QE (Step 4b): SchusterAndHemsalech

QE agent (Opus), 2026-10-04. I read everything in the project folder: the manifest, the digest, the deep read, facts.json, claims.md, sites.md, all four research notes, the three reviews, arbiter.md, primer.md and example.py. I also checked the extracted text (_text/TheConstoftheElectSpark-4e82a8.txt) directly. Tags: [read] project document, [web] URL, [calc] my own arithmetic, [memory] with confidence.

Verdict: MINOR

The single project document was read in full, with a deep read and page images for the equation and table pages. The verdicts are applied consistently and the primer is good: it is accurate, it suits the reader profile, and example.py runs. The defects are about attribution, one arithmetic error in the arbiter that the primer corrects but the arbiter still carries, stale how_read notes in facts.json and claims.md, missing reviewer web checks, and some source material missing from the primer. None of them changes a verdict on the project document. One larger gap lies outside the manifest: the full 1900 paper (Phil. Trans. A 193, 189-213), which is free to view, was never obtained. Reading it could change verdict 1c (the 6 microsecond against 560 m/s mismatch) and could strengthen or weaken claim 5. That is a task for the Investigator recheck, not a deep read, because the paper is not in the manifest.

1. Coverage

No document was skipped. research/files/ is empty (see section 5).

Note: digest/TheConstoftheElectSpark-4e82a8.md (21 KB) sits in digest/ but is deep-read material. It does no harm, but it is a misplaced file and should be listed in the artifact index or left alone. No action is needed for this review.

2. Shortcuts

• The only document is short, and it was read in full twice (digest and deep read), with page images for the equation and table pages. No sampling was done. Acceptable.
• The real shortcut is about scope. Every reviewer, the arbiter and the primer correctly say the document is only an abstract, and that the full paper with tables for Al, Mg, Bi, Cd and Hg and the photographs was read by no one. The Investigator found the full paper free to view (DOI 10.1098/rsta.1900.0006) and did not download it, giving as the reason "file downloads need the owner's explicit permission in this session" (sites.md). prompt.txt Step I says to keep copies of useful documents "when downloadable". The Investigator also used about 3.5 of its 10 minutes and left 8 sites "not reached", among them NASA ADS (Schaeffer 1908, Schenck 1901), Gallica (Hemsalech's Comptes rendus notes) and Royds 1908 pp. 334-347. Claims 1c, 3 (which electrode supplies the vapour) and 5 turn on these sources. This is a defect of the Investigator step. It does not justify a MAJOR verdict, because every verdict says plainly that it rests on the abstract.

3. Missed material (from the digest and the source text)

1. The calcium H and K observation belongs to Schuster and Hemsalech, not to Royds. Royds 1908 p.333 lists it as point (5) of his summary of the 1899/1900 results: calcium impurity lines "became luminous first in the centre of the gap, and were propagated towards the electrodes" (research/2026-10-04-royalsocietypublishing.md, item 3: "not in the abstract; in the full paper"). The primer (section 4, "Royds also reports...", and section 7, "Why did calcium impurity light from the gap centre outward (Royds 1908...)") and the arbiter's open questions present it as Royds's own finding. The contrarian.md wording is ambiguous too. This matters for the record: the inventors themselves recorded the observation that fits the "luminosity front" reading best, in their own full paper, and the abstract leaves it out. It should be attributed to them, "as summarised by Royds 1908; full paper not read".
2. "The metallic vapours remain luminous in the centre of the spark for a longer period than near the poles" [read: p.6]. This is the authors' own observation and bears directly on the matter-against-front question (claims 1 and 2). It is consistent with the calcium observation. facts.json, the arbiter and the primer leave it out.
3. The full timing set on p.6. Air stays luminous about 5 x 10^-7 s. Light persists about 1.5 x 10^-5 s with six jars. The vapours "then begin to diffuse" after the air glow and reach the centre in 6 x 10^-6 s [read: p.6]. The primer gives only the 6 microsecond figure. The phrase "then begin" lets the 6 microseconds be counted from the start of diffusion, which the arbiter allowed for (909 m/s). The primer's worked example (step 4) should mention it, and add the 5 x 10^-7 s and 1.5 x 10^-5 s figures to the timeline. The physicist's point that 5 x 10^-7 s is below the film's resolution of about 1 microsecond at 100 m/s is pertinent and appears in no later file.
4. Bismuth line 3793 slower than zinc and cadmium; bismuth taken with 2 jars instead of 6; mercury tried with amalgamated zinc and cadmium poles; magnesium too scattered to measure, and Feddersen's view "partly true" for magnesium [read: p.4-5]. The primer's evidence table gives Bi 1420 m/s only. Because bismuth is the main "anomaly", the primer should say that it was measured under different conditions (2 jars), which the authors state.
5. Crookes's referee request about metal purity (research/2026-10-04-makingscience-royalsociety.md). The arbiter has it, the primer does not. It belongs beside item 1, because the calcium is an impurity.
6. The energy scale. The physicist's [calc] gives 15 J stored at about 30 kV, "ample to ablate". It is not used later. This is optional for the primer.

4. Accuracy spot checks (against the extracted text)

example.py runs cleanly under the .venv (exit 0) and reproduces every number in the primer's section 5. I checked the answers to the check-yourself questions by hand: 1200 and 1431 m/s; 1190 K and 1277 K; 6.3 microseconds. All correct.

5. Retention check

• Kept: Hull 1906 (with its limits stated fairly), Royds's stationary-vapour remark, Hemsalech 1901 with its two groups of metals (tagged as an interested party), the three-way disagreement over which electrode supplies the vapour, Schaeffer's credit, the referee reports, the NIST Zn II identification, and the absence-of-evidence analysis. The primer keeps the case for the authors' reading beside the challenge. Nothing was dropped because it was judged negatively, and the reviewers state the material for both sides.
• Defect (attribution, not deletion): item 3.1. The calcium observation is kept but credited to the wrong party. It understates that the inventors recorded evidence against their own simplest reading.
• Thin retention: items 3.2 to 3.5 are pertinent observations from the source that are missing from the primer. They were not omitted because of a negative judgement, so they count as missed material rather than as a retention defect. They should still be added.
• The Investigator's unreached leads (Schenck 1901, Milner, Strutt 1913, Royds 1910, Hemsalech's Comptes rendus 1899-1906 notes, Beilis 2018) are listed as "not read" in the primer's further reading and evidence table. Retention is acceptable.

6. Process

• Source tags: present and honest throughout. The primer and arbiter separate [web, via Investigator] from direct reading well (for example, Hull's text was read directly only by the Contrarian).
• facts.json / claims.md how_read is stale: facts.json says "no separate deep read file exists... reconstructed by a numerical check, not from a page image", and claims.md says "No deep reads were needed". Both are false: the deep read exists and rendered p.3 and p.5. The mathematician and the arbiter flagged this, and it is still not fixed. facts.json key_claim 1 says the vapour "crosses a 1 cm gap"; the text says "reach the centre" (arbiter correction, not applied). mainstream_conflict "none evident" should note the contested interpretation (Hull 1906, Royds 1908).
• Web checks: the physicist and the mathematician did no web checks of their own ("I did not repeat web searches"). The Contrarian checked NIST and Hull, and the arbiter checked NIST once. Several medium-confidence [memory] facts were left unchecked:
• spark-channel temperatures of 10^4 K or more, used against 2700 K;
• D of 1 to 5 cm^2/s;
• cadmium vapour being monatomic;
• Hull's initials and Dartmouth affiliation;
• vacuum-arc jet speeds.

The arbiter says no verdict turns on them. For the D value the diffusion conclusion holds by a factor of more than 60, so that is fair. The spark-temperature figure supports the "Historically superseded" half of verdict 4, though, and should be checked or tagged in the verdict line. This is a minor process defect. - Verdict scale: applied consistently, and every split verdict is stated in parts. The three reviewers differed in their labels for claim 5. The arbiter resolved this with a reason (a null statement about scattered data is "Not testable as stated"). Acceptable. - Primer against STEP 4 and the reader profile: all nine sections are present and in order. Every symbol is defined (gamma, R, M, rho relative to hydrogen, v_f). Zn I and Zn II and the angstrom are explained. Textbooks are named without unsure chapter numbers. The worked example uses real numbers, with a script. The "Source gap" note is repeated. Good. - Investigator: a premature stop (3.5 of 10 minutes), and the key free primary source was not saved (section 2).

7. Remediation actions

Deep reads: none required. The one manifest document is fully covered. Its page images for the table and the formula were already checked, and the p.6 timing text is clean in the text layer.

Outside the deep-read mechanism (highest value, for the Investigator recheck or the next run): - R1. Get Schuster and Hemsalech, Phil. Trans. A 193 (1900) 189-213 (DOI 10.1098/rsta.1900.0006) and read it. Check whether the 6 microsecond and 560 m/s figures use different averages (verdict 1c). Get the tables for Al, Mg, Bi, Cd and Hg and the line list (to confirm Zn II for 4925), and the calcium H and K passage. Also get Royds 1908 pp. 334-347 (the velocity table by line type). If 1c is reconciled, change its verdict in an arbiter QE revision.

facts.json: - F1. Correct how_read: "digest + deep read pp.1-6 (page images p.3, p.5)". Delete the "no separate deep read" and "not from a page image" statements. - F2. Key claim 1: change "crosses a 1 cm gap" to "reaches the centre (5 mm) of a 1 cm gap". - F3. mainstream_conflict: change to "interpretation of slope as vapour velocity contested (Hull 1906 Doppler limit; Royds 1908)". - F4. Add the p.3 wording slip, the "centre luminous longer than poles" observation, and the bismuth 2-jar condition.

claims.md: - C1. Remove "No deep reads were needed" and replace it with a note that the deep read exists. - C2. Apply the arbiter's corrections, which have not been made yet: item 2 should read "was taken by the authors to measure", with the Hull and Royds challenge; item 5 should include the abstract's "Al and Mg faster" half.

arbiter.md: add a "QE revision" section that does the following: - A1. Correct numbers point 5 and the "Corrections" bullet. The diffusion length is 0.035 to 0.077 mm, short by a factor of about 65 to 145. The physicist's "about 0.1 mm, fifty times" was roughly right and should not be "corrected". The conclusion is unchanged and stronger. - A2. Re-attribute the calcium H and K observation to Schuster and Hemsalech's full 1900 paper, as summarised by Royds 1908 p.333. Add the abstract's own statement that the centre stays luminous longer than the poles [read: p.6] as evidence bearing on claims 1b and 2. No verdict change. It adds weight to "Never properly tested" rather than to "Established" for the matter reading. - A3. Either web-check the spark-channel temperature (order 10^4 K) used in claim 4 or mark that half of the verdict as resting on [memory: medium].

primer.md: - P1. Section 4 and section 7: the calcium observation is the authors' own (full paper), reported by Royds. Say so, and add that the inventors recorded it. - P2. Fix "A. Royds" to "T. Royds". - P3. Section 5, step 4: add the full p.6 timeline (5 x 10^-7 s of air glow, below film resolution; 6 x 10^-6 s, possibly counted from when the vapour "begins"; 1.5 x 10^-5 s of persistence; the centre stays luminous longer than the poles). - P4. Evidence table, claim 5: note that bismuth was measured with 2 jars and that line 3793 was slower. Mention the mercury amalgam attempt and that magnesium could not be measured. - P5. Add Crookes's purity request beside the calcium-impurity point. - P6. The primer's own correction note on the arbiter's diffusion figure can stay. Once A1 is made, change it to "agrees with arbiter QE revision".

Recheck

Recheck (2026-10-04), against the current files.

• R1 (obtain the full 1900 paper, Phil. Trans. A 193, 189-213): REMAINING. research/files/ is still empty. The primer and claims.md state the source gap honestly, and no verdict claims more than the abstract supports. This is a source limitation, not a defect in the written files.
• F1 (facts.json how_read): FIXED. It now cites the digest, the deep read and the page images, and says the full paper is unread.
• F2 (key claim 1 wording): FIXED. It now reads "reaches the centre (5 mm) of a 1 cm gap".
• F3 (mainstream_conflict): FIXED. It now notes the contested interpretation (Hull 1906, Royds 1908).
• F4 (wording slip, centre luminous longer, bismuth 2 jars): FIXED. The bismuth 2-jar condition and the p.6 centre-luminosity observation are present. The slip appears in the primer.
• C1 (claims.md "no deep reads"): FIXED. It now says the deep read exists.
• C2 (claims 2 and 5 corrections): FIXED. Claim 2 reads "taken by the authors to measure" with the Hull and Royds challenge. Claim 5 includes "Al and Mg faster".
• A1 (diffusion arithmetic): FIXED. arbiter.md has a QE revision section, and its item 5 is corrected to 0.035 and 0.077 mm, 65 to 145 times too short.
• A2 (calcium re-attribution and p.6 centre observation): FIXED. Both are in the QE revision. No verdict changed.
• A3 (spark-temperature tag): FIXED. The "superseded" half of verdict 4 is now marked as resting on [memory: medium], unchecked.
• P1 (calcium attribution): FIXED. Primer sections 4 and 7 now attribute it to the authors' full paper, as summarised by Royds.
• P2 (Royds initial): FIXED. No "A. Royds" remains.
• P3 (full p.6 timeline): FIXED. The 5 x 10^-7 s, the film resolution and the 1.5 x 10^-5 s persistence are present.
• P4 (bismuth 2 jars, line 3793, mercury amalgam, Mg too scattered): FIXED. They are in the claim 5 evidence row. The mercury amalgam mention is brief.
• P5 (Crookes purity request): FIXED. It is in the primer next to the calcium point.
• P6 (primer note on the diffusion figure): FIXED. It now says "agrees with the arbiter's QE revision".

The remaining open item is R1, which needs a download the Investigator could not make, and which the earlier text already treats as a stated limitation. It changes no verdict as written.

Final: PASS