Fedderson-Daedalus: Arbiter (Step 3)

Source gap (repeated, updated after QE). Both PDFs are Google Books scans; Step 0 has since produced OCR text for 405 of 406 and 26 of 27 pages, and the tables were checked against page images. Earlier [read] tags in this file come from rendered page images; see the QE revision at the end, which supersedes anything it contradicts. "Daedalus-Fedderson-109.pdf" is NOT Feddersen's work and NOT the journal Daedalus: it is the Proceedings of the American Academy of Arts and Sciences vol. 25 (1890), 406 pp., of which only the contents, Trowbridge and Sabine pp. 109-123 and Trowbridge pp. 192-194 matter here. Feddersen's own 1857-1862 papers (Poggendorff's Annalen) are not in the folder. Claim 1's attribution to Feddersen rests on the 1890 authors' citations and on [memory]. Rich (1917) is the only experimental paper with data in the folder. Trowbridge-Sabine journal pp. 109-123 now have two independent full reads (mathematician; QE deep read).

Tags: [read] = from the document (page image); [web] = URL; [memory] = recalled, with confidence.

How the arbiter worked

• I re-read Rich pp. 151-154 as page images to settle the reviewers' conflict over the Cu-Zn table (below), and recomputed the statistics and the energy budget.
• Web budget: web search was exhausted (quota). Two WebFetch attempts (Wikipedia "Oscillating circuit" and "Spark-gap transmitter") returned nothing on Kelvin/Feddersen. So the check on "Kelvin predicted it in 1853" failed; it stays [memory, high] and no verdict depends on it. The Feddersen date (1857-1862 vs 1859) stays unresolved: [web] https://en.wikipedia.org/wiki/Berend_Wilhelm_Feddersen (cited by two reviewers, one via search summary) gives 1859 for the demonstration; the 1890 paper's footnote cites his Annalen papers 1858-1862 [read: p.109]. No verdict depends on it. Work-function values are [web] https://en.wikipedia.org/wiki/Work_function (as cited by all three reviewers; I did not re-fetch them). Spark-resistance and ion-mobility statements remain [memory, medium], unchecked.
• "No replication of Rich found" means not found, not searched exhaustively (search quota gone). It must not be read as "failed to replicate".

Disagreements resolved

1. Cu-Zn table (physicist vs mathematician and contrarian). I read p.154. The headings are Cu-Zn+ : 32 half-cycles, 14 trains, 1.27 (footnote: "11 half-cycles only"), 21 half-cycles with no train; Cu+Zn- : 30 half-cycles, 143 trains, 4.77 [read: Rich p.154]. The physicist is right. The mathematician and contrarian transposed the headings, so their finding that "Cu-Zn contradicts the series, copper negative favoured" is wrong. With Zn negative (Cu+Zn-) the spark starts much more readily, which agrees with Zn above Cu. Per half-cycle the weak column is 14/32 = 0.44, not 1.27, so the printed asymmetry understates the true one (the denominator is inconsistent with the other tables).
2. Series consistency (mathematician and contrarian vs physicist). Reading headings only: Fe-Bi+ (Fe negative) 5.21 vs 2.86 [p.152]; Cu+Bi- (Bi negative) 2.16 vs 1.06 [p.152]; Fe-Zn+ (Fe negative) 3.1 vs 1.41 [p.153]; Cu+Zn- (Zn negative) 4.77 vs 0.44-1.27 [p.154]; Cu+Fe- (Fe negative) 6.83 vs 3.16 [p.148, per reviewers]. In every table the favoured half-cycle is the one where the metal ranked higher in the series is NEGATIVE. These give Fe > Bi, Fe > Zn, Fe > Cu, Bi > Cu, Zn > Cu, all consistent with Fe, Bi, Zn, Cu. The only missing link is Bi vs Zn: Rich says "all possible combinations" were used [p.151] but shows no Bi-Zn table, so Bi above Zn is unsupported in the pages. So the claim that the tables contradict the series is wrong.
3. The real internal inconsistency is wording, not numbers. Rich's prose says "iron anode", "bismuth anode", "zinc anode" for the favoured electrode [pp.152-154], but the headings put these metals negative. On p.152 he also writes "always assuming that the discharge is electronic in nature, so that the current flows from anode to cathode" [read], and his summary speaks of electron emission from iron. Electrons leave the negative electrode, so "anode" is a sign slip or a loose use of the word for "emitting electrode". Tables are the reliable part (headings agree with each other and with electron emission by the cathode). I rate the series' wording Internally inconsistent, the data internally consistent.
4. Cu-Fe average. Printed 6.2; 205/30 = 6.83 [my calculation; two reviewers transcribed the totals]. The asymmetry is larger than printed (6.83 vs 3.16).
5. Statistics (my recomputation, Poisson z for difference in train rate per half-cycle). Fe-Bi 4.4, Cu-Bi 3.4, Fe-Zn 4.4, Cu-Zn 10.4, Cu-Fe 6.4, Cu-Cu (control) -0.6. This matches the physicist. The mathematician's Welch t = 9.8 for Cu-Fe (from per-half-cycle counts) is larger because it uses actual variance. Both caveats apply: half-cycles within one film are not independent, one "typical" film per pair is shown, no blinding, no electrode-surface control.
6. Claim 3 verdict. Physicist and mathematician: Established; contrarian: Plausible, untested. Resolution: two different things. The statement "atoms are not carried across the gap in step with the oscillation" follows from the energy budget: crossing 6 mm in a half period of 1.5e-7 s needs v = 4e4 m/s, i.e. 0.5 m v^2 = 464 eV for a 56 u iron atom [my calculation; reproduces the mathematician]. Collisions at atmospheric pressure limit drift to far less [memory, medium]. So the hypothesis Trowbridge tested is Refuted by calculation and the null is expected. Trowbridge's experiment itself is one unquantified, unreplicated photograph pair and adds little; his stated sensitivity ("1/100 of a wave length") cannot be read reliably and, taken literally, would be too coarse to see a ~0.5 Angstrom shift ("a tenth meter" = 1 Angstrom, old unit) [read p.193 per three reviewers, uncertain legibility]. The contrarian's "Plausible, untested" understates the calculation.
7. Claim 2. Physicist: deviation Established; mathematician: Plausible, untested. Resolution: the plain observation (periods of 2.9-3.4 microseconds against a computed 3.04; I recomputed 2 pi sqrt(39700 x 5317.9)/(3e10) = 3.04 microseconds with the factor 1/v the mathematician identified [read: pp.110, 115]) is within about 10% and is expected from spark resistance and stray elements. A 5% lengthening needs about 50 ohm series resistance, plausible for a young spark channel [mathematician's calculation; spark-channel resistance [memory, medium], unchecked]. [SUPERSEDED, see QE revision items 2-3: steady resistance is not the explanation, and most of the mean deviation disappears when the listed wire capacity is included.] The authors rejected the spark-resistance alternative with one glass-jar test and no equations [read pp.120, 122 per mathematician], so it was not excluded. Verdict on the observation: Plausible, untested as a precise claim (no error bars, no neutral replication); mainstream explanation is the likelier one.
8. Claims 4 and 5 verdict label. Physicist: Plausible, untested; mathematician and contrarian: Never properly tested. The scale distinguishes "no real test yet" (consistent with known physics) from "a testable claim with no neutral, controlled test on record". Rich's work IS a test with controls (Cu-Cu symmetric 2.03 vs 2.24 trains, 3.97 vs 3.93 sparks per train [p.149]; DC make/break polarity check [pp.150-151]; two years' work), but it is a single-lab dissertation, polarity assigned by hand, no neutral repeat. The better fit is Never properly tested, with the note that it was not dismissed; it was simply not followed up in anything found.
9. Physicist's "Refuted by calculation" for the work-function reading of claim 5. Rejected. The work function does not predict the order (Fe 4.67-4.81, Bi 4.31, Zn 3.63-4.9, Cu 4.53-5.10 eV [web] https://en.wikipedia.org/wiki/Work_function), so Rich's interpretation is in tension with data, but breakdown in air at 0.1-4 mm gaps depends on secondary emission, cathode-spot formation, oxide films and vapour supply, not on clean-metal work function alone ([web] https://en.wikipedia.org/wiki/Paschen%27s_law as cited by the physicist; spot-formation mechanism [memory, medium]). An energy budget it is not. Verdict: "not supported by tabulated work functions; mechanism open".

Tested / never tested / dismissed (kept visible)

Final verdicts and reasoning

Claim 1. Condenser discharge is oscillatory, T = 2 pi sqrt(LC), seen in a rotating mirror

All three agree. Check: L = 25 mH, C = 0.125 microfarad gives T = 2 pi sqrt(0.025 x 1.25e-7) = 0.351 ms [my calculation]; in the 1890 CGS circuit the paper's 3.04 microsecond theoretical period is reproduced. Needs R < 2 sqrt(L/C) (damping ratio for the 1890 wire resistance about 0.009). Settling experiment: none needed. Verdict: Established. (QE addition: Rich's own film gives a spark spacing implying an effective C of about 0.039 microfarad against the 0.125-0.5 microfarad he quotes at low frequency; the formula is fine, the capacitance is rate-dependent. Verdict unchanged. Attribution to Feddersen specifically and the date: [memory, high]/[web] partial; not material.)

Claim 2. Periods deviate from the LC formula; cause "dielectric viscosity" of air

Split (revised after QE, see below). (a) Mean period deviates from LC theory: Tested, not supported by the authors' own data (right-terminal mean 3.11 us against 3.10 us once the listed wire capacity of 200 cm is added; left 3.17, +2%). (b) A systematic "periodicity" by oscillation number: Tested, not supported by reanalysis of their published tables (ANOVA/permutation p = 0.53 right, 0.10 left); interval-to-interval scatter of 4-6% is real but is not shown to be anything but spark behaviour and measurement. (c) Explanation "dielectric viscosity / hysteresis of air": Historically superseded but instructive (the real effect, slow recovery of a solid dielectric, is dielectric absorption in glass; their own cited Boltzmann-Klemencic result concerns polar molecules, not air). (d) Magnetic-field-in-glass remark: Never properly tested.

Claim 3. Atoms are not carried to and fro; iron lines show no Doppler shift

Split. The physical statement that atoms are not carried across in step with the oscillation: Established (the opposite hypothesis is Refuted by calculation: 464 eV needed; J. J. Thomson had published this energy argument in 1890 and Trowbridge cites it as his motivation). Trowbridge's experiment as evidence: weak, one photograph pair, unreadable sensitivity, unreplicated.

Claim 4. Unlike-metal spark gap starts more readily on one polarity (partial rectification)

Never properly tested (single-lab, controlled by a Cu-Cu null and DC polarity check, asymmetry 2:1 to 6.8:1 in trains per half-cycle, z = 3.4-10; not replicated on record; not dismissed). Physical content is ordinary electrode and cathode physics, so nothing mainstream would change. The odd/even evidence is Not testable as stated (Rich himself says no reliance can be placed on it).

Claim 5. Order Fe, Bi, Zn, Cu as ease of electron emission; material not negligible

Split. "Electrode material is not negligible": Never properly tested (as 4). The empirical order by negative-electrode headings is consistent across five tables (Bi vs Zn not shown): Never properly tested. Rich's prose wording ("anode" for the favoured, negative electrode in several passages) is Internally inconsistent with his headings and his closing "electrons are emitted from iron", although the tables themselves are consistent; the tables themselves are consistent and the earlier reviewer charge that Cu-Zn reverses the series is retracted. Electron-emission reading as the mechanism: Plausible, untested (ordinary cathode physics, but the tabulated work functions do not reproduce the order; not Refuted by calculation). Rich's wording: mixed usage (see QE revision item 6).

Corrected claim list

1. A condenser discharge through a spark gap is oscillatory with T = 2 pi sqrt(LC) in the low-resistance limit (T = (2 pi / v) sqrt(LC) in mixed ESU/EMU units, v = 3e8 m/s); photographed by Feddersen in a rotating mirror (Feddersen date 1857-1862, 1859 per one web source).
2. In 1890 Trowbridge and Sabine measured successive periods of 2.6-3.6 microseconds (small air condenser) against a computed 3.04 microseconds and attributed a "periodicity" to "dielectric viscosity" of air. Their own tables show the mean matches theory (3.11 against 3.10 microseconds) once their listed wire capacity (200 cm) is included, the scatter (4-6%) is real but no repeatable pattern by oscillation number is shown, and the resistance alternative was not excluded by their glass-jar test (which actually shows +2.2%, 0.4 against 1.3 cm). Round 2 caveat: the quoted 5317.9 cm does not follow from the printed geometry (height is 30.47 cm, not 80.47; 18 gaps in parallel give about 4,540 cm, 17% lower, which would put theory at 2.87 microseconds and the measured excess at 8-10%), so the capacity itself is uncertain by about 15% and the data cannot settle a few-percent departure either way. Verdict: Tested, not supported on the authors' own stated inputs, with that caveat; explanation historically superseded.
3. Metal atoms do not cross the gap in step with the oscillation (it would need about 460 eV per iron atom; J. J. Thomson's 1890 energy argument). Trowbridge's single Doppler photograph pair (double oscillation 3e-7 s, gap 6 mm, 20,000-line grating) is consistent with that; the sensitivity prints as 1/100 of a wave length (about 85% confidence, Round 2 image read) and its unit is ambiguous: taken as a fraction of the light's wavelength it is too coarse to see the ~0.5 Angstrom shift, taken as a fraction of a line width it would be fine enough; the paper does not say which, so the experiment adds little. Verdict: Established / opposite hypothesis Refuted by calculation.
4. In Rich's films (gap length not stated for any table; Cu-Cu control run at different damping), trains per half-cycle depend on polarity for unlike-metal pairs (Fe-Bi 5.21 vs 2.86; Cu-Bi 2.16 vs 1.06; Fe-Zn 3.1 vs 1.41; Cu-Zn 4.77 vs 0.44 per half-cycle (1.27 as printed, 11 half-cycles only); Cu-Fe 6.83 vs 3.16 (printed 6.2 is a slip)), with Cu-Cu symmetric.
5. The favoured half-cycle is the one where the higher-ranked metal is the negative electrode, giving the series Fe, Bi, Zn, Cu (Bi-Zn link unshown; no Bi-Zn data anywhere in the paper). In some passages Rich's prose calls the favoured electrode the "anode", which conflicts with his headings and his closing statement about electron emission. The work-function explanation is not supported by tabulated values (verdict: Plausible, untested as a mechanism).

What would settle it

• Claims 4 and 5: a one-day replication with a fast camera or a photodiode pair and a signal generator or full-wave HV supply instead of a leaky transformer; cleaned, polished and conditioned electrodes of Fe, Bi, Zn, Cu at fixed gap (0.1-4 mm); blinded polarity assignment; many films with film-to-film variance; include Bi-Zn and Cu-Cu controls and oxide/vacuum or inert-gas variants. Record the gap length for every film (Rich gives only a blanket 0.1-4 mm range, so gap is an uncontrolled variable and the Cu-Cu control, at about 4 sparks/train, was evidently not run under the CuFe conditions). If the asymmetry survives in inert gas, it is cathode emission; if it vanishes, surface films were the cause.
• Claim 2: a measured series resistance of the spark channel versus time (voltage-current oscilloscope trace of the same circuit); if R(t) predicts period lengthening of 5-10%, "dielectric viscosity" is unnecessary. Capacitor loss-tangent measurement of the air capacitor alone would rule hysteresis in or out.
• Claim 3: the energy and collision-limited drift calculation already settles it; a modern spectrum with measured Stark and thermal width would show the 0.5 Angstrom coherent shift is far below line width.
• Claim 1: nothing needed.

Open questions

• Was Rich's result ever replicated, cited or refuted? Not found; search was incomplete (quota).
• Why does iron rank above bismuth and zinc if work function does not explain it? Candidates (my suggestion, not tested): oxide films, vapour supply and cathode-spot formation, thermal properties.
• What is the Bi-Zn relation? No data in the pages read.
• Trowbridge's sensitivity fraction reads 1/100 of a wave length (glyph small, about 85% confidence after Round 2); the intended reference unit is unclear, so whether the Doppler test could detect 0.5 Angstrom remains open.
• (Closed by QE) Trowbridge-Sabine period data now have a second independent reading.
• (Round 2) Why the quoted small-condenser capacity (5317.9 cm) is 17% above the geometry as printed; a zinc thickness near 0.75 mm would reconcile it but the paper gives no value.
• Feddersen's own papers are needed to check the claims in the folder-name's namesake; none exist in the folder.

QE revision (after Step 4b, verdict MINOR)

Inputs: qe.md; two new Opus/Sonnet deep reads in deep/ (Trowbridge and Sabine PDF pp.145-159, second independent read; Rich PDF pp.6-24, tables against page images); and my own page-image read of Trowbridge, "Motion of Atoms", journal p.193 (PDF p.229) [read: page image PDF p.229]. At the time of this first revision, deep reads for Dolbear, Nott, the Joule memorial and Safford-Hofman had NOT been produced; item 10 below rests on the QE's OCR quotes only and is revised by Round 2 below. No web checks were run in this revision; the medium-confidence [memory] items listed in qe.md section 5 (air loss at MHz, ion mobility, spark-channel behaviour, Rich citations/replications) remain unchecked, and "no replication of Rich found" still means not searched.

New evidence

1. Trowbridge and Sabine data [read: deep read]. Wire capacity. Their small-condenser circuit lists "Capacity of wire, 200" beside C = 5317.9 cm; the quoted theoretical 3.04 microseconds uses 5317.9 only. With 5517.9 the theory is 3.10 microseconds, against measured full-period means of 3.11 (right terminal) and 3.17 (left, +2%). Their headline "deviation" is mostly an unapplied correction.
2. Scatter is real, pattern is not. Single-reading repeatability is 0.063 microseconds; within-negative scatter is 0.13 (right) and 0.19 (left), 2-3 times larger. But ANOVA and permutation tests by oscillation number give p = 0.53 (right), 0.10 (left); the wave seen in the mean curves comes from the first ordinate. Successive deviations are anticorrelated (lag-1 about -0.3 to -0.4), which is what setting error produces and contradicts their remark that this is "conspicuously not the case".
3. Resistance test. Glass jar, spark 0.4 cm against 1.3 cm (QE's 1.8 cm and 4.5-fold were misreadings; ratio 3.25): full-period means 3.33 and 3.41 microseconds, +2.2%, Welch t = 2.3, p about 0.04 (one negative per length, fragile). So the test does not show "no variation", and it was run on glass, not on the air condenser. Linear damping alone cannot be the mechanism: sqrt(L/C) = 82 ohm for the small circuit, so 10 ohm shifts the period 0.19% and about 32 ohm is needed for 2%. The primer's point also stands: a steady 50 ohm would damp the oscillation to about 1e-7 in 8 periods, inconsistent with the 8-9 periods photographed. The earlier arbiter item 7 ("50 ohm plausible") is withdrawn; the realistic candidate is the spark as a nonlinear, time-varying element, which is standard spark-gap physics [memory, medium] and untested in the paper.
4. Errata to qe.md. Chronograph resolution is 1/500 s (not 1/315). The "large against small condenser" comparison has no data in the paper (no capacity, circuit or period table for the large condenser; conclusion 3 is assertion). The Boltzmann and Klemencic passage is prose with no formula and no numbers; the "marked difference" is for polar molecules (SO2, ethyl bromide), which modern theory explains by orientational polarisation that cannot follow high frequency [memory, high]; it says nothing about air. The authors' own source thus points against their extension to air. The paper has internal slips (right neg 1 interval 6 printed 2.91 and 3.00; two printed column means off; coil radii inverted). An unresolved side check: the small condenser's stated geometry gives about 11,900 cm if 18 gaps act in parallel, against the quoted 5317.9 cm.
5. Magnetic-field experiment. Glass jar 2.5 cm by 28 cm in a 728-turn coil at about 30 A, field about 0.037 T [calc]; no numbers given; the text says some runs "seemed to show a decrease in the total number" of oscillations while conclusion 4 says "apparently unaffected". Qualitative null only.
6. Rich, "anode" wording [read: deep read, 13 passages]. The headings (items 3, 5, 7) and the closing summary ("electrons are emitted from iron") make the favoured electrode the NEGATIVE one. The prose uses "anode" for that electrode in several passages (Cu-Bi "bismuth anode", Fe-Zn "iron anode", Cu-Zn "zinc anode", Guthe comparison), while elsewhere it says "iron is negative". The usage is mixed, not uniformly wrong; the earlier statement "Rich's prose calls the favoured electrode the anode" is softened to "in some passages". Tables remain the reliable part.
7. Rich, period check [read: p.14; calc]. L = 25 mH, C = 0.125-0.5 microfarad (low frequency, "absorption was very bad", depending on charge duration), film 42,000 mm/s, observed spark spacing 4.1 mm = 97.6 microseconds. Sparks alternate electrodes, so spacing is a half-period: T = 195 microseconds, effective C = T^2/(4 pi^2 L) = 0.039 microfarad (0.0097 if a full period). The naive LC prediction would give 7.4-14.75 mm spacing. Probable (interpretation, not in the paper) causes: rate-dependent glass capacity and iron-wire coil losses. The formula survives; the capacitance does not stay at its slow-method value. This is a measured instance of the very effect Trowbridge and Sabine invoked, but for a solid dielectric.
8. Rich, gap length and Bi-Zn. No table states a gap length; the only statement is a blanket 0.1 mm to 3-4 mm range. Train lengths differ between tables (Cu-Cu mostly 4 sparks per train; Cu-Fe 2-3; Fe-Zn Fe-negative mostly 1) which suggests differing conditions, so the Cu-Cu control is not matched. No Bi-Zn data exist anywhere in pp.6-24; Bi above Zn rests on Guthe's coherer series. If anything Rich's numbers put Zn level with or above Bi. Cu-Zn: 14 trains in 32 half-cycles (0.44 per half-cycle) against 143 in 30 (4.77), about 10.9-fold; printed 1.27 is per active half-cycle (footnote). Other printed slips: Cu-Cu even-side sparks 292 by rows against 291 printed; Cu-Fe 6.2 should be 6.83.
9. Trowbridge Doppler paper [read: page image PDF p.229]. Confirmed: double oscillation t = .0000003 s (3e-7, as the reviewers gave) for one circuit and .0000024 s for the other; 6 mm gap; 20,000 lines/inch Rowland grating; broadening "might amount to a whole tenth meter" (1 Angstrom); one photograph with terminals perpendicular to the slit and a comparison with terminals parallel. The sensitivity prints as a small fraction that I read as 1/100 of a wave length (glyph small; medium confidence). A half-period of 1.5e-7 s across 6 mm needs 4e4 m/s, matching the 464 eV figure. The paper cites an energy argument (J. J. Thomson, 1890) as motivation, so the decisive calculation was not original to this project's reviewers.
10. Items from digests/OCR only (no deep read made; QE quotes) [read: OCR via qe.md]. Dolbear (PDF pp.309-315; digest wrongly says "Dodge"): flexing magnetised iron rings induces currents; "magnetic sound"; speculation on ring-shaped magnetic atoms, ether waves outside a closed ring magnet and a "Vortex Ring theory of atoms". Induced currents on flexing are ordinarily explained by magnetostriction (Villari effect) and flux change [memory, medium]; the ether and vortex-atom statements are speculation without a measurement. Not a project claim; assessed as Not testable as stated (the atom model) and Plausible but ordinary (the induced currents). Nott (pp.281-296): condenser capacity varies with charging time; at 0.002 s "the discharge exceeded the charge, but the results were too irregular", a classic dielectric-absorption artefact, not a gain. Safford and Hofman (pp.15-18): capacity "varied with rate". Joule memorial (p.387): Royal Society members called his paper "nonsense from beginning to end" and refused to publish it in their Transactions; Thomson the only friendly voice. This is a documented case of resistance to a correct result, useful as a counterweight, though it is not about the project's claims. The Rumford Fund grant to Trowbridge (PDF pp.335-344) answers the "was funding blocked" question: no suppression is documented in these files.

Changed verdicts and labels

Claims.md was already updated for these items; the Corrected claim list above has been edited to match.

Round 2 (second-round deep reads, appended to deep/Daedalus-Fedderson-109-c85345.md)

Inputs: five new sections in the Trowbridge-Sabine deep-read file (Doppler page, Joule memorial, Dolbear paper, small-condenser geometry, Nott paper). The brief named five reads under one slug; the folder holds them as sections of that one file. A read of Safford and Hofman is not among them and remains undone. I did not re-open the page images myself; the points below are as reported by the deep reader [read: deep read], and no web checks were run.

What the new reads found

1. Small-condenser geometry (changes the strength of the Claim 2a basis). The height is printed 30.47 cm, not 80.47 (OCR and digest misread 3 as 8) [read: deep read of page image p.150]. The "about 11,900 cm" side check in item 4 above is therefore withdrawn. With 30.47 cm and 18 gaps in parallel the formula the authors give (C = l / (2 ln(b/a)) per gap) yields about 4,511 cm bare, 4,544 cm with the Maxwell end strip, against the quoted 5317.9: 15-17% low, not 2.2 times high. Fewer active gaps make it worse. A zinc thickness of about 0.075 cm would reconcile it; the paper prints the symbol beta but no value, so this is an inference only. 5317.9 is confirmed as the value actually used (it reproduces the printed 3.043 and 3.096 microsecond periods). Consequence: with the geometric 4,544 plus 200 cm wire, theory would be 2.87 microseconds and the measured 3.1-3.2 would exceed it by 8-10%. So the Round 1 finding that the deviation "disappears" with wire capacity holds only if the authors' quoted 5317.9 is right, and that cannot be checked from the printed dimensions better than about 15%. The 1890 data can neither establish nor exclude a few-percent departure from LC. The mirror radius is 313 cm, not 813 (optics confirm).
2. Doppler page (Trowbridge, journal p.193). Confirms t = 3.0e-7 s and 2.4e-6 s (OCR's 8e-7 is wrong), 6 mm gap, 20,000 lines per inch grating, "a whole tenth meter" (1 Angstrom), and a visual null on one photograph pair with no repeats [read: deep read of page image p.229]. Sensitivity "1/100 of a wave length", about 85% confidence (a printed 1/1000 is not fully excluded). The deep reader stresses that the unit is ambiguous (fraction of wavelength versus of a line width); the paper gives no calculation, L, C, wavelength or exposure data. His half-period arithmetic (4e4 m/s, 0.5-0.6 Angstrom for the shorter period, about 0.07 Angstrom for the 2.4 microsecond period) matches mine. Trowbridge himself says convection "would take place too slowly to give a spectroscopic effect". The test addresses bulk atom convection, not the electron-carrier question.
3. Nott (condenser comparison, journal pp.244-258; PDF pp.282-296). A bench comparison of solid dielectrics (paraffined paper, hard rubber, mica); no air, no megahertz, and the word "absorption" never appears. Residuals of 2-27% of the full charge come back in instalments over 1-28 minute intervals and rise again after waiting (No. 72: 0.1 at 2-minute intervals, then 1.2, 1.5, 3.0 after 6, 22, 28 minutes), the standard signature of dielectric absorption [memory, high], which Nott records without naming. Capacity rises with charging time by 16-85% for leaky units and 1.6-5.5% for good ones between 1 ms and 30 s. The "discharge exceeded the charge" sentence at 0.002 s is quoted exactly but has no numbers and Nott calls it too irregular for any conclusion; the one footnoted "apparent gain" (No. 9, 0.5%) is within scatter. Corrections to my item 10: the residual trains are minutes, not hundreds of minutes (the long times belong to the leakage table); the best condenser is No. 3, not "No. 8" (OCR); the "absorption artefact" is the reviewers' interpretation, not Nott's, and a charge-through-galvanometer versus discharge-throw instrument offset is not excluded; the paper bears on glass or paraffine jars, never on air.
4. Dolbear (journal pp.271-277; PDF pp.309-315). Fully read. It is a magnetic and acoustic paper (flexed iron rings, galvanometer, telephone, a gas-pipe compass test); nothing wireless, electrostatic or about speed or Feddersen. Quantities: 1,300 ampere-turns (6 V over 4.6 ohm, about 1,000 turns, his estimate), 30-40 degree deflections, a one-yard range. No unmagnetised-ring control, and Earth's field matters by his own Exp. XI. His conclusion 3 (sympathetic and forced vibration between magnets) is not separated from mechanical or airborne coupling. The ring-atom, vortex and "ether waves outside a closed ring" ideas are explicitly suggestions, with no proposed test. Page range is 309-315 (p.316 is Davenport), and the paper was presented 14 January 1891. My item 10 labels (induced currents Plausible but ordinary; atom model Not testable as stated; not a project claim) stand.
5. Joule memorial (journal pp.846-851; PDF pp.384-389). Read in full; no author named. The Royal Society quotation ("nonsense from beginning to end"), Thomson as the only friendly voice and the 1850 publication are accurate against the text [read: deep read]. New points: the memorial dates the Cork paper 21 August 1848 yet says publication in 1850 came "seven years after", which fits 1843 and not 1848 (the date inconsistency is the memorial's own; the web check is still not done); the 1849 British Association committee refused to print "Friction consists in the conversion of mechanical power into heat"; the memorial blends a refusal with a delay and cites no source for the quotations beyond Roscoe, 40-47 years after the events, so it is a secondary report. Values: 838 (1843), 781.8 (1847), 772 (1849), 772.55 (1878). It also conflates the equivalent J with the later unit joule. It names none of Mayer, Colding, Helmholtz or Carnot. None of this touches a project claim; it stays a documented case of resistance, with the source caveat.

Verdicts changed by Round 2. No label on the verdict scale changes. One basis changes and one statement is withdrawn:

Why 2a keeps its label: the scale offers no better fit, and the test the authors ran, measured period against theory from their own stated inputs, did not support a deviation. The honest reading is that the experiment was not accurate enough to confirm or deny one.

Corrected claim list: items 2 and 3 above have been edited to carry the capacity caveat and the sensitivity ambiguity. Claim list items 1, 4 and 5 are unchanged.

Remaining

Not done: web checks listed in qe.md (including the Cork 1843 versus 1848 date of Joule's paper); a deep read of Safford and Hofman (still unread beyond digest and QE quotes); Feddersen's own papers. Done in Round 2: deep reads of Dolbear, Nott, the Joule memorial, the Doppler page and the small-condenser geometry.

Files: review/Fedderson-Daedalus/arbiter.md