Farnsworth-PT: Quality Engineering (Step 4b)

Inputs checked: _text/manifest.json, digest/ (1 digest), facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py, and the full extracted text _text/US2172152-Radio-Freq-Multipactor-Amplifier-b4b016.txt (4 pages; I read all of it). No deep/ folder exists.

Verdict: MINOR. Coverage is complete and the one document was read in full by every reviewer. One reading error spread from the contrarian through the arbiter into the primer: what the "one to four" gauze ratio applies to. Two further arbiter statements were corrected in the primer but never in the arbiter itself. None of these changes a verdict. All of them can be fixed by editing facts, the arbiter (adding a QE revision section) and the primer.

---

1. Coverage

• The manifest lists exactly one document. It has one digest and one facts.json entry. No documents were skipped.
• Page 1 is the drawing sheet. Its only text is the title block, so it is acceptable that it has no prose digest.
• No defects.

2. Shortcuts

• Step 1 ran before digests existed. Because the document is only about 19.7k characters, Step 1 read the whole text instead of sampling it. facts.json how_read says "full text". That is honest, though it does not use the prescribed vocabulary (digest | digest + deep read ...). This is a cosmetic deviation.
• No deep read was made. None was needed for coverage, since three reviewers each read the full 4-page text themselves [read: their own statements]. No central claim rests on a digest alone.
• Process shortcut with a consequence. The arbiter states "I did not re-read the patent PDF; I rely on the reviewers' three independent ... readings" (arbiter.md, line 3). Where the reviewers disagreed on a reading (the "one to four" passage), the arbiter settled it without checking the text and chose the wrong side (see 4.1). For a 4-page source, the arbiter should check disputed readings against the text.

3. Missed material

I read the digest and the full text myself. Items that facts.json, claims.md, the reviews, the arbiter or the primer missed or got wrong:

1. Which electrons the "one to four" ratio refers to (accuracy defect, see 4.1). The patent applies it to the electrons arriving from the helix that pass through the gauze holes, not to secondaries.
2. The patent prefers the gun to light as the seed, and says why. "...inasmuch as I preferably utilize an electron gun as described above, since it possesses advantages in regard to modulation" (p.3) [read]. facts.json treats light and the gun as equal alternatives, and the primer does not mention the stated preference. This matters for claim 4. The patent does not frame the seed mainly as a photo-cathode device: its preferred embodiment omits the light source and optics ("not shown in the figure").
3. Three starting mechanisms are listed among the objects: "ionized particles existing therein, by photo emission, or by an electron gun" (p.2) [read]. The primer table mentions stray ionised particles. claims.md and facts.json key_claims do not. This is minor, but it bears on the arbiter's open question about how seeding works.
4. Gun electrons follow the same energy rule as the primer's Step 2. The coupling coil 36 ties the screen (cathode 27) to ground through its grounded side (p.2-3) [read], so the screen sits at d.c. ground. If the gun cathode is also near ground, which the patent does not state, then gun electrons that reach the screen gain no net energy from the d.c. bias either. This adds a little support to the primer's main insight. Optional.
5. Electrons go to the collector at one phase and return at another. The patent says that secondaries leaving the screen "when the anode 30 is least positive" go to the collector, and that those leaving "when ... most positive" partly return (p.3) [read]. The primer uses this in its "puzzle" (Step 2c). The arbiter does not use it. No change is needed beyond the arbiter revision below.
6. Formal details, peripheral, for completeness. The inventor's residence is Springfield Township, Montgomery County, Pa. The patent was assigned "by mesne assignments" to Farnsworth Television & Radio Corporation, Dover, Del. Its classification is Cl. 179-171.5. It mentions "other applications now in prosecution" (p.2) [read]. No action needed.

Nothing in the patent was missed that would add a new central claim. claims.md's five claims are an adequate deduplicated set.

4. Accuracy spot checks (against _text/...b4b016.txt)

4.1 The "one to four" reading (the defect)

Full passage, p.3 [read]: "Free electrons in the multiplying space will be attracted by the positive potential on the anode 30, but due to the small area presented by the anode, the greater number of electrons will pass through and continue toward the screen 21, on which the greater number will strike, producing secondaries. Some will pass through, however, the ratio being very nearly that of the apertured area to the total area, or one to four. Those electrons passing through will be further accelerated by the potential on collecting anode 25..."

The subject of "Some will pass through" is the electrons approaching the screen from the helix, not the secondaries. Of those, most strike and make secondaries, and about one in four go through the holes to the collector. The screen specification on p.2 repeats the number as a geometric property: "the ratio of apertured area to total area is of the order of one to four".

• The mathematician read it this way (collimated primaries on a mesh). The contrarian said it is "a statement about secondaries". The arbiter sided with the contrarian and called the mathematician's reading "the less accurate of the two". That is backwards.
• The primer repeats the error in §4 ("the gauze's open-area fraction is equated with the fraction of secondaries that pass through. Secondaries leave the wires in all directions, so this is a rough geometric guess"). Its criticism is therefore aimed at the wrong thing. For electrons arriving roughly radially, geometric transparency is a reasonable first approximation, subject to field-penetration ("lens") effects at the holes.
• facts.json key_claim 6 ("About one fourth of electrons striking the wire-gauze screen pass through it") is close but loosely worded: electrons that strike do not pass. It should say "arriving at".
• The primer §5 Step 3 gain model applies t = 1/4 to the secondaries made per impact (collected = tδI, round trip g_rt = δr). A model that follows the patent applies t to the arriving electrons: per arrival, t is collected, (1 − t) strike and yield δ(1 − t) secondaries, and a fraction r of those return. That gives g_rt = (1 − t)δr and G = t / (1 − (1 − t)δr). The qualitative conclusion does not change: high gain needs g_rt near 1, where the gain becomes very sensitive to δ. The threshold value of r moves from 1/δ = 0.385 to 1/((1 − t)δ) = 0.513 for δ = 2.6, and the table values change.
• No verdict changes. Every reviewer and the arbiter said no verdict depends on this figure.

4.2 Other numeric checks

• example.py reproduces the mathematician's and arbiter's two-plate figures exactly (62.483 V / 31.749 eV at 100 MHz and 1 cm; 15.621 V / 7.937 eV; 0.625 V / 0.317 eV) and the contrarian correction (3.37 ns, 148 MHz instead of 1.69 ns, 297 MHz). The RK4 cross-check lands at 0.50000 periods. Confirmed.
• δ¹⁰ values 1,024, 59,049 and 9,765,625 are confirmed.

5. Process

Source tags. These are present and mostly honest throughout. Problems: - In arbiter.md "Open questions", "the two 2023/2024 Lyden patents" names an inventor that appears nowhere else. It presumably comes from the Google Patents page the arbiter fetched, but this bullet is not tagged. Add [web] with the URL. - arbiter.md "Unread in the folder: the earlier Farnsworth multipactor patents (2,071,515, 2,071,516, 2,091,439)". These patents are not in the folder. They are cited works that were not obtained. The wording should be fixed. - The primer uses a [calc] tag. It is not in the prompt's tag set, but the primer defines it and it is acceptable. - The contrarian says "all 3 text pages". The PDF has 4 pages, and page 1 is the drawing sheet. Trivial.

Web checks. Physicist about 2, mathematician 2, contrarian 3, arbiter 2. All are within budget. The verdict-relevant [memory, medium] items are either web-checked (multipactor attribution, dates, the Halloran article) or explicitly excluded from verdicts ("overtaken by triodes and pentodes" is set aside and not used). The mathematician's ScienceDirect yields (silver δ ≈ 2.6 at 300 eV; Cs oxide ≈ 10) were never re-fetched by the arbiter. They are illustrative only. The primer's citations of Vaughan (1988) and Kishek et al. (1998) are [memory] because the web budget was exhausted, and the primer says so. This is acceptable.

Verdict scale. Applied consistently, with one wobble. The arbiter's argument for claim 2 ("Plausible, untested" is reserved for claims with no real test, and the Halloran report exists, so "Never properly tested") would apply equally to claim 1, yet claim 1 keeps "Plausible, untested" for "this tube". The arbiter should either say why the two differ (Halloran concerns gain, not the bare multiplication) or align them. The difference between the two labels is small and has no practical effect here.

Arbiter errors that the primer corrected but the arbiter still carries: 1. Arbiter §1 accepts the mathematician's inference that, because RF-only impact energies are low, "the DC bias must matter", which in the mathematician's words is "the DC bias on the helix, not the RF, must supply the impact energy". The primer §5 Step 2 shows, with example.py Part 2, that in the patent's screen to helix to screen geometry the d.c. bias contributes zero net impact energy: it sets the timing, and the RF swing supplies the energy (K_imp − K_launch = e(⟨V_h⟩_in − ⟨V_h⟩_out), at most 2eV_rf). The arbiter's text and claim 1 reasoning should record this. 2. Arbiter claim 4 says "build-up time scales as N/f, so higher frequency gives shorter time per trip and a larger relative need for a seed". As the primer's answer 5 notes, N/f build-up gets faster at high f, so this does not explain the patent's remark. The stray-seeds-per-cycle (R/f) argument does. The arbiter should correct its reasoning. The verdict stands.

Primer against STEP 4 and the reader profile. All nine sections are present and in order, with answers at the end. Symbols are defined (e, m, δ, V₀, ω, φ, f, T, d, K, g_rt, t, r, G). Acronyms RF, d.c. and RK4 are defined, but "AM" is used without being expanded ("AM radio"), so expand it. The worked example uses real numbers and is reproduced by example.py. The evidence table follows the required columns. Prerequisites name textbooks without unsafe chapter numbers. The primer is of good quality. Apart from the one-in-four error, its §5 Step 2 is the strongest analysis in the project.

example.py. It runs cleanly with .venv/bin/python (about 1.5 s). Its outputs match every number the primer prints, and its comments explain each step. Part 3(b)/(c) uses the secondary-transmission model that 4.1 finds does not match the patent. It needs an update.

6. Verdict

MINOR. Coverage is complete, the full text was read, and no verdict changes. The defects are one propagated misreading of the source (the "one to four" ratio), two arbiter statements already refuted in the primer but not corrected in the arbiter, one untagged name, one wording error about what is in the folder, and one small verdict-consistency note.

7. Remediation actions

Deep read (optional, cheap; confirms 4.1 and items 3.2 and 3.3 for the record): - US2172152-Radio-Freq-Multipactor-Amplifier-b4b016, pp. 2-3, Sonnet: Which electrons does "Some will pass through ... one to four" refer to? List every operating statement about RF phase, seed sources (ionised particles, light, gun) and the stated preference for the gun. Render the page images if the two-column OCR interleaving is ambiguous.

facts.json - key_claim 6: change it to "About one fourth of the electrons arriving at the screen from the helix pass through its apertures to the collector; the rest strike and release secondaries (the ratio equals the apertured-to-total area)". - Summary: add the patent's stated preference for the gun over light "since it possesses advantages in regard to modulation", and the three listed starting mechanisms (ionised particles, photoemission, gun). - Optional: give how_read in the prescribed vocabulary, e.g. "full text (4 pages, 19.7k chars), no deep read needed".

claims.md - Claim 4: add "(the patent prefers the gun, citing modulation advantages; ionised particles are also listed)". No other change.

arbiter.md: add a "QE revision" section that: 1. Reverses §1's judgment on the one-in-four reading: the mathematician's reading (arriving electrons on a mesh) is the correct one [read: p.3 quote], and the contrarian's correction 1 is wrong. 2. Records the primer/example.py result that the helix d.c. bias contributes no net impact energy in the screen to helix to screen geometry. The bias controls transit timing, and the RF supplies the energy (at most 2eV_rf). This supersedes "the DC bias must matter" as the mathematician meant it. 3. Corrects the claim 4 reasoning: N/f build-up does not explain a high-frequency need for seeds, while seeds per cycle (R/f) does. 4. Adds [web] and the Google Patents URL to the "Lyden" citing-patents mention, and rewrites "Unread in the folder" as "Not in the folder (cited, not obtained)". 5. Explains or removes the claim 1 / claim 2 inconsistency in applying "Plausible, untested" versus "Never properly tested". 6. States that no final verdict changes.

primer.md - §4 table row "Output" and the "One in four" bullet: restate the ratio as applying to electrons arriving from the helix. Replace the criticism with: transmission equal to open area is a fair first approximation for radially arriving electrons, modified by field penetration at the holes. It remains an assertion, with no measurement. - §5 Step 3: re-derive with g_rt = (1 − t)δr and G = t/(1 − (1 − t)δr). Update the threshold (r = 0.513 for δ = 2.6, t = 0.25), the runaway trip counts and the sensitivity table. Keep the conclusion. - §8 Q3 and its answer: recompute under the corrected model. Under the corrected formula G = 5 needs (1 − t)δr = 0.95, so r ≈ 0.487. - §1 or §4: mention that the patent prefers the gun as the seed for modulation reasons. - Define "AM" (amplitude modulation) at first use.

example.py - Part 3(b)/(c): switch to the corrected transmission model to match the revised primer, keeping the comments. Rerun with the venv Python and paste the new numbers into the primer.

Recheck (Sonnet) should confirm all of the above and append a "Recheck" section here.

---

Recheck (Sonnet, 2026-10-03)

Checked against the current facts.json, claims.md, arbiter.md, primer.md, example.py and deep/ read. example.py was run with the venv Python and its Part 3 output (threshold r = 0.513, G = 5 at r = 0.487) matches the primer.

Defects (section 4 and 5) - 4.1 "One to four" misreading: FIXED. facts.json key_claim 6 and summary, claims.md, arbiter (QE revision 1, with [SUPERSEDED] flags in section 1) and primer (section 4 bullet) now say the ratio applies to electrons arriving from the helix. - Verdict-consistency wobble (claim 1 vs claim 2): FIXED. Arbiter QE revision 4 gives claim 1's tube-level part "Never properly tested". Final-verdict line 42 updated. - Arbiter "DC bias must matter" error: FIXED (QE revision 2; superseded flag at line 14). - Arbiter claim 4 N/f reasoning: FIXED (QE revision 3; flag at line 59). - Untagged Lyden citation: FIXED ([web] Google Patents URL at line 88). - "Unread in the folder" wording: FIXED ("Not in the folder (cited, not obtained)", line 89). - Contrarian "all 3 text pages": remains as written in contrarian.md (trivial, acknowledged in arbiter QE revision 5; no action required).

Remediation actions (section 7) - Deep read of pp.2-3 (Sonnet): DONE (deep/US2172152-Radio-Freq-Multipactor-Amplifier-b4b016.md). - facts.json (key_claim 6, gun preference, three seed mechanisms, how_read): FIXED. - claims.md (claim 4 gun preference): FIXED. - arbiter.md "QE revision" section, including the statement that no final verdict changes beyond the claim 1 label: FIXED. - primer.md: section 4 ratio wording FIXED; section 5 Step 3 re-derived with G = t/(1-(1-t)deltar) and threshold 0.513 FIXED; Q3 and its answer (r = 0.487) FIXED; gun preference added FIXED; "AM" defined at first use FIXED. - example.py Part 3(b)/(c) switched to the corrected model and runs cleanly: FIXED.

No defects remain open.

Final: PASS