failed to attenuate orientation signals by the reflex"	OBSERVED-REPLICATED	Suga & Jen 1975, J Exp Biol 62(2):277–311	Re-measure reflex latency; a latency short enough for the reflex to attenuate the emission after all
MEM timing vs vocalisation	synchronous — driven by an efference copy, not by the sound and not by a pre-vocal lead	—	stapedius/tensor tympani, M. lucifugus	OBSERVED-REPLICATED	Suga & Jen 1975: the muscles "received a message from the vocalization system when the bat vocalized, and contracted synchronously with vocalization"	EMG showing muscle onset leading or lagging call onset, with a stated sign
MEM pre-vocal lead time	—	ms	onset of middle-ear-muscle contraction relative to call onset, with a sign	NOT-SOURCED in this pass	— Struck: an earlier version of this row printed "4–6 ms before vocalisation" and "8.8 ± 2.2 ms" as an unresolved discrepancy. Neither figure is in Suga & Jen 1975 (the row's only citation), the 8.8 ± 2.2 could not be sourced anywhere in this pass, and Suga & Jen report the opposite sign — "synchronously". Two unsourced numbers dressed as a live controversy is not a contested row; it is a fabrication with a hedge on it	Find an EMG study reporting MEM onset relative to call onset with a stated sign
A_MEM	17–25	dB attenuation of self-generated signal	middle-ear muscle contraction during emission	OBSERVED-REPLICATED (spread real; some sources report ~20–30)	Suga & Jen 1975; Henson 1965 (cited by, not read here)	Measure with stapedius intact vs ablated
AGC	~6	dB sensitivity drop per halving of target distance	approach phase, attributed to middle-ear muscles	OBSERVED-REPLICATED	Jakobsen et al. 2013, citing Suga & Jen 1975	Measure receiver gain vs range and find no schedule
cost of echolocation in flight	negligible at low intensity; exorbitant above ~130 dB SPL @ 0.1 m for small bats	—	Rhogeessa io (5 g); Pipistrellus nathusii	OBSERVED-CONTESTED → resolved along intensity	Speakman & Racey 1991, Nature 350:421–3, DOI 10.1038/350421a0; Voigt & Lewanzik 2012, J Comp Physiol B 182:831–40, DOI 10.1007/s00360-012-0663-x; Currie et al. 2020, Nat Ecol Evol 4(9):1174–7, DOI 10.1038/s41559-020-1249-8	Measure flight metabolic rate vs call intensity in a third species; a knee absent, or at a very different SPL
rate_pulse,flight	19.7 ± 2.7 (range 15.3–25.8)	pulses/s	Rhogeessa io, in flight (non-buzz)	OBSERVED-REPLICATED	Voigt & Lewanzik 2012	Replication outside range
St_cruise	0.2–0.4	dimensionless	flying + swimming animals at cruise, bats included	OBSERVED-REPLICATED	Taylor, Nudds & Thomas 2003, Nature 425:707–11, DOI 10.1038/nature02000	A cruising animal well outside the band under the same St = fA/U convention
St_Glossophaga	0.17–0.22 (4–6 m/s); 0.25–0.40 (3.4–4 m/s); 0.5–0.68 (<3 m/s)	dimensionless	Glossophaga soricina, wind tunnel, 1.23–7.52 m/s	OBSERVED-REPLICATED	Lindhe Norberg & Winter 2006, J Exp Biol 209(19):3887–97, DOI 10.1242/jeb.02446	High-speed replication moving the bands
wing loading / aspect ratio numeric ranges	—	N/m² / dimensionless	bats	NOT-SOURCED in this pass	Norberg & Rayner 1987 pattern is sourced; the numbers are not	Read Norberg & Rayner 1987, Phil Trans R Soc B 316:335–427, DOI 10.1098/rstb.1987.0030; state full- vs half-span convention
AR/wing-loading pattern	high AR → open air; low AR → clutter; low wing loading → slower flight	—	PCA over 200+ species	OBSERVED-REPLICATED	Norberg & Rayner 1987	Re-run on a modern phylogeny with phylogenetic correction; pattern vanishes
capture_clicking	6.8 vs 71 (silenced)	% capture success	Bertholdia trigona vs Eptesicus fuscus, field	OBSERVED-REPLICATED	Corcoran & Conner 2012, J Exp Biol 215(24):4278–87, DOI 10.1242/jeb.076943	Field replication with a defence ratio near 1
defence_ratio	10.4 (= 71/6.8)	dimensionless	as above	MODELED	Computed by Corcoran & Conner 2012 from their own rates	as above
miss_distance	~15–20	cm	jammed bats; matches ranging-interference prediction	OBSERVED-REPLICATED	Corcoran, Barber, Hristov & Conner 2011, J Exp Biol 214:2416–25, DOI 10.1242/jeb.054783	A method discriminating the three hypotheses and favouring phantom echo or masking
dose_response	77% capture at 0% duty cycle; odds −4% per +1% duty cycle	%	E. fuscus, playback	OBSERVED-REPLICATED	Fernández, Dowdy & Conner 2022, J Exp Biol 225(18):jeb244187, DOI 10.1242/jeb.244187	Replicate the dose–response and find no slope
t_jam	~2	ms window before echo arrival	click must land inside it to jam	OBSERVED-REPLICATED	Fernández et al. 2022	Vary click timing; jamming persists far outside 2 ms
d_detect,moth	20–25 (E. fuscus) vs <1 (E. maculatum)	m	average noctuoid moth's detection distance	OBSERVED-REPLICATED	Fullard & Dawson 1997	Neurophysiology giving a different threshold; a moth detecting E. maculatum at >5 m
f_click,B.trigona	up to 4,500	clicks/s	Bertholdia trigona — chain-flagged	NOT-CONFIRMED	tertiary source citing Corcoran et al. 2009; primary not read in this pass	Read Corcoran et al. 2009 (Science 325:325–7) and confirm or strike

Falsifier (operable)

The chapter's structural claim: a bat's call parameters are not fixed traits but state-dependent actions, and each one moves in the direction that keeps a specific inference well-posed — with the emission cost, the parameter change, and the uncertainty being resolved all separately measurable.

It is refuted by exhibiting an echolocating bat that hunts successfully with a call whose duration, rate, intensity and frequency do not covary with target range and closing speed — specifically: one that keeps tau at search-phase values (15–20 ms) through the terminal approach while ranging targets inside c·tau/2, or a CF-FM bat that lets its echo drift out of its measured fovea while continuing to discriminate flutter at the same precision. Every measured phase transition (search → approach → buzz), every DSC trace, and the 6 dB/halving gain schedule are the claim's exposed surface. They are all recordable with a calibrated microphone and a high-speed camera. Go and record one.

Secondary falsifiers are row-local: any number found outside its stated scope under its stated conditions moves that row only. A refuted row does not refute the chapter. A bat that does not adapt its emission does.

A cheaper falsifier, for the impatient: the buzz-ceiling result already shows the method works. The overlap hypothesis was elegant, arithmetically correct, and wrong. It lost to a work-loop assay on a muscle. If the structural claim above is doing real work, more of it should lose the same way.

Recorded INADMISSIBLE / NEGATIVE (first-class, inline)

• NEGATIVE — "pulse–echo overlap sets the terminal-buzz call-rate ceiling." Arithmetically tidy (c·PI/2 = 1.07 m at 160 calls/s), widely repeatable, refuted: Elemans et al. 2011 measured the actuator and found laryngeal motor performance binds first. Recorded because it is exactly the failure mode this wing exists to catch — a derivation that predicts the right number for the wrong reason.
• NEGATIVE / OBSERVED-CONTESTED — the naive wavelength argument, applied within the bat band. lambda = c/f explains why echolocators are ultrasonic rather than sonic. It does not survive contact with real prey: Waters, Rydell & Jones (1995) measured insect target strength as frequency-independent over 20–100 kHz, against sphere/disk model predictions. The chapter's best derivation is scope-limited and the scope is printed.
• NEGATIVE / sign reversal — "middle-ear muscle contraction begins before vocalisation." Carried in an earlier version of this chapter as the load-bearing premise of the whole reafference section, cited to Suga & Jen 1975. The cited paper says the opposite: the muscles "contracted synchronously with vocalization." Jen & Suga 1976 (Science, DOI 10.1126/science.1251206) report middle-ear muscle action potentials ~3 ms after those of the laryngeal muscles; the Frontiers 2021 review (DOI 10.3389/fevo.2021.661216) has stapedius contraction "coincident with the onset of the out-going signal." Nothing found says "before." The error's anatomy is worth recording: Suga & Jen's reflex latencies (3–4 / 4–8 ms), which the paper reports in order to reject the reflex, were re-signed into a pre-vocal lead — a refutation flipped into its opposite and then used as evidence. The forward-model reading survives on the paper's actual finding (efference copy, not reflex), which never needed the lead. Recorded because the failure was invisible: the sentence read as more precise, not less.
• NEGATIVE / fabricated precision — "8.8 ± 2.2 ms." Printed alongside "~4–6 ms" as an unresolved discrepancy "also reported in this lineage," under a row citing only Suga & Jen 1975. Neither number is in that paper, and the 8.8 ± 2.2 could not be sourced anywhere in this pass. The ± term did the damage: it reads as an instrument talking. A hedge is not a citation — "the discrepancy is NOT-RESOLVED and both are carried" dressed two unsourced values as a live controversy, which is strictly worse than printing one, because it borrows the chapter's own honesty vocabulary to launder them. Both struck; the row is now open with a falsifier.
• NEGATIVE / convention trap — mixing dB conventions. Source levels appear as rms SPL and as peak-equivalent (pe) SPL (e.g. Holderied & von Helversen 2003 report up to 133 dB pe SPL). These are not interchangeable and averaging across them is an error. Any dB figure without re 20 µPa and a reference distance and rms-vs-pe is not a number.
• NEGATIVE / convention trap — aspect ratio. Bat aspect ratio is reported under both full-span and half-span conventions. A "half-span AR of 2.75–3.75" and a "full-span AR of 5.5–7.5" can describe the same animal. Quoting one against the other manufactures a difference that is not there.
• NEGATIVE / scope trap — "echolocation is free for flying bats." True as measured (Speakman & Racey 1991; Voigt & Lewanzik 2012) and false above ~130 dB SPL @ 0.1 m for small bats (Currie et al. 2020). Quoting the 1991 title without the 2020 bound is a scope error, not a citation.
• INADMISSIBLE — "bats prove that consciousness/intention is required for intelligent behaviour." Nagel's question (what it is like to be a bat) is a real philosophical question and is not answered, addressed, or bounded by any measurement in this chapter. As a claim about the measurements, it names no observation that could refute it. Recorded, not mocked: it is a good question in the wrong ledger.
• INADMISSIBLE — "bat sonar demonstrates a natural implementation of the free-energy principle." As stated, unfalsifiable: no observation is specified that would show a bat not implementing it. The admissible version is narrow and is what this chapter claims: specific measured behaviours (DSC, call-duration collapse, the buzz, the gain schedule) are consistent with ambiguity-minimising action, and the fit has not been quantitatively tested here. The difference between those two sentences is the whole method.
• NOT-CONFIRMED (M22 chain flags): the Cloeotis percivali 212 kHz primary measurement (chain: Thiagavel 2017 → Bell & Fenton 1984, unread here); the B. trigona 4,500 clicks/s figure (tertiary → Corcoran et al. 2009, unread here); the 11–212 kHz span's attribution — Fenton et al. 1998's confirmed subject is the 20–60 kHz assemblage result, and the span was not confirmed to that paper in this pass (closure: locate the span in Fenton 1998 or in a review that carries it, and re-attribute). All printed with the chain visible. Closure: read the primaries.
• NOT-SOURCED in this pass: numeric wing-loading and aspect-ratio ranges; species FM sweep bandwidths; CF component durations; a primary reference for c in air; bat range-discrimination ("jitter") thresholds; the Corcoran et al. 2009 in-paper capture statistics (the 2012 field figures are used instead, and are sourced); any pre-vocal lead time of middle-ear-muscle onset relative to call onset (the previously printed "~4–6 ms" and "8.8 ± 2.2 ms" are struck — see the NEGATIVE ledger above); whether the anticipatory reading of the 6 dB/halving gain schedule can be discriminated from an echo-amplitude-driven one (the schedule itself is sourced to Jakobsen et al. 2013; only the anticipation is unsourced).
• NOT-MEASURED: atmospheric absorption at 212 kHz; within-species pairing of fovea width and DSC precision; any fitted coupled-agent free-energy model of the bat–moth arms race.

HONEST FENCE — MODELED

This chapter is fenced MODELED. Most individual rows are OBSERVED-REPLICATED; several are OBSERVED-CONTESTED, NOT-SOURCED, or NOT-MEASURED, and carry their own class. But the chapter as an artifact composes measured constants through stated assumptions — c = 343 m/s at 20 °C (echolocation happens at other temperatures and humidities, and every derived metre moves with c); a stationary reflector in the Doppler algebra; free-field spherical spreading; matched-filter resolution bounds that assume a receiver the bat may or may not implement. The assumptions are the fence. Change c by 5% and every range in the chapter moves by 5%; the conclusions survive because they turn on the structure — that resolution trades against range, that duration trades against overlap — not on the prefactors.

The active-inference reading is HYPOTHESIZED throughout and is a lens, not a finding. No G(pi) has been computed for any bat here. No policy space has been enumerated, no preference distribution C specified, no gamma fitted. The claim is the weaker and honest one: these measured behaviours have the shape the epistemic and ambiguity terms describe, and the shape is close enough to be worth testing. The measurements are the asset and they stand alone. If the lens were withdrawn tomorrow, DSC would still hold the echo to 110 Hz.

Per Gould & Lewontin (1979), "The Spandrels of San Marco and the Panglossian Paradigm": none of this establishes that any bat trait is an optimum. Phylogenetic inertia, drift, developmental constraint and frozen accidents produce traits that solve nothing, and the bat literature has its own candidates — a larynx that had to be repurposed, a cochlea that cannot move its fovea and forces the animal to compensate with motor output instead. Nature's authority here is precise and limited: it already ran the search under real physical constraints and deleted the failures. Convergence — bats and dolphins arriving independently at broadband clicks and range-from-delay — is evidence of a constraint-optimum. It is a hypothesis generator, never a proof. Per repo rule M7, any sonar or active-sensing design taken from this chapter must beat a tuned conventional baseline on a pre-registered metric, with a discriminator that collapses the claimed gain, or it is recorded NEGATIVE. "Bats do it this way" is not an argument. It is a place to start looking.

Not claimed

• Not claimed: that a bat is conscious, aware, or has experience. This chapter takes no position on Nagel's question and produces no evidence bearing on it. It is a permanent OPEN QUESTION, in a different ledger.
• Not claimed: that a bat computes expected free energy. Nothing here shows a bat evaluating G(pi) over a policy space, or anything isomorphic to it. The bat emits, listens, and adjusts. "Minimising the ambiguity term" is our description of its behaviour, in our vocabulary, for our purposes. The bat is not doing our arithmetic. Anyone who reads this chapter as evidence that active inference is implemented in a bat brain has crossed the exact lane the chapter was written to hold.
• Not claimed: that the active-inference reading is required to explain any measurement here. Classical sensorimotor control, signal-detection theory and plain optimal-foraging accounts predict much of it. No discriminating experiment separating those accounts from the active-inference account is offered, and none is known to me in this pass. Until one exists, the lens earns nothing it has not paid for.
• Not claimed: that DSC is gamma (policy precision, NA-02). It is an action that lowers the ambiguity term by moving sensory input into a high-precision likelihood band. The two precisions are different objects with different units. Conflating them is a category error and this chapter refuses it.
• Not claimed: any bat range-discrimination threshold. The psychophysics is contested and NOT-SOURCED here.
• Not claimed: any numeric bat wing loading or aspect ratio. NOT-SOURCED here; the pattern is sourced, the numbers are not, and the convention trap is live.
• Not claimed: that the bat–moth arms race has been modelled as coupled surprise minimisation. That reading is HYPOTHESIZED and unfitted.
• Not claimed: that any citation above raises any UNI rung. A nature citation is NEVER a UNI gate. Reading Elemans et al. 2011 does not make any UNI claim proven, designed, or built. The NATURA vocabulary (OBSERVED-REPLICATED / OBSERVED-CONTESTED / MODELED / HYPOTHESIZED / INADMISSIBLE / NOT-MEASURED) and the UNI ledger vocabulary (proven / designed / hypothesized / not-yet-built) describe different kinds of claim and never merge. This chapter contains zero UNI claims.
• QUAESTIO-APERTA: "full human" and "beyond human" appear nowhere here as a target, milestone, or deliverable. They are permanent open questions. A bat has been running measurable active sensing for tens of millions of years and it has no bearing on them whatsoever.