Goodenough-JB: Quality Engineering (Step 4b)

Inputs checked: _text/manifest.json, digest/ (1 file, all read), facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py. Spot checks were made against _text/Alt-Strat-for-Safe-Rech-Battery-d78469.txt, and pages 3, 5, 6 and 7 of the source PDF were rendered and read as images by QE [read: page image p.3, p.5, p.6, p.7]. No deep/ folder exists.

Source gap (repeated from Step 1). Only one document is in the folder: Braga, Grundish, Murchison, Goodenough, Energy Environ. Sci. (2016), doi:10.1039/c6ee02888h. The cited works that carry the evidence for the glass (Braga 2014 J. Mater. Chem. A 2, 5470; Braga 2016 EES 9, 948; Braga 2017 JES 164, A1) were not provided. All later files repeat this note correctly.

Verdict: MINOR

The central verdicts (claims 3 and 4 refuted by calculation; the excess capacity and cycling never properly tested; the 10.5 Wh/g figure internally inconsistent) rest on the paper's own text and on Fig. 1, which reviewers rendered and QE re-checked. These verdicts hold. But no reviewer rendered pages 4 to 7. Those pages hold Figs. 2, 3, 5, 6, 7 and 9, and several statements in the arbiter and primer are wrong or incomplete because of it. The most important is that the arbiter and primer say Fig. 9 gives "no current density". In fact the figure shows ±3 mA/cm² for about 200 h. Fixing these points changes some wording, one open question and possibly the weak-form verdict on claim 2. It does not change the central verdicts.

1. Coverage

• The manifest lists one document and none was skipped. Coverage is complete.
• Step 1 predates digests. facts.json's how_read says "full text", which is honest: its sources field says the figures were not rendered. The value is not in the prescribed vocabulary ("digest + deep read (pages)" or "page images"), so it should be updated to state what was actually done.

2. Shortcuts

• Acceptable: the document is only 7 pages, and every reviewer read its full text layer. A separate deep read of the prose was not needed.
• Defect (figure data never read): this paper is figure-heavy and two-column. pdftotext loses the figure contents and also the minus signs in the text. Only Fig. 1 (p.3) was rendered. The verdicts on claims 1, 2 and 5 include absence statements ("no current density", "ferrocene voltage not stated", "no retention data", "1.65 V vs 2.61 V unexplained") that the figures on p.5 to p.7 partly contradict or answer. See section 3.
• Defect (Fig. 2, p.4, never viewed): the claim that "EDS cannot show Li" is applied to Fig. 2 without anyone looking at the panels. The [memory, high] general point is sound, but what the panels actually show (photos, back-scatter images, the Au-coated zones, any EDS maps) should be read.

3. Missed material (found by QE in the text and page images)

1. Fig. 9 [read: page image p.7]. The symmetric Cu//Li/Li-glass/Li//Cu C2032 cell was run at I = ±3 mA cm⁻² for about 200 h, with voltage excursions of about ±0.01 V after an initial transient (spikes up to about 0.075 V near 30 to 40 h). The figure is annotated "σ_optimized = 5–10 mS cm⁻¹ at 25 °C". - The arbiter (claim 2: "no critical current density"; "Settles it: ... at mA/cm²") and the primer evidence table ("No current density, capacity or cycle count given") are factually wrong. The authors did report a practical-level current density. Areal capacity per half-cycle cannot be read without the switching period. - The figure's 5 to 10 mS/cm (5e-3 to 1e-2 S/cm) is below the text's claim of "σ > 10⁻² S cm⁻¹". This is an internal inconsistency on claim 1 that no file notes.
2. Fig. 7 [read: page image p.7 of the PDF, printed p.6]. The time axis runs to about 2,600 h and is labelled "cycle 34 ... cycle 270". Cycles 1 to 33 ran at I ≈ 2 mA/g, and cycles from 34 on at 25 mA/g ("≈ C/4 MnO₂"). The inset shows capacity normalised to up to about 300 % of MnO₂. - The arbiter frames the cycle count as "170 (caption) vs over 250 (abstract)". The figure itself labels 270. Report all three numbers (170, "over 250", 270). - "C/4 at 25 mA/g" implies an MnO₂ capacity basis of about 100 mAh/g. Commonly quoted MnO₂ values are about 300 mAh/g [memory, medium], which would make it about C/12. Either way, this is not the "commercial rates" claimed in the Broader-context paragraph [read p.2]. The arbiter does not discuss the 3× MnO₂ excess capacity on the 34th cycle, which is the same anomaly as the 10.8× sulfur excess.
3. Fig. 6 [read: page image p.6]. The Na–ferrocene cell ran at I = 40 mA/g of ferrocene, and its time axis is in minutes: cycles 1 to 200 are plotted over about 120 min. Voltages are roughly 1.9 to 2.6 V. - So "ferrocene voltage not stated" (physicist, mathematician, arbiter, primer) is wrong. The figure gives roughly 2.0 to 2.5 V vs Na. - Very short cycles (minutes) carry very little charge per cycle. This weakens the "200 cycles" point, and no file says so. - The text says "the ferrocene molecule was not oxidized" [read p.6]. Ferrocene's usual couple is the oxidation Fc → Fc⁺ [memory, high], so its role as a reducible "redox center" during discharge is unexplained. Nobody raised this.
4. Fig. 5 [read: page image p.6]. The open-circuit voltage (VOC) curve was fitted with V = V₀ + (V_max − V₀)(1 − e^(−t/τ_Ci)), τ_Ci = 94 h, rising to about 2.65 V at 240 h. A second fit, V = (V_max/2)(1 + e^(−t/τ_Cf)), τ_Cf = 346 h, then falls to about 1.65 V by about 650 h. - This answers the arbiter's open question "why 1.65 V vs 2.61 V". The 1.65 V is reached after weeks on open circuit, not at a different state of charge. - A sustained open-circuit decay from 2.65 V toward V_max/2 is itself a self-discharge-like observation that the capacitor model is fitted to. It is worth adding to the primer as an example of a fitted model with free time constants.
5. Fig. 4 and the sign of μ [read: page image p.5]. The paper writes μ_A = −3.05 V and μ_C = −0.40 V vs SHE, and plots E_F = −μ > 0. The minus signs were lost in the text layer, and all files quote positive values. The difference, 2.65 V, is unchanged. - Fig. 4 gives the energy gap as "E_g > 8 V", while the text gives "> 9 eV". - Fig. 4 also predicts V_cell ≈ 2.65 V "just assembled" and about 1.6 V after "days/weeks". The μ_C value appears to be fixed from the same cells, so "3.05 − 0.40 = 2.65 = V_max" is partly circular. The contrarian said this ("looks like an inserted fit"). The arbiter and primer instead present it as evidence that the cell is "the ordinary Li/S couple". The primer also says conventional Li–S plateaus are about 2.3–2.4 V and 2.1 V [memory, high], which a 2.65–2.68 V start does not match. That argument should be softened.
6. The paper's own voltage mechanism is not engaged directly. The abstract and Broader context say the voltage is "limited by the energy difference between the Fermi levels of the anode and the cathode current collector" and "tuned by a catalytic relay" [read p.2]. The paper also says the cell reads about 3.5 V with no redox center [read p.7]. The reviews refute the Li-to-Li energy budget, but none says directly that a Fermi-level (work-function) difference between two different metals is a contact potential. That potential equalises once the two metals are connected and cannot drive a sustained current. This is the cleanest teaching point for primer section 4.
7. Electrostatic storage admission [read p.4]. "The voltage of the cell was limited to avoid electrostatic energy storage at a high voltage, as has been demonstrated can occur (ref. 2)". Only the mathematician mentions this. It bears directly on the capacitor-energy argument and belongs in the arbiter and primer.
8. Possible reactants not considered. The membrane is glass in recycled paper (cellulose) or fiberglass, cast from an ethanol slurry and heated only to above 130 °C [read p.3]. Residual ethanol, water and cellulose –OH groups react with Li metal [memory, high]. No file lists them among the candidate sources of the about 36 mAh, although the arbiter lists "water/hydroxide".
9. Published comment and reply not found. [memory, medium] A formal comment on this paper arguing a first-law problem was published in Energy Environ. Sci. (about 2018, by D. A. Steingart and V. Viswanathan), together with a reply by Braga and Goodenough. The reviewers relied on a blog (Lacey) and Wikipedia instead. This exchange should be web-verified. If confirmed, it is the primary critique and reply, and should replace or supplement the blog. It would also add to the arbiter's account of how the work was received.
10. Fig. 3 [read: page image p.5]. The Li–S cycling window is about 2.50 to 2.66 V: discharge 2.53→2.51 V and constant-voltage charge about 2.66 V, never near 2.34 V. Each discharge passes 40 mA/g × 10 h = 400 mAh/g. Assuming "g" means g of S, which the paper does not state, that is about 24 % of sulfur capacity per cycle. If so, the 46-cycle result could in principle be carried by sulfur alone, and this should be stated.

4. Accuracy spot checks

5. Process

• Source tags: present and mostly honest throughout.
• Physicist tags the Li₁₀GeP₂S₁₂ barrier (0.25 eV) as [memory, high]. The arbiter and primer tag the same fact [memory, medium]. Harmonise to medium.
• Physicist cites the Graz Chem. Mater. paper as [web] but did not read it. The arbiter correctly downgraded it to secondary and unverified. The primer's InsideEVs URL returned 502 to the arbiter. The primer marks it "not verified", which is acceptable.
• Web checks:
• Physicist: about 3 checks. Contrarian: 1 successful check (Wikipedia; search refused). Arbiter: 4 of 8 calls, 2 succeeded.
• The medium-confidence memory facts left unverified (Li₃OCl and LGPS conductivities, permittivity ranges) are correctly declared non-load-bearing.
• The external critique literature was under-searched (missed item 9).
• Verdict scale: applied consistently, with split verdicts clearly stated.
• Using "Internally inconsistent" for a number (10.5 Wh/g) is a reasonable use of the scale.
• The physicist's split on claim 1 ("Tested, not supported (composition)") was properly overruled by the arbiter, with reasons given.
• Primer:
• Follows the nine-part STEP 4 order. Symbols are defined (F, R, T, activity, ε₀, Fermi level, band gap vs window), acronyms are expanded, and the worked example uses real numbers.
• The answers to the check-yourself questions are correct. QE recomputed Q2: 75 µF × 2.65 V = 2.0e-4 C, a fraction of 1.5e-6. QE recomputed Q5: exp(0.06 × 8.9) = 1.7.
• Reader-profile level is appropriate.
• example.py: runs cleanly with the venv (exit 0). Output matches primer section 5: 129 C, 315 J, mean 2.45 V, 8.5 and 7.0 Wh/g, 48.6 F, a factor of 1.6e6, 171 J vs 315 J, ε_r 2.2e12, and 10^44.8.

6. Remediation actions

Deep reads (page images required; the text layer loses figure content and minus signs): 1. Alt-Strat-for-Safe-Rech-Battery-d78469, pages 5-7, Opus. Render Figs. 3, 5, 6, 7, 8 and 9 and extract every quantitative value: - Fig. 9: current density, half-cycle period, areal capacity per half-cycle, overpotential, and the "σ_optimized 5–10 mS/cm" note. - Fig. 7: cycle labels (34, 270), currents (2 and 25 mA/g), and inset capacity in % of MnO₂. - Fig. 6: Na–ferrocene voltages and the time per cycle. - Fig. 5: the fitted VOC model and the elapsed time at 1.65 V. - Fig. 3: charge and discharge voltage window and charge per cycle. - Fig. 4: μ signs and E_g > 8 V.

State which arbiter and primer statements each figure confirms or contradicts. 2. Alt-Strat-for-Safe-Rech-Battery-d78469, pages 4-4, Sonnet. Render Fig. 2 (panels a to i). What do the photos and SEM back-scatter images actually show? Is there any EDS spectrum or map? Which zones were Au-coated? Does the figure support "Li plated on the cathode" beyond visual appearance?

Web check (arbiter, during the QE revision): find and read the Energy Environ. Sci. comment on doi:10.1039/c6ee02888h (Steingart and Viswanathan, about 2018 [memory, medium]) and the Braga–Goodenough reply. Retag them [web] with URLs, or record that the check failed.

facts.json: - Update how_read to "full text + page images p.3 to p.7 (QE)". - Correct the μ signs. - Add the Fig. 9 current density (3 mA/cm², about 200 h) and the 5–10 mS/cm annotation. - Add the cycle-270 label, the Na–ferrocene voltage (about 2.0–2.5 V) and its minute-scale cycles, and the Fig. 5 time constants. - Add the paper's "electrostatic energy storage" admission. - Add E_g > 8 V (Fig. 4) vs > 9 eV (text).

claims.md: - Claim 1: note that the figure gives 5–10 mS/cm against the text's > 10 mS/cm. - Claim 2: add "symmetric cell at ±3 mA/cm² for about 200 h (authors)". - Claim 4: replace "about 2.5 to 3.1 V" with the figure values (Na–Fc about 2.0–2.5 V vs Na; Li–S 2.55–2.68 V; Li–MnO₂ 3.04–3.08 V). - Claim 5: give the cycle counts as 46 / 200 (minute-scale) / 170, 250 or 270.

arbiter.md: add a "QE revision" section that: - (a) corrects "no current density" on claim 2 and re-decides the weak-form verdict on the Fig. 9 evidence. "Plausible, untested" is likely still right because the data are the authors' only, but the reasoning must change; - (b) adds the 5–10 vs > 10 mS/cm internal inconsistency to claim 1; - (c) answers the 1.65 V vs 2.61 V open question from Fig. 5; - (d) qualifies the "3.05 − 0.40 = 2.65 = ordinary Li/S couple" argument as partly circular (μ_C is fitted from the same cells) and in tension with conventional Li–S plateaus; - (e) adds the contact-potential (Fermi-level difference) rebuttal to the paper's stated mechanism and to the "3.5 V without redox center" claim; - (f) adds the 3× MnO₂ excess as a second instance of the excess-capacity anomaly; - (g) adds residual ethanol, water and cellulose as candidate reactants for the excess charge; - (h) records the outcome of the EES comment/reply web check.

primer.md: - Section 4: add a contact-potential explanation. Two metals with different work functions show a Volta potential, but connecting them equalises the Fermi levels, so no sustained current flows. Explain it with a numerical example, citing a standard source or marking it [memory]. - Section 4: soften "the 2.65 V coincidence". - Section 6: fix the claim-2 row (Fig. 9 at 3 mA/cm²), the claim-1 row (5–10 mS/cm in the figure), the claim-4 row (ferrocene voltage) and the claim-5 row (cycle labels; minute-scale Na cycles). - Section 7, "Genuinely open": drop the resolved 1.65 V question and add the ferrocene-oxidation puzzle. - Correct the sign of μ wherever it is quoted. - Section 9: add the EES comment and reply if verified.

example.py: no change needed.

QE recheck (Sonnet): confirm each item above was addressed and append a "Recheck" section here.

Recheck

Rechecked against the current facts.json, claims.md, arbiter.md (incl. "QE revision"), primer.md, example.py and deep/Alt-Strat-for-Safe-Rech-Battery-d78469.md.

Deep reads 1. Pages 5-7, Figs. 3-9 (Opus): FIXED. deep/ file exists; its values feed facts.json, claims.md, the arbiter revision and the primer. 2. Page 4, Fig. 2 (Sonnet): FIXED. Arbiter QE revision item 11 and facts.json record what the panels show (no EDS spectrum or map, dark crusty deposit, caption-only assignments).

Web check (EES comment and reply): FIXED as permitted by the QE action ("or record that the check failed"). The arbiter records that the check failed (search refused, RSC page 403) and keeps the item [memory, medium]. The primer's further reading says "not verified". The item stays unverified.

facts.json: FIXED. how_read updated; mu signs corrected; Fig. 9 (3 mA/cm2, about 200 h, 5-10 mS/cm); cycle 270; ferrocene voltages (QE's "minute-scale cycles" was corrected by the deep read to about 2 h half-cycles; the deep read is the better source); Fig. 5 time constants; electrostatic-storage admission; E_g > 8 V vs > 9 eV.

claims.md: FIXED. Claim 1 (5-10 vs > 10 mS/cm), claim 2 (+-3 mA/cm2), claim 4 (figure voltages), claim 5 (cycle counts 46 / 200 / 170, 250 or 270) all updated; source-gap note kept.

arbiter.md QE revision (a)-(h): FIXED. (a) Fig. 9 correction, weak form kept Plausible, untested with new reasoning; (b) 5-10 vs > 10 mS/cm; (c) 1.65 V vs 2.61 V resolved as cell age; (d) 2.65 V coincidence qualified as partly circular; (e) contact-potential rebuttal; (f) MnO2 excess (sized at about 1.6x theoretical, a weaker anomaly than QE assumed); (g) ethanol, water, cellulose; (h) web-check outcome. Stale statements in the body are marked superseded and the corrected claim list is updated.

primer.md: FIXED. Section 4 has the contact-potential explanation and a worked number (about 4e-10 C), the softened 2.65 V coincidence and the mu sign note; section 6 rows for claims 1, 2, 4 and 5 corrected; section 7 drops the resolved 1.65 V question and adds the ferrocene puzzle and the voltage-jump anomaly; section 9 lists the comment and reply as unverified.

example.py: no change required; runs with exit 0.

Non-blocking note (not a remediation action): physicist.md still tags the Li10GeP2S12 barrier [memory, high] while arbiter.md and primer.md use medium. No verdict depends on it, and the downstream files carry the more cautious tag. Left as is.

Final: PASS