Heaviside: Quality Engineering (Step 4b)

Inputs checked: _text/manifest.json, all 3 digests, facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py. No deep/ folder exists (none was requested). Spot checks were made against the extracted text and against the saved page images in the source folder. I copied the images to the scratchpad and viewed them there, and ran tesseract on the Electrical Papers contents pages. Nothing in the source folder was modified.

1. Coverage

The real content is not in the HTML. It is in the extensionless image files content(1), content(2) and content(3) inside each *_files/ folder: nine PNG page scans, each about 1025-1280 by 1670-2070 pixels. Step 0 (textprep) did not list or OCR these files, and the manifest says nothing about them. This is a gap in the tool. When I ran tesseract on the Electrical Papers images, it read them cleanly, so they could have been extracted. Step 1 made up for the gap by reading all nine images directly (facts.json how_read: "page images"), and the reviewers, the arbiter and the primer author re-read the Waves pages. So every page of content in the folder was read in full. No text that should have been read was missed. The remaining 66 files in _files/ are browser junk (JS, CSS, icons, cover thumbnails content at 49x80 to 128x198 pixels, an ad GIF). facts.json records them in one entry, as the prompt requires.

Source gap: claims.md, the arbiter and the primer all state it prominently. They correctly say that only claim 4 rests on pages actually read. This is honest and complete.

2. Shortcuts

• No sampling took place. Each "document" is 1 HTML page plus 3 images, and all of it was read.
• No deep reads were needed for documents in the folder. A digest-only reading would have been a defect, because the digests describe only viewer chrome. That was avoided by reading the images.
• Claims 1, 2, 3, 5, 6 and 7 rest on contents headings plus [web] (Wikipedia) and [memory]. This is not a shortcut: the folder holds nothing more. Every downstream file declares it.
• Acceptable overall.

3. Missed material (from my own reading of the digests and page images)

1. Operational notation is in the folder. p.142, sec. 10 "Effect of Impressed Force": "In equations (23), (24), let p = ni, where n/2π = frequency of alternations" [read: page image p.142]. This is Heaviside's operator p (= d/dt) set equal to iω for sinusoidal driving. It is exactly the bridge from operational calculus to phasors that primer section 2 ("Complex exponentials") teaches. The primer (3.5) and the arbiter (claim 5) both say "no document in the folder contains the calculus". That is slightly wrong: the folder holds a direct, if small, use of the operator. (The printed text reads "n2π = frequency", an old typesetting of n/2π; I am not changing the meaning.)
2. Limit of high frequency (p.142): as n grows, k and g act only as small attenuation exponents, and E and H come into phase. This is a clean statement of the low-loss limit. No reviewer used it.
3. Electric-magnetic duality (p.143): "To produce dissipation of H with persistence of E requires the plate to be a magnetic, not an electric conductor", and on the same page, "the vorticity of e produces magnetic induction, that of h produces electric displacement". The primer mentions the k/g balance from p.139 but not this duality statement. It would strengthen the primer's point that the fictitious magnetic conductivity g is the origin of the distortionless condition.
4. Electromagnetic Theory contents (p. xv) also lists sec. 63 "The Cardinal Feature of Maxwell's System. Advice to anti-Maxwellians", sec. 73 "Uncertainty regarding the General Application of the Electromagnetic Stress" and sec. 79 "Resultant Action on Solid Body independent of the Internal Stress, which is statically indeterminate". Sec. 79 is directly relevant to the arbiter's 7a note that different force densities give the same total force. Heaviside's own heading says the same thing, so 7a can cite it as [read: contents sec. 79]. The contrarian mentions sec. 73 and sec. 79; the arbiter does not connect sec. 79 to 7a.
5. Electrical Papers contents also include "Transference of Energy. Ohm's Law", "The Conservation of Energy", "The Minimum Heat Property in Conductors" and the thermo-electricity sections [read: page image, tesseract]. facts.json summarises these adequately. No defect.
6. facts.json gives Electromagnetic Theory Chapter II as "sections 20-85, pp. 20-131". The page I viewed shows sec. 85 at p. 107. Where "131" comes from was not checked. This is trivial and does not change any claim.

No missed claims, experiments or people would change any verdict.

4. Accuracy spot checks

Every check passed. I found no fabricated or misquoted text.

5. Process

• Source tags: present and used honestly throughout. [read] is used only for the page images. Wikipedia-based facts are tagged [web] with URLs, and recall is tagged [memory] with a confidence level. The primer adds an [arbiter] tag and defines it. Acceptable.
• Web checks: the physicist and the mathematician each used 1 web call out of a budget of about 6. The mathematician left the Gravity Probe B (GP-B) and Appleton dates as "check failed to be completed". The arbiter then web-checked both, using 5 of 8 calls: the Kennelly-Heaviside layer, GP-B, gravitoelectromagnetism (GEM) and the loading coil. The verdict-relevant facts that were medium-confidence are now [web]. The process is weak at Step 2 but the gap was repaired at Step 3. Facts left as [memory] (Hertz, Aharonov-Bohm, lattice spacing) are high-confidence or not load-bearing.
• Verdict scale: applied consistently. Split verdicts (claims 2, 4, 5) are stated part by part. One caveat: 7a "Not testable as stated (from these sources)" and 7c use that label to mean "unread", not "untestable". The arbiter says so explicitly, so this is acceptable but a little loose.
• Primer vs STEP 4: all nine sections are present and in order, with answers at the end. Symbols are defined, CGS units are explained, and the worked example uses real numbers. It fits the reader profile: no tensors, and phasors and the Laplace transform are introduced from scratch.
• example.py: runs cleanly with the .venv Python. Its output matches every number in primer sections 5.3-5.6 (8.4883e-09 cm; 0.0849/0.0892/0.0915 nm; T/R/A = 1/4, 1/4, 1/2; the α table; phase velocity 2.8284e3 km/s).

6. Verdict: MINOR

Coverage is complete: every page of content in the folder was read in full. The spot checks are exact and the verdicts are sound. The defects are small and fixable in facts.json, the arbiter and the primer: one inaccurate statement about the folder's content (point 3.1), a few unused details from the pages that were read (3.3, 3.4), and a gap in the Step 0 tool. No central claim or verdict changes.

7. Remediation actions

Deep reads: none. Every document in the folder has been read in full, and the remaining gaps (Electromagnetic Theory full text, pp. 136-138 and 140-141 of the Waves paper) are not in the folder.

1. primer.md 3.5 and arbiter.md claim 5: replace "No document in the folder contains the calculus" with: the folder shows only one use of the operator, "let p = ni" on p.142 [read: page image p.142], tying it to phasors; the expansion theorem itself is not in the folder.
2. primer.md 4.1 (optional): add the p.143 duality sentence, that a magnetic rather than electric conductor would dissipate H and preserve E, beside the k/g no-tail remark.
3. arbiter.md "QE revision" section: under 7a, cite Theory contents sec. 79 ("Resultant Action on Solid Body independent of the Internal Stress, which is statically indeterminate") [read: contents p. xv] as Heaviside's own statement that supports the total-force point. Verdict unchanged.
4. facts.json: for Electromagnetic Theory, either correct "pp. 20-131" to what the images show (secs. 62-85 run to p. 107) or drop the page span. For every book, add to how_read that the images are the extensionless _files/content(1)-(3) files, not OCR'd by Step 0.
5. Tooling (outside this project's files; report to the batch owner): tools/textprep.py should detect extensionless image files in saved-page _files/ folders (by magic bytes, at least about 800 px tall) and OCR them, or at least list them in the manifest as skipped. Saved Google Books pages are common in this collection.
6. No changes to the verdicts in physicist.md, mathematician.md or contrarian.md are needed.

Recheck

Checked against the current arbiter.md, primer.md, facts.json and example.py.

Defects (section 3): - 3.1 "No document contains the calculus" statement: FIXED. The primer (3.5 and the evidence table) and the arbiter (claim 5, QE revision item 1) now cite "let p = ni" [read: p.142] and say the expansion theorem is absent. - 3.2 High-frequency limit unused: FIXED. Arbiter QE revision item 2 states it. - 3.3 p.143 duality statement missing: FIXED. Primer section 4 and arbiter QE revision item 2 quote it. - 3.4 Theory sec. 79 / sec. 73 / sec. 63 not connected to 7a: FIXED. Arbiter QE revision item 3 and primer evidence row 7a and open questions cite them [read: contents p. xv]. Verdict unchanged. - 3.5 Electrical Papers extra contents: no defect, no action needed. - 3.6 "pp. 20-131" span: FIXED. The span is dropped in facts.json with a QE note.

Remediation actions (section 7): 1. primer 3.5 and arbiter claim 5 wording: FIXED. 2. primer 4.1 duality sentence: FIXED. 3. arbiter "QE revision" section with sec. 79 under 7a: FIXED (section exists, verdicts unchanged). 4. facts.json span correction and how_read note on the extensionless content(1)-(3) images: FIXED for all three books. 5. tools/textprep.py handling of extensionless image files: REMAINS (grep of tools/textprep.py finds no such handling). This is outside the project's files and was assigned to the batch owner. It does not affect any Heaviside file or verdict. 6. No verdict changes needed in the three reviews: nothing to do.

example.py runs cleanly with the .venv Python.

Final: PASS

Evidence-rule check (2026-10-05)

Read arbiter.md (170 lines), primer.md (301 lines) and, for the keep-pertinent-information rule, contrarian.md. Checked against the four SOURCE HONESTY rules in prompt.txt (lines 96-119).

1. No suppression of evidence: PASS. Single-source material is quoted with its source and weighed: Pupin's 1894 mountain account (arbiter l.46, primer l.241), the Eves quotation (arbiter l.73, primer l.262), the Preece-motive article (arbiter l.18, l.44; primer l.239), the Royal Society referee's ground and Heaviside's reply, marked as a search summary not checked verbatim (arbiter l.62; primer l.88). The grep for dismissal labels ("uncorroborated", "single source", "hearsay") finds none. "Not testable as stated" on 7c and on the local flow in claim 2 is explained, not used as a dismissal (arbiter l.38, l.71-73; primer l.222). The withdrawn "not always safe [memory]" line is removed (arbiter l.62, l.163). Evidence against a claim is stated (GP-B vs "confirmed", arbiter l.16, l.72; eq. 63 unphysical thickness, l.12).
2. Interested parties: PASS. Preece, AT&T, Pupin, the Royal Society referee and Eves are each reported as that party's position with its stake beside Heaviside's own (arbiter l.43-48; primer l.237-244). The Heaviside article's statement on Preece's motive is attributed as the article's characterization, its primary basis is flagged as not cited, and the "utter blockhead" quotation is marked hostile. The Post Office record is stated as non-adoption, not a failed test.
3. Absence of evidence: PASS. Four silences are analysed with competing explanations, evidence for and against, and deciding records (arbiter l.147-154; primer l.246-253): the 1887 manuscript, the unpublished fourth volume, Operators Part III, and the 7a argument. Both documents state that no Investigator run exists, so these are leads, not findings. The memory-low recollection of papers lost after his death is flagged as unverified.
4. Keep pertinent information: PASS. Every reviewer position that differs from the arbiter is retained (7c's three verdicts, arbiter l.73; primer l.222; contrarian's 3/2 reading, arbiter l.11). The contrarian's tested/dismissed distinction, the AT&T commission, the 1887 block and the Operators rejection all appear in arbiter l.75-89 and primer l.244. Minor items from contrarian.md not carried into the arbiter or primer (the permalloy and mu-metal loaded cables of 1923-24; the digestion paraphrase, which the contrarian itself said not to rely on) are peripheral and do not bear on any verdict. No material was dropped because of a negative judgement.

Violations: none.

Final: PASS

Investigator update check (2026-10-05)

Read: prompt.txt (STEP 4b, SOURCE HONESTY), the six research/ notes, sites.md, the "Investigator findings (2026-10-05)" sections of physicist.md, mathematician.md, contrarian.md and arbiter.md, and the updated arbiter.md and primer.md. No new web calls.

Checked and holding: - Weighing: every major lead in sites.md "Leads for review" (1887 text, Preece block and its two mechanisms, Biggs, 1873 Culley exchange, Kelvin 1887 letter, Preece draft, Arthur's later stance, Burnside, Britannica 1902 text, sec. 83, App. B, Vol. 3 keyword counts, patent 1407, notebook 005/I/1/12, Josephs/floorboards, 1932 GPO report, Pupin's three payment figures, Kennelly unread) appears in the reviews, the arbiter and, with one exception below, the primer. The contrarian's "Kennelly's reflection" clause and the 3/2 glyph, the u exponent slip and the 1925 obituary vs Hunt/Bruton conflict are stated and weighed, not dropped. - Source tags: Investigator material is tagged [web, via Investigator] with URL; the arbiter's own fetch of the EMT1 OCR is tagged [web] with its URL and its extra fetches are disclosed. Snippet-only items (Burnside, deadreckonings, Kennelly citation) are marked as such. - SOURCE HONESTY: single-source items (Biggs's hint, Pupin's 1894 account, 1925 obituaries) are quoted with their source and weighed, not labelled away; no dismissal labels in the new text. Interested parties (Preece, AT&T, Pupin, Wiener, admirer obituarists, Heaviside's own 1922 letter) are attributed as that party's account with the stake stated. Silences are analysed with competing explanations and deciding records (six items, arbiter and primer agree). - Primer vs arbiter: the verdict table, evidence table, 7a/7c distinction, 7b exponent note, block mechanisms, interested-party list and six "records that go quiet" match. No verdict changed category in either file.

Defects: 1. Retention, claim 3: Heaviside's own cited experimental evidence in EP2 Sec. XL (good long-distance telephony on copper-covered steel wire, where current entering the steel "lowers the resistance and increases the inductance", offered with "I do not wish to magnify its importance") and his self-flagged "unbelievable result" caveat (sinusoidal current at the far end of an Atlantic cable exceeding the steady current, blamed on reflections) are in research/archive-org.md and in all three reviews (physicist l.122, mathematician l.127-130, contrarian l.123-124) but absent from arbiter.md and primer.md. Pertinent pre-1899 evidence for and against his own claim was dropped at condensation. Fix: add both to arbiter claim 3 and primer 3.4, with the reviewers' reading. 2. Retention, own work and experiments: EP2 reports by Oliver of Arthur's observations (oscillations making telegraph ABC indicator pointers jump, which prompted part of the induction work) and Arthur's tube-return experiment finding no inductive action outside the tube (EP2 pp. ~185-186) [research/archive-org.md, section B]. Not in any review, the arbiter or the primer. It is an experiment and a witness account bearing on the brothers' collaboration and on claim 3. Fix: add one line to arbiter claim 3 and the primer's block narrative. 3. Primer vs arbiter, retention: the arbiter's "Material kept in the record" list includes the Stevenson correspondence (SC MSS 005/I/6/43, "Heaviside's brother's career and being blocked by Preece"), Preece's "America Revisited - 1907" proof with Heaviside's MS notes (005/I/2/057) and Yavetz 2011; the primer's "Kept in the record, unweighed" paragraph lists only patent 1407 and does not carry these. Fix: add them to that paragraph. 4. Retention, peripheral (low weight): Hunt's 1894 Royal Society Relief Fund refusal and the 1896 civil-list pension (GBP 120, later 220), the IET blog's note that Heaviside was expelled earlier for unpaid membership fees before the 1908 IEE honorary membership (Bruton dates the striking-off 1881), the Google snippet about a letter to Bethenod on refusing the Hughes medal, and Föppl 1894 (in the primer, not the arbiter) appear nowhere or only in one of the two. They bear on poverty, institutional relations and the outsider story. Fix: add one line each to the arbiter's kept-material list and primer. 5. Stale or inconsistent statements: (a) arbiter "Evidence-rule revision" item 10 says no Investigator run exists for this project, and the "Web-check log" heading says "8 budget, 5 calls used" while the evidence-rule section says 8 of 8 used; both are earlier text left uncorrected beside the update (add "superseded" notes). (b) sites.md still lists "(u/v)^2 = 10^-6 for u = 30 km/s", the figure all three reviewers and the arbiter showed is arithmetically wrong (should be 3 x 10^7 cm/s = 300 km/s, exponent probably 7); annotate it.

Verdicts and source discipline are sound; the open items are omissions and stale lines, fixable by adding text only.

Final: OPEN (5 issues)

Recheck (2026-10-05)

Re-read the current arbiter.md, primer.md, the three reviews and the fix notes in the arbiter's QE revision section, and grepped both files for each item.

1. Claim 3, Heaviside's own evidence and caveat: FIXED. Arbiter claim 3 (l.60) and primer 3.4 (l.95-96) carry the copper-covered steel wire telephony ("lowers the resistance and increases the inductance", "I do not wish to magnify its importance") and the "unbelievable result" Atlantic-cable caveat, tagged [web, via Investigator] with the archive.org URL.
2. Arthur's observations and tube-return experiment: FIXED. Arbiter l.61 and primer l.295 report the ABC indicator pointer observations and the tube experiment as Oliver's report of Arthur's work, noting no reviewer weighed them.
3. Primer kept-material paragraph: FIXED. Primer l.301 now carries the Stevenson correspondence, the "America Revisited" proof and Yavetz 2011, matching the arbiter's list.
4. Peripheral retention: FIXED. The 1894 Relief Fund refusal, the civil-list pension (GBP 120, later 220), the expulsion for unpaid fees (IET blog; Bruton 1881), the Bethenod/Hughes-medal snippet (marked unread) and Foppl 1894 are in the arbiter's kept-material list and in primer l.303. 5a. Stale statements: FIXED. Arbiter evidence-rule item 10 and the web-check log heading now carry "Superseded" notes (5 calls at Step 3, 8 of 8 after the evidence-rule revision, 3 disclosed over-budget fetches, 11 in all). 5b. sites.md exponent: FIXED for the deliverables. sites.md is the Investigator's file, not edited by this step, and its line l.39 still reads "10^-6 for u = 30 km/s". The arbiter's QE revision records the arithmetic (30 km/s gives 10^-8; 300 km/s gives 10^-6; exponent 7 fits the printed words; page image still the deciding record), and the arbiter and primer 7b carry the exponent note. The stale line is a raw research note, not a review or deliverable, and is flagged where a reader would use it.

No verdict changed category and no new defects found. All additions are tagged and attributed, and none rely on dismissal labels.

Final: PASS