--- slug: cordic passed: 2026-10-14 by: the assay line (research/assay-line/cited-1014/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex checked: 130 confirmed: 83 wrong: 47 unverifiable: 0 observed: 0 stale: 0 --- This was the cordic repair and source-reread pass directed by research/assay-line/cited-1014/cordic/triage.json against 7a9bcb5442. The five layer files matched the checkers' pin 80213b8a47 before editing. The fixer re-read Cochran, Walther pp. 379-381, Lomont pp. 1-3 and 10, AN5325 p. 6, the page recurrences, verifier and README for the edited claims, and retained the other CONFIRMED source rereads with explicit Codex N2 provenance. All 47 triage items and 83 checker-confirmed entries are accounted for; N2:024 is reclassified WRONG because its floor invitation repeats X13, while X26's STALE inference is rejected using the placard's source-date definition. The 21 listed MINOR items remain unchanged and count in wrong; no MINOR correction lines were added. Literal \u2014 and \u007c preserve original punctuation without changing record delimiters. ## Claims - WRONG | X1: "A pocket calculator finds the sine of an angle without ever multiplying two numbers together." | https://www.hpmemoryproject.org/wb_pages/d_cochran_01.htm | Re-read Cochran's accuracy paragraph: the HP-35 sine routine multiplies before and after its iterative core. Replaced the opening and markdown dek with the narrower no-general-hardware-multiplier claim. - MINOR | X2: "It walks a single vector through a fixed ladder of ever-halving angles \u2014 arctan 1, arctan ½, arctan ¼ \u2014 and at each rung it does only three things a chip finds trivial: shift some bits, add, and choose a sign. That is CORDIC." | https://dspguru.com/dsp/faqs/cordic/ | The tangents halve exactly; atan(1/2)/atan(1)=0.590334470601733. The body already says half-ish. Directing triage keeps this precision slip at MINOR, tier C, in sentences separate from the WRONG fixes; left unchanged. - MINOR | X3: "In 1959 Jack Volder needed to compute trigonometry aboard a B-58 bomber, in real time, with 1950s hardware that had no multiplier worth the name \u2014 just shifters and adders." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Walther p. 379 dates Volder's development to 1956; the paper appeared in 1959. Directing triage treats the sentence as a chronology-precision blur, not an explicit first-development claim; tier C, left unchanged. - WRONG | X4: "CORDIC's rotation multiplies by 2⁻ⁱ \u2014 which, in binary, is just \"move the point i places right.\" No multiplier required." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Re-read Walther p. 381, Use of Shifters, and the local recurrence. Binary 1.0 times 2^-1 is 0.1. Preserved the original illustrative quotation and identified its reversed point direction; the explanation now distinguishes rightward digits from leftward point motion. - MINOR | X5: "Every CORDIC machine carries one small read-only table: the arctangents of the negative powers of two." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Walther Table I gives arctangent, linear and inverse-hyperbolic-tangent constants for the three modes. Directing triage keeps the circular-table scope overstatement MINOR; tier C, left unchanged. - MINOR | X6: "the whole angle table, computed once and frozen Σ αᵢ = the reach ↓" | git show 80213b8a47:public/strata/cordic/index.html | The core allocates 64 table entries but renders eight rows; the preceding sentence calls them the first rungs. Directing triage keeps this caption imprecision MINOR; tier C, left unchanged. - MINOR | X7: "After enough steps the tip lands, to full precision, on the unit circle at exactly (cos θ, sin θ)." | git show 80213b8a47:public/strata/cordic/index.html; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | The geometry uses 28 iterations and displays a residual; its finite endpoint is approximate. Directing triage keeps the limiting-endpoint wording MINOR because the displayed error discloses the approximation; tier C, separate from the conserved-path error, left unchanged. - MINOR | X8: "≈ 13 correct bits" | git show 80213b8a47:public/strata/cordic/index.html | At theta=40 degrees and N=14 the maximum error is 6.517392226701446e-5, about 14 bits; the initial fallback says 1.2e-4 and 13 bits. Directing triage keeps this JavaScript-replaced fallback mismatch MINOR; tier C, left unchanged. - MINOR | X9: "the sign choices dᵢ \u2014 turn toward θ (+) or back (−)" | git show 80213b8a47:public/strata/cordic/index.html | The code uses d=(z>=0?1:-1); at a negative target the first negative step turns toward the target. Directing triage keeps the default-example caption ambiguity MINOR; tier C, left unchanged. - MINOR | X10: "Three integer operations per step \u2014 a shift, an add or subtract, and reading one sign bit \u2014 repeated a couple of dozen times. Nothing that deserves to be called a multiply." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:public/strata/cordic/index.html | Each step updates x, y and z; shift, add/subtract and sign selection name operation types, not three individual instructions. Directing triage keeps the loose count MINOR; tier C, left unchanged. - MINOR | X11: "The running product of the stretch factors converges to the total gain A = 1/K; its reciprocal settles onto the constant every CORDIC chip stores." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; https://dspguru.com/dsp/faqs/cordic/ | Walther says initial compensation may be used, and the checker's FAQ reread permits omitting it for phase-only results. Directing triage keeps the universal storage overstatement MINOR while preserving the valid running product; tier C, left unchanged. - MINOR | X12: "Converged, to the last bit a double can hold: K = 0.60725293500888, 1/K = 1.64676025812107." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs | The calculation gives 0.6072529350088813 but the display uses toFixed(14). Directing triage keeps the displayed-precision overstatement MINOR because the rounded constant is valid; tier C, left unchanged. - WRONG | X13: "The angle table and the running arithmetic are themselves only stored to finite precision, so past a certain point new iterations stop buying anything \u2014 the error flattens, and in double precision it even bounces around a floor near 10⁻⁹." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - WRONG | X14: "A logarithmic plot of the CORDIC sine and cosine error against iteration count at a fixed angle. The error falls roughly one bit per iteration along a straight line, then flattens onto a floor near ten to the minus nine and stops improving. Checkpoint values are tabulated below." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - WRONG | X15: "the sloped rows gain ~1 bit each; the last rows sit on the floorrecomputed live" | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - MINOR | X16: "A hardware designer reads this curve backwards: to guarantee b good bits, run about b iterations and store the table to a couple of guard bits beyond that." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | AN5325 p. 6 describes a 24-bit engine with 19-bit result precision and quantisation in both the table and arithmetic. It does not prove the stated universal guarantee. Directing triage keeps this overconfident rule of thumb MINOR; tier C, left unchanged. - WRONG | X17: "The quantity the iteration secretly conserves is x² + m·y²." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X18: "The conserved curve for the chosen geometry \u2014 a circle, a vertical line, or a hyperbola \u2014 with the point walking along it from the start to the computed result. The result and its cross-check against the math library are reported below." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X19: "conserved: x² + y² = 1 (a circle)point → (cos θ, sin θ)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X20: "Both families make transcendentals cheap, and both, at bottom, exploit the logarithm \u2014 but by opposite mechanisms." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; https://www.lomont.org/papers/2003/InvSqrt.pdf | Re-read the circular ATAN table and recurrence, and Lomont pp. 1-3. Circular CORDIC accumulates arctangent rotations. Inverse square root is algebraic (x*y^2=1). Corrected the section heading, comparison and related note without denying hyperbolic CORDIC logarithms. - MINOR | X21: "Nothing here is a stored answer." | git show 80213b8a47:public/strata/cordic/index.html | The page contains the literal gain string 1.6468 although other panels calculate it. Directing triage explicitly keeps the live-recomputation overclaim MINOR; tier C, this standalone sentence is left unchanged. - MINOR | X22: "CORDIC stays optimal specifically where multipliers are scarce: FPGA/ASIC/DSP and dedicated blocks." | https://dspguru.com/dsp/faqs/cordic/ | The checker's FAQ reread supports attractiveness without multipliers, not universal optimality across workloads. Directing triage keeps the comparative overstatement MINOR; tier C, left unchanged. - WRONG | X23: "Past roughly N ≈ 24 in double precision the error stops falling and bounces around ~10⁻⁹; §4 shows this rather than pretending the line drops forever." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - MINOR | X24: "the survey confirming where CORDIC is and is not used today)." | https://eprints.soton.ac.uk/267873/; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | A 2009 survey does not establish an exhaustive 2026 use/non-use account. Directing triage keeps this citation-scope overstatement MINOR, rather than STALE, because no change in the underlying facts was established; tier C, left unchanged. - MINOR | X25: "Every displayed figure is recomputed by this page's live code and independently by research/cordic/verify-cordic.mjs." | git show 80213b8a47:research/cordic/verify-cordic.mjs; git show 80213b8a47:public/strata/cordic/index.html | The verifier checks selected ladder rows and a different gain-proof vector from the live page. Directing triage explicitly keeps the checking overclaim MINOR; tier C, the standalone footer sentence is left unchanged. X33 fixes the separate sentence containing the WRONG attribution. - CONFIRMED | X26: "Last verified 3 July 2026, when it was written." | assay/lib.mjs; scripts/check-placard.mjs; research/self-check/verifier-runs.json | Rejected the STALE inference: assay/lib.mjs defines verified as the date claims were re-read against sources, or the writing date when no claims record exists. scripts/check-placard.mjs uses that dossier date. An August numerical verifier run is not evidence of a later source reread. The new 2026-10-14 claims record now independently warrants regenerating the date, without a Corrected or Updated line. - WRONG | X27: "A pocket calculator finds the sine of an angle without ever multiplying two numbers." | https://www.hpmemoryproject.org/wb_pages/d_cochran_01.htm | Re-read Cochran's accuracy paragraph: the HP-35 sine routine multiplies before and after its iterative core. Replaced the opening and markdown dek with the narrower no-general-hardware-multiplier claim. - MINOR | X28: "It walks one vector through a fixed ladder of ever-halving angles (arctan 1, arctan ½, arctan ¼), and at each rung it does only three things a chip finds trivial: shift some bits, add, and choose a sign." | https://dspguru.com/dsp/faqs/cordic/ | The tangents halve exactly; atan(1/2)/atan(1)=0.590334470601733. The body already says half-ish. Directing triage keeps this precision slip at MINOR, tier C, in sentences separate from the WRONG fixes; left unchanged. - WRONG | X29: "Watch the vector zig-zag onto the unit circle, watch the error fall about one bit per step and then honestly flatten at the floating-point floor, then flip one parameter m to turn the very same machine into a circle, a straight line, or a hyperbola: computing sin, then multiply, then exp and ln (Walther, 1971)." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - WRONG | X30: "Both make a transcendental cheap by exploiting a logarithm \u2014 but by opposite mechanisms." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; https://www.lomont.org/papers/2003/InvSqrt.pdf | Re-read the circular ATAN table and recurrence, and Lomont pp. 1-3. Circular CORDIC accumulates arctangent rotations. Inverse square root is algebraic (x*y^2=1). Corrected the section heading, comparison and related note without denying hyperbolic CORDIC logarithms. - WRONG | X31: "Fast inverse square root is a single one-shot bit reinterpretation: read a float's bits as an integer and log₂ is just there, at a fixed few-percent error." | https://www.lomont.org/papers/2003/InvSqrt.pdf | Re-read Lomont pp. 1, 3 and 10: bit manipulation generates an initial guess, then the code takes a Newton step. The table distinguishes the 3.43756 percent seed error from 0.175228 percent after one refinement. Corrected the related note and clarified the comparison. - WRONG | X32: "For an immersive stratum the body is not what renders \u2014 the real page is the self-contained interactive at the `href` above (`public/strata/cordic/index.html`): a vector that zig-zags onto the unit circle one shift-and-add at a time; a proof, made operable, of why the gain **K = 0.6072529…** is constant no matter the angle (the sign steers direction, never length); an error curve that falls about one bit per iteration and then honestly *flattens* at the double-precision floor; and the flagship \u2014 Walther's 1971 unification, where flipping a single parameter **m ∈ {+1, 0, −1}** turns the identical machine into a circle, a line, or a hyperbola, computing sine, then multiplication, then exp and ln." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - WRONG | X33: "Every number on the page is recomputed live in the browser and independently by `research/cordic/verify-cordic.mjs` (33/33), which also carries the record-corrections: the HP-35 ran a *CORDIC-family* variant, and modern multiplier-rich CPUs use polynomials, not CORDIC." | git show 80213b8a47:research/cordic/verify-cordic.mjs; git show 80213b8a47:research/cordic/README.md | Re-read the complete verifier and README. The historical/CPU qualifications occur in the page and README, not the numerical verifier. Corrected the source attribution and narrowed the every-number claim within this same sentence to selected values and modes. - MINOR | X34: "A calculator finds sine by walking one vector through a fixed ladder of ever-halving angles: each step just a bit-shift, an add, and a sign choice, no multiply." | https://dspguru.com/dsp/faqs/cordic/ | The tangents halve exactly; atan(1/2)/atan(1)=0.590334470601733. The body already says half-ish. Directing triage keeps this precision slip at MINOR, tier C, in sentences separate from the WRONG fixes; left unchanged. - WRONG | X35: "Watch the vector zig-zag onto the unit circle, watch the error fall about one bit per step then flatten at the precision floor, and flip one parameter m to turn the same machine into a circle, a line, or a hyperbola: computing sin, then multiply, then exp and ln (Walther, 1971)." | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - WRONG | X36: "A calculator finds sine by walking a unit vector through a fixed ladder of ever-halving angles \u2014 each step a bit-shift, an add, and a sign choice, with no multiply." | git show 80213b8a47:public/strata/cordic/index.html; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Re-read the structured description and circTrace initialization: x=K_INF, y=0. Recomputed K_INF=0.6072529350088813. Corrected the starting length and the roughly-halving qualification within this same sentence. - MINOR | X37: "CORDIC cos θ / sin θ 0.7660 / 0.6428 from shifts & adds only" | git show 80213b8a47:public/strata/cordic/index.html | The default CORDIC sine is 0.6428527836088063, rendered 0.6429, but the fallback says 0.6428. Directing triage keeps this last-digit fallback mismatch MINOR because JavaScript replaces it; tier C, left unchanged. - MINOR | X38: "\u007cerror\u007c (max of the two) 1.2e-4 ≈ 13 correct bits" | git show 80213b8a47:public/strata/cordic/index.html | At theta=40 degrees and N=14 the maximum error is 6.517392226701446e-5, about 14 bits; the initial fallback says 1.2e-4 and 13 bits. Directing triage keeps this JavaScript-replaced fallback mismatch MINOR; tier C, left unchanged. - MINOR | X39: "length after + step 1.1180 \u007cv\u007c · sᵢ" | git show 80213b8a47:public/strata/cordic/index.html | The proof vector has length 0.62, so its next length at i=1 is about 0.6932; 1.1180 is the scale factor. Directing triage keeps this live-replaced fallback mismatch MINOR; tier C, left unchanged. - MINOR | X40: "length after − step 1.1180 identical → K is constant" | git show 80213b8a47:public/strata/cordic/index.html | The proof vector has length 0.62, so its next length at i=1 is about 0.6932; 1.1180 is the scale factor. Directing triage keeps this live-replaced fallback mismatch MINOR; tier C, left unchanged. - WRONG | X41: "conserved: x² − y² = 1 (a hyperbola)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X42: "exp err vs Math.exp" | git show 80213b8a47:public/strata/cordic/index.html | Re-read drawGeo and evaluated uniTrace at theta=-0.8: cosh error 1.6054886309291305e-9, sinh error 2.4177708723982505e-9, exp error 8.122821304468175e-10. The readout takes their maximum. Corrected its label and kept the arithmetic. - WRONG | X43: "Circular reach is ±99.88°. The point walks the unit circle to (cos θ, sin θ)." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X44: "circle: x²+y² conserved" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X45: "N \u007csin error\u007c bits vs. slope 4 8.50e-3 7 \u2014 8 9.04e-4 10 still gaining 12 2.23e-4 12 still gaining 16 1.32e-5 16 still gaining 20 9.27e-8 23 still gaining 24 1.57e-9 29 still gaining 28 4.13e-9 28 worse \u2014 floor" | git show 80213b8a47:public/strata/cordic/index.html; git show 80213b8a47:research/cordic/verify-cordic.mjs; https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. - WRONG | X46: "conserved: x² + y² = 1 (a circle)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - WRONG | X47: "CORDIC cos θ / sin θ 0.7660 / 0.6428 circle: x²+y² conserved Math.cos / Math.sin 0.7660 / 0.6428 the answer to beat \u007cerror\u007c 4.18e-9 ≈ 28 bits" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. - CONFIRMED | N2:003: "Watch it work, watch why its magic constant never changes, then flip one parameter and turn the same machine into a circle, a straight line, or a hyperbola." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [003 one-parameter-intro] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:006: "The idea is almost impudent: to rotate a vector by some target angle θ, don't rotate it by θ. Rotate it by arctan 1 = 45°. Overshot? Rotate back by arctan ½ ≈ 26.57°. Still not there? Forward by arctan ¼ ≈ 14.04°. Each rung is half-ish the last, so the running total zeroes in on θ like a scale balancing \u2014 a binary search across the arc." | https://dspguru.com/dsp/faqs/cordic/; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [006 angle-example] The FAQ table gives 45, 26.56505 and 14.03624 degrees and describes a "binary search" on phase. Recomputed atan(1), atan(1/2), atan(1/4) agree. The direction example is illustrative; the actual sign is chosen from the residual. - CONFIRMED | N2:007: "The reason those particular angles were chosen \u2014 the arctangents of 2⁻ⁱ \u2014 is the whole trick." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [007 binary-shift-1] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:008: "Rotating by an angle whose tangent is exactly 2⁻ⁱ means the rotation's arithmetic collapses into a bit-shift." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [008 binary-shift-2] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:009: "A true rotation multiplies by a 2×2 matrix full of sines and cosines." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [009 binary-shift-3] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:014: "Add every rung and you get the farthest angle a bare CORDIC can reach: Σ arctan 2⁻ⁱ = 1.743 rad ≈ 99.88°." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [014 bare-range] Walther p. 381 bounds convergence by the sum of rotations. Recomputed the circular sequence starting at i=0: 1.74328662047234 radians, 99.88296583468963 degrees. The printed rounded values agree. - CONFIRMED | N2:015: "Ask for more than that and you must first fold the angle into a quadrant (the calculator does; our instrument stays inside the honest range)." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [015 fold-quadrant] Volder p. 330 explicitly describes a first exact 90-degree rotation; p. 332 gives the full angular range. This supports quadrant reduction, although the record excerpt about pseudo rotations alone does not. The page controls are restricted to -99..99 degrees. - CONFIRMED | N2:022: "Native CORDIC reaches ±99.88°." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [022 native-range-hint-1] Walther p. 381 bounds convergence by the sum of rotations. Recomputed the circular sequence starting at i=0: 1.74328662047234 radians, 99.88296583468963 degrees. The printed rounded values agree. - CONFIRMED | N2:023: "Beyond that a real chip folds by quadrant first." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [023 native-range-hint-2] Volder p. 330 explicitly describes a first exact 90-degree rotation; p. 332 gives the full angular range. This supports quadrant reduction, although the record excerpt about pseudo rotations alone does not. The page controls are restricted to -99..99 degrees. - WRONG | N2:024: "Each step wins about one more bit of accuracy \u2014 until it can't. Drag to feel the floor." | public/strata/cordic/index.html; research/cordic/verify-cordic.mjs | The checker confirmed the general convergence rate, but the drag-to-feel-the-floor clause repeats the false floor claim. The control ends at 28; the same core improves at 40 and 56. Corrected with X13. - CONFIRMED | N2:025: "from shifts & adds only" | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [025 readout-shift-only] Volder equations 6-9 implement the core with shifts and adds. The browser models that arithmetic with multiplication by signed powers of two; its float64 idealisation is explicitly disclosed at HTML line 519. This label is accurate at the algorithm level. - CONFIRMED | N2:030: "A true rotation by angle α keeps the vector's length exactly the same." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [030 pseudo-rotation-2] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:031: "CORDIC's step does not." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [031 pseudo-rotation-3] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:032: "When it \"rotates\" by using tan α = 2⁻ⁱ, it quietly skips the cos α that a real rotation carries \u2014 so each step is a rotation and a stretch." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [032 pseudo-rotation-4] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:033: "This is why it is called a pseudo-rotation." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [033 pseudo-rotation-5] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:034: "xi+1 = xi − d·yi·2⁻ⁱ # d = ±1, the sign choice yi+1 = yi + d·xi·2⁻ⁱ" | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [034 circular-recurrence] Read Volder original pp. 330-331, reprint pp. 226-227. Equations 3-9 show ordinary circular coordinates, the two shifted cross additions, and their radial factor. The excerpt "two simultaneous shift-and-add operations" directly supports the implementation; the surrounding equations, rather than that excerpt alone, support the rotation and stretch claims. - CONFIRMED | N2:035: "\u007cvi+1\u007c² = (x − d·y·2⁻ⁱ)² + (y + d·x·2⁻ⁱ)² = x² − 2·d·x·y·2⁻ⁱ + y²·2⁻²ⁱ + y² + 2·d·x·y·2⁻ⁱ + x²·2⁻²ⁱ = (x² + y²)·(1 + 2⁻²ⁱ) # the two red terms cancel \u2014 for either sign of d" | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [035 norm-identity] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:036: "It appears only in the two red cross-terms \u2014 and those are equal and opposite, so they cancel for d = +1 and for d = −1 alike (all that matters is d² = 1)." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [036 cancellation-2] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:037: "What survives, the stretch factor √(1+2⁻²ⁱ), is set by the shift magnitude i alone." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [037 cancellation-3] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:038: "The sign steers the angle; it has no vote on the length." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [038 cancellation-4] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:039: "So over N steps the vector's length is multiplied by the same product no matter what angle you steered toward \u2014 a fixed number." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [039 fixed-product] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:040: "Its reciprocal is K. Start the vector at length K and the accumulated stretch lands it back at length exactly 1, on the unit circle, where its coordinates are the cosine and sine." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [040 reciprocal-and-unit-circle] Walther p. 381 says "the constant 1/K may be used as an initial value". For a fixed finite sequence, initialising with the reciprocal of that sequence gain cancels the total scale; the infinite-sequence constant is the limit. The final angular value still has the selected finite precision. - CONFIRMED | N2:043: "The stretch is √(1+2⁻²ⁱ). Bigger i → tinier stretch, closer to a true rotation." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [043 stretch-shrinks] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:050: "This single number is the price of admission, paid once, for every trig call the chip will ever make." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [050 constant-paid-once] Walther p. 381 says "the constant 1/K may be used as an initial value". For a fixed finite sequence, initialising with the reciprocal of that sequence gain cancels the total scale; the infinite-sequence constant is the limit. The final angular value still has the selected finite precision. - CONFIRMED | N2:051: "The folklore says CORDIC gains \"about one bit of accuracy per iteration,\" and the sloped part of the curve below shows exactly that: each rung roughly halves the error." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [051 folklore-one-bit] AN5325 p. 6 states "1 binary digit per iteration". This is a convergence-rate description within available arithmetic precision, not strict improvement at every rung. The page quotes anonymous folklore; the source record should use kind=fact for entry 51 because the ST sentence is a paraphrase, not the exact quoted wording. - CONFIRMED | N2:058: "The cost is linear in the number of result bits \u2014 which, in a moment, is exactly why CORDIC lost the general-purpose CPU." | https://dspguru.com/dsp/faqs/cordic/; https://www.intel.com/content/www/us/en/developer/articles/technical/the-difference-between-x87-instructions-and-mathematical-functions.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [058 linear-cost-cpu] The FAQ explicitly says multiplier-equipped implementations using lookup or power series are "generally faster than CORDIC". Intel independently reports replacing CORDIC with polynomial approximations in the 1990s. The basic sequential CORDIC loop adds roughly one result bit per iteration. - CONFIRMED | N2:059: "Everything so far was Volder's 1959 circular CORDIC." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; https://api.crossref.org/works/10.1109/TEC.1959.5222693 | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [059 walther-unification-1] Volder original paper is dated September 1959 and describes circular rotation and vectoring. Crossref agrees on title, author, issue and date. This is a publication-date attribution, unlike entry 4 historical-origin wording. - CONFIRMED | N2:060: "Twelve years later, John Walther at Hewlett-Packard noticed that the identical shift-add-sign iteration \u2014 change nothing but one sign in the update and swap the little angle table \u2014 computes an entire library of functions, not just sine." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [060 walther-unification-2] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:061: "He added a single parameter m:" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [061 walther-unification-3] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:062: "xi+1 = xi − m·d·yi·2⁻ⁱ yi+1 = yi + d·xi·2⁻ⁱ zi+1 = zi − d·αi # α from the table chosen by m" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [062 unified-recurrence] Walther equations 3-8 plus the sign-steering convention give the three displayed recurrences. His sign convention is opposite but equivalent. Delta=±2^-i gives the stated x/y updates and the table supplies alpha_i. - CONFIRMED | N2:064: "Set m = +1 and that is a circle \u2014 sine, cosine, arctangent." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [064 circular-functions] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:065: "Set m = 0 and the conserved shape degenerates to a straight line \u2014 and the machine multiplies and divides." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [065 linear-functions] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:066: "Set m = −1 and it is a hyperbola \u2014 giving exp, ln, √, and the hyperbolic functions." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [066 hyperbolic-functions] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:067: "Same rotations. Same silicon. Flip m and watch the point walk a different curve." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [067 same-silicon] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. This confirms the three function modes and target geometries; the false conserved intermediate-path interpretation is corrected under X17. - CONFIRMED | N2:071: "Iterations 4, 13, 40, 121, … (each 3k+1 after the last) must be repeated, or exp and ln silently drift." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [071 hyperbolic-repeated-rungs] Walther p. 381, Table II footnote lists "4, 13, 40, 121" and the 3k+1 recurrence. The same page explains why the hyperbolic convergence condition needs repeats. Computed reach is 1.1181730118332 radians for the repeated sequence. - CONFIRMED | N2:076: "It was that one tiny state machine, by swapping a table and a sign, spans trigonometry, exponentials, logarithms, multiplication, division and square roots." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [076 unified-library] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:077: "Which is why it never left the places where multipliers are expensive or absent: FPGAs and ASICs, DSP blocks, and dedicated units like the CORDIC coprocessor STMicroelectronics still ships inside its STM32 chips." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf; https://dspguru.com/dsp/faqs/cordic/; https://eprints.soton.ac.uk/267873/ | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [077 fpga-asic-dsp-stm32] AN5325 Rev 8, June 2026, lists current STM32 products and describes the CORDIC accelerator. The Meher author-repository abstract confirms uses in signal/image processing and communications, and the FAQ explicitly identifies FPGA implementation where multipliers are costly. The single ST excerpt covers only the STM32 portion; the wider sources supply the remainder. - CONFIRMED | N2:078: "In 1962 \u2014 three years after Volder \u2014 John Mitchell showed you can multiply and divide by faking a logarithm: because a binary number's position already is roughly its log₂, you can approximate the log by counting and shifting bits, add instead of multiply, then shift back." | https://ntrs.nasa.gov/api/citations/19650021535/downloads/19650021535.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [078 mitchell-1962-method] Opened NASA printed p. 106 (PDF page 122), which reproduces the Mitchell abstract and cites IRE EC-11(4), August 1962, pp. 512-517. Its text says the binary logarithm can be obtained approximately by "simple shifting and counting". This matches the page, including approximate rather than exact multiplication. - CONFIRMED | N2:081: "The bit-log trick is a single, one-shot reinterpretation: read the bits differently and the log is just there, no loop, at the cost of a fixed few percent of error." | https://www.lomont.org/papers/2003/InvSqrt.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [081 bit-log-single-shot-error] Read Lomont pp. 1-3 and p. 10. The bit-level initial estimate has worst relative error "3.43756" percent before Newton refinement. The bit-log description is an approximation involving exponent and mantissa, not an exact logarithm. Entry 81 refers to the initial bit trick, rather than the complete Quake routine. - CONFIRMED | N2:082: "CORDIC is iterative accumulation: it never reinterprets anything; it grinds out one more correct bit per rung by tabulated rotation, as accurate as you are patient." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [082 cordic-no-reinterpretation] AN5325 p. 6 states "1 binary digit per iteration". This is a convergence-rate description within available arithmetic precision, not strict improvement at every rung. The page quotes anonymous folklore; the source record should use kind=fact for entry 51 because the ST sentence is a paraphrase, not the exact quoted wording. - CONFIRMED | N2:083: "One reads a logarithm the hardware already wrote down; the other builds an answer up from rotations." | https://www.lomont.org/papers/2003/InvSqrt.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [083 contrast-reads-log-builds-rotations] Read Lomont pp. 1-3 and p. 10. The bit-level initial estimate has worst relative error "3.43756" percent before Newton refinement. The bit-log description is an approximation involving exponent and mantissa, not an exact logarithm. Entry 81 refers to the initial bit trick, rather than the complete Quake routine. - CONFIRMED | N2:094: "The gain K = 0.6072529350088813 is constant, and the reason is the cancellation shown in §3 \u2014 the per-step stretch √(1+2⁻²ⁱ) depends on the shift i only, never on the sign." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [094 exact-gain-constant] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:095: "Circular CORDIC converges for \u007cθ\u007c ≤ Σ arctan 2⁻ⁱ = 1.7433 rad ≈ 99.88°; the linear range is Σ 2⁻ⁱ = 2; the hyperbolic range is ≈ 1.1182 rad and needs the repeated iterations." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [095 three-convergence-ranges] Recomputed all three sums: circular 1.74328662047234; linear sum from i=0 is 2; repeated hyperbolic sequence approximately 1.1181730118332. Walther Table II uses a linear sequence starting at 1 and consequently lists 1.0. That is not a contradiction, but the record excerpt alone does not establish the page variant with range 2. - CONFIRMED | N2:098: "A real fixed-point CORDIC also carries quantisation error in the stored angle table; the shape of the curve (slope, then floor) is the same, the floor just sits where that word length puts it." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [098 fixed-point-quantisation] AN5325 p. 6 explicitly explains quantisation from the angle table, shifters/adders and input/output registers, and the eventual loss of further convergence. This supports a finite-precision limit in principle, not the separate claim that float64 bottoms out near 1e-9. - CONFIRMED | N2:099: "The HP-35 (1972) ran a CORDIC-family method \u2014 David Cochran's implementation used the Meggitt pseudo-multiplication/pseudo-division variant, and Jacques Laporte's reverse-engineering shows its exp/ln path differs from textbook circular CORDIC." | https://www.hpmemoryproject.org/wb_pages/d_cochran_01.htm; https://archived.hpcalc.org/laporte/Logarithm_1.htm; https://www.hp.com/hpinfo/abouthp/histnfacts/museum/personalsystems/0023/0023history.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [099 hp35-1972-method] Read Cochran and Laporte. Laporte names "D. COCHRAN" and "directly derived from J.E. MEGGITT" for the HP-35 log routine; Cochran describes the iterative pseudo-division/pseudo-multiplication choice. The tool false-negatives disappear after Windows-1252 decoding: all three supplied Laporte excerpts match. These sources support the qualified family attribution, not a verbatim circular implementation. The HP Virtual Museum independently dates the HP-35 to 1972. - CONFIRMED | N2:100: "So: CORDIC-family, yes; the precise rotation this instrument shows, not verbatim." | https://www.hpmemoryproject.org/wb_pages/d_cochran_01.htm; https://archived.hpcalc.org/laporte/Logarithm_1.htm; https://www.hp.com/hpinfo/abouthp/histnfacts/museum/personalsystems/0023/0023history.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [100 hp35-not-verbatim] Read Cochran and Laporte. Laporte names "D. COCHRAN" and "directly derived from J.E. MEGGITT" for the HP-35 log routine; Cochran describes the iterative pseudo-division/pseudo-multiplication choice. The tool false-negatives disappear after Windows-1252 decoding: all three supplied Laporte excerpts match. These sources support the qualified family attribution, not a verbatim circular implementation. The HP Virtual Museum independently dates the HP-35 to 1972. - CONFIRMED | N2:101: "\"HP calculators used CORDIC.\" Nearly, not exactly." | https://www.hpmemoryproject.org/wb_pages/d_cochran_01.htm; https://archived.hpcalc.org/laporte/Logarithm_1.htm; https://www.hp.com/hpinfo/abouthp/histnfacts/museum/personalsystems/0023/0023history.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [101 hp-calculators-qualified] Read Cochran and Laporte. Laporte names "D. COCHRAN" and "directly derived from J.E. MEGGITT" for the HP-35 log routine; Cochran describes the iterative pseudo-division/pseudo-multiplication choice. The tool false-negatives disappear after Windows-1252 decoding: all three supplied Laporte excerpts match. These sources support the qualified family attribution, not a verbatim circular implementation. The HP Virtual Museum independently dates the HP-35 to 1972. - CONFIRMED | N2:102: "\"Modern CPUs compute sine with CORDIC.\" Generally false." | https://www.intel.com/content/www/us/en/developer/articles/technical/the-difference-between-x87-instructions-and-mathematical-functions.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [102 modern-cpus-general-denial] Intel says "In the 1990s Intel replaced the 8087’s CORDIC-based approximations" with polynomial-based approximations. This primary vendor account directly supports the x87 correction and the qualified general statement. - CONFIRMED | N2:103: "Where a fast hardware multiplier exists, argument reduction plus a minimax polynomial (or accurate-table methods) beats CORDIC, because CORDIC costs about one add-shift per result bit." | https://dspguru.com/dsp/faqs/cordic/; https://www.intel.com/content/www/us/en/developer/articles/technical/the-difference-between-x87-instructions-and-mathematical-functions.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [103 hardware-multiplier-methods] The FAQ explicitly says multiplier-equipped implementations using lookup or power series are "generally faster than CORDIC". Intel independently reports replacing CORDIC with polynomial approximations in the 1990s. The basic sequential CORDIC loop adds roughly one result bit per iteration. - CONFIRMED | N2:104: "In particular, do not believe the common claim that Intel's x87 FSIN \"uses CORDIC\" \u2014 Intel's transcendentals are table/polynomial." | https://www.intel.com/content/www/us/en/developer/articles/technical/the-difference-between-x87-instructions-and-mathematical-functions.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [104 x87-polynomials] Intel says "In the 1990s Intel replaced the 8087’s CORDIC-based approximations" with polynomial-based approximations. This primary vendor account directly supports the x87 correction and the qualified general statement. - CONFIRMED | N2:107: "\"More iterations, more accuracy.\" Only up to the floor." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [107 more-iterations-qualified] AN5325 p. 6 explicitly explains quantisation from the angle table, shifters/adders and input/output registers, and the eventual loss of further convergence. This supports a finite-precision limit in principle, not the separate claim that float64 bottoms out near 1e-9. - CONFIRMED | N2:108: "J. E. Volder, \"The CORDIC Trigonometric Computing Technique,\" IRE Transactions on Electronic Computers, EC-8(3):330–334, Sept 1959" | https://api.crossref.org/works/10.1109/TEC.1959.5222693; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [108 citation-volder] Crossref and the original scan agree: Jack E. Volder; September 1959; EC-8(3); pages 330-334; the cited title and IRE journal. - CONFIRMED | N2:109: "J. S. Walther, \"A unified algorithm for elementary functions,\" Proc. AFIPS Spring Joint Computer Conference, 1971, pp. 379–385" | https://api.crossref.org/works/10.1145/1478786.1478840; https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [109 citation-walther] Crossref confirms J. S. Walther, the title, 1971 conference, and first page 379. The seven-page paper scan supplies final page 385; the source-record Crossref excerpt alone does not contain the ending page. - CONFIRMED | N2:110: "J. N. Mitchell, \"Computer Multiplication and Division Using Binary Logarithms,\" IRE Transactions on Electronic Computers, EC-11(4):512–517, 1962" | https://api.crossref.org/works/10.1109/TEC.1962.5219391; https://ntrs.nasa.gov/api/citations/19650021535/downloads/19650021535.pdf; https://ethw-images.s3.us-east-va.perf.cloud.ovh.us/ethw/c/c1/2011-04-liu.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [110 citation-mitchell] Crossref confirms John N. Mitchell, the title, EC-11(4), 512-517, August 1962. Its container-title says IEEE, but NASA printed p. 106 and the IEEE journal history establish the contemporary IRE name. The page is right; the Crossref field must not be copied uncritically. - CONFIRMED | N2:111: "P. K. Meher et al., \"50 Years of CORDIC,\" IEEE Trans. Circuits & Systems I, 56(9):1893–1907, 2009" | https://api.crossref.org/works/10.1109/TCSI.2009.2025803; https://eprints.soton.ac.uk/267873/ | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [111 citation-meher] Crossref and the author repository agree on Meher et al., 2009, 56(9), 1893-1907. The page uses the recognizable short title; the full title adds Algorithms, Architectures, and Applications. - CONFIRMED | N2:112: "the original algorithm, built to replace the B-58's analog resolver)." | https://link.springer.com/article/10.1023/A:1008110704586; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [112 original-b58-resolver] Volder autobiography abstract directly identifies replacement of the B-58 analog navigation computer. His 1959 paper describes the digital equivalent of an analog resolver. This supports the stated application. - CONFIRMED | N2:113: "the m ∈ {+1, 0, −1} unification)." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [113 walther-m-parenthesis] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:114: "note the journal was IRE, not IEEE, until 1963)." | https://ethw-images.s3.us-east-va.perf.cloud.ovh.us/ethw/c/c1/2011-04-liu.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [114 ire-until-1963] Read the IEEE Transactions on Computers April 2011 editorial, printed p. 449. It says the journal became IEEE Transactions on Electronic Computers in August 1963. This independently resolves the misleading retrospective Crossref container name for Mitchell. - CONFIRMED | N2:116: "HP-35 attribution after D. S. Cochran and J. Laporte's reverse-engineering." | https://www.hpmemoryproject.org/wb_pages/d_cochran_01.htm; https://archived.hpcalc.org/laporte/Logarithm_1.htm; https://www.hp.com/hpinfo/abouthp/histnfacts/museum/personalsystems/0023/0023history.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [116 hp-attribution-reference] Read Cochran and Laporte. Laporte names "D. COCHRAN" and "directly derived from J.E. MEGGITT" for the HP-35 log routine; Cochran describes the iterative pseudo-division/pseudo-multiplication choice. The tool false-negatives disappear after Windows-1252 decoding: all three supplied Laporte excerpts match. These sources support the qualified family attribution, not a verbatim circular implementation. The HP Virtual Museum independently dates the HP-35 to 1972. - CONFIRMED | N2:125: "That is CORDIC (Volder, 1959), and its secret is a constant, K = 0.6072529…, that never changes no matter which angle you ask for, because every pseudo-rotation stretches the vector by the same factor √(1+2⁻²ⁱ), fixed by the shift alone and blind to the sign that steers the angle." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [125 dek-3] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:130: "CORDIC never reinterprets anything; it grinds out one correct bit per rung by tabulated rotation, as accurate as you are patient." | https://www.st.com.cn/resource/en/application_note/an5325-getting-started-with-the-cordic-accelerator-using-stm32cubeg4-mcu-package-stmicroelectronics.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [130 related-1-4] AN5325 p. 6 states "1 binary digit per iteration". This is a convergence-rate description within available arithmetic precision, not strict improvement at every rung. The page quotes anonymous folklore; the source record should use kind=fact for entry 51 because the ST sentence is a paraphrase, not the exact quoted wording. - CONFIRMED | N2:132: "Peaucellier's linkage turns pure rotation into an exact straight line through bars and pivots; CORDIC turns a ladder of fixed rotations into a sine through shifts and adds." | https://webhomes.maths.ed.ac.uk/~csangwin/howroundcom/straightline/exact.html; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [132 related-2-2] Read the cited mathematical construction, including the Peaucellier linkage and inversion argument. It states "the geometry guarantees an exact straight line movement". Volder provides the separate shift/add CORDIC mechanism. - CONFIRMED | N2:133: "The Antikythera mechanism computes the irregular motions of the heavens by stacking gear ratios \u2014 rational approximations to transcendental cycles, assembled from the one primitive bronze allows." | https://www.nature.com/articles/nature05357 | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [133 related-3-1] Read the Nature primary-research abstract: reconstruction finds gear implementation of Hipparchos lunar-motion theory. This supports the mechanical representation of irregular lunar motion and rational gear ratios. The bronze/primitive phrase is metaphor, not a literal materials-science exclusivity claim. - CONFIRMED | N2:134: "CORDIC computes transcendental functions by stacking pseudo-rotations \u2014 the one primitive a multiplier-less chip allows." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [134 related-3-2] Volder describes pseudo rotations implemented with shift registers and add/subtract units. This supports the factual mechanism; the exclusive primitive wording is figurative in this comparison. - CONFIRMED | N2:135: "There, a sum of spinning epicycles traces any closed curve (the Fourier series made mechanical); here, a sum of ever-smaller rotations lands a vector exactly on the unit circle, and its coordinates are the cosine and sine." | https://in.mathworks.com/matlabcentral/fileexchange/184897-fourier-epicycles-studio-draw-anything-rebuilt-from-circles; https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [135 related-4-2] Opened the named MathWorks implementation description and verified its Fourier/rotating-circle statement. For ordinary parametrised closed curves this is Fourier approximation, exact in the appropriate infinite-series limit. Circular CORDIC likewise has an ideal limiting endpoint; a finite browser computation has the explicitly displayed residual. - CONFIRMED | N2:139: "That is CORDIC (Volder, 1959)." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; https://api.crossref.org/works/10.1109/TEC.1959.5222693 | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [139 meta-description-2] Volder original paper is dated September 1959 and describes circular rotation and vectoring. Crossref agrees on title, author, issue and date. This is a publication-date attribution, unlike entry 4 historical-origin wording. - CONFIRMED | N2:140: "Its secret is a constant, K = 0.6072529…, that never changes with the angle, because every pseudo-rotation stretches the vector by the same factor √(1+2⁻²ⁱ), set by the shift alone and blind to the sign that steers the angle." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [140 meta-description-3] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:143: "How a calculator computes sine with no multiplier: shift, add, choose a sign." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [143 meta-og-description-1] Volder supports a dedicated arithmetic unit built from shift registers and add/subtractors. These descriptions promise no hardware multiplier, which is distinct from the opening statement that a complete calculator routine never multiplies two numbers. - CONFIRMED | N2:144: "Watch the vector zig-zag onto the unit circle, watch why the gain K never changes, then flip one parameter to compute exp and ln on the same machine." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [144 meta-og-description-2] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:146: "A calculator computes sine with no multiplier: just shift, add, and a sign choice." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [146 meta-twitter-description-1] Volder supports a dedicated arithmetic unit built from shift registers and add/subtractors. These descriptions promise no hardware multiplier, which is distinct from the opening statement that a complete calculator routine never multiplies two numbers. - CONFIRMED | N2:149: "That is CORDIC (Volder, 1959)." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; https://api.crossref.org/works/10.1109/TEC.1959.5222693 | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [149 structured-description-2] Volder original paper is dated September 1959 and describes circular rotation and vectoring. Crossref agrees on title, author, issue and date. This is a publication-date attribution, unlike entry 4 historical-origin wording. - CONFIRMED | N2:150: "Its gain K = 0.6072529… is constant because every pseudo-rotation scales the vector by √(1+2⁻²ⁱ), fixed by the shift and blind to the sign." | https://archived.hpcalc.org/laporte/Volder_CORDIC.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [150 structured-description-3] Expanded the two squares independently: the cross terms cancel and d^2=1, giving (x^2+y^2)(1+2^(-2i)). Volder p. 331 says "either of the two choices of direction produces the same change in magnitude". The N-step product is sign-independent. Computed K=0.6072529350088813, matching the exact-gain claim and shorter rounded versions. - CONFIRMED | N2:151: "Flip one parameter m and the same machine computes multiply, exp and ln (Walther, 1971)." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [151 structured-description-4] Read Walther pp. 379-381, including equations 3-8, Table I, Figure 2 and equations 17-21. The summary names "multiplication, division, sin, cos, tan, arctan, sinh, cosh, tanh, arctanh, ln, exp and square-root". It defines circular, linear and hyperbolic systems by m. The record summary alone is insufficient for the mechanics; the tables supply them. Hyperbolic shift repeats and different initial values are also needed, as the page later acknowledges. - CONFIRMED | N2:163: "conserved: x = a (a straight line)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [163 script-caption-162] For m=0 the recurrence leaves x=a and updates y by signed a*2^-i while z approaches zero from b. Hence y approaches a*b; generalized radius x is unchanged; the sum of shifts starting at zero gives \u007cb\u007c<=2. The page and default linear results reproduce these properties. - CONFIRMED | N2:165: "point → (cosh θ, sinh θ)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [165 script-caption-164] Walther Figure 2 gives the gain-compensated limiting coordinates (cos,sin) and (cosh,sinh). The arrows describe those target endpoints, with finite residuals explicitly displayed beside them; this does not validate the separate conserved-trajectory captions. - CONFIRMED | N2:168: "x is held fixed at a." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [168 script-caption-167] For m=0 the recurrence leaves x=a and updates y by signed a*2^-i while z approaches zero from b. Hence y approaches a*b; generalized radius x is unchanged; the sum of shifts starting at zero gives \u007cb\u007c<=2. The page and default linear results reproduce these properties. - CONFIRMED | N2:169: "z is driven from b to 0; y climbs to a·b. Linear reach \u007cb\u007c ≤ 2." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [169 script-caption-168] For m=0 the recurrence leaves x=a and updates y by signed a*2^-i while z approaches zero from b. Hence y approaches a*b; generalized radius x is unchanged; the sum of shifts starting at zero gives \u007cb\u007c<=2. The page and default linear results reproduce these properties. - CONFIRMED | N2:170: "Hyperbolic reach is ±1.1182 rad; iterations 4,13,40,121 repeat." | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:research/cordic/verify-cordic.mjs | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [170 script-caption-169] Walther p. 381, Table II footnote lists "4, 13, 40, 121" and the 3k+1 recurrence. The same page explains why the hyperbolic convergence condition needs repeats. Computed reach is 1.1181730118332 radians for the repeated sequence. - CONFIRMED | N2:172: "line: x held constant" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [172 script-caption-171] For m=0 the recurrence leaves x=a and updates y by signed a*2^-i while z approaches zero from b. Hence y approaches a*b; generalized radius x is unchanged; the sum of shifts starting at zero gives \u007cb\u007c<=2. The page and default linear results reproduce these properties. - CONFIRMED | N2:173: "point → (a, a·b)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [173 script-caption-172] For m=0 the recurrence leaves x=a and updates y by signed a*2^-i while z approaches zero from b. Hence y approaches a*b; generalized radius x is unchanged; the sum of shifts starting at zero gives \u007cb\u007c<=2. The page and default linear results reproduce these properties. - CONFIRMED | N2:174: "linear · m = 0 · gain = 1.0000 (no stretch) · reach \u007cb\u007c ≤ 2" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [174 script-caption-173] For m=0 the recurrence leaves x=a and updates y by signed a*2^-i while z approaches zero from b. Hence y approaches a*b; generalized radius x is unchanged; the sum of shifts starting at zero gives \u007cb\u007c<=2. The page and default linear results reproduce these properties. - CONFIRMED | N2:180: "point → (cos θ, sin θ)" | https://archived.hpcalc.org/laporte/Welther-Unified%20Algorithm.pdf; git show 80213b8a47:public/strata/cordic/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [180 live-circular-geo-cap-r] Walther Figure 2 gives the gain-compensated limiting coordinates (cos,sin) and (cosh,sinh). The arrows describe those target endpoints, with finite residuals explicitly displayed beside them; this does not validate the separate conserved-trajectory captions. - CONFIRMED | N2:194: (added) "The alphabet end of the spine, and the one member that cannot use the page's own frontier." | git show 80213b8a47:public/strata/filed-before-the-question/index.html | Codex N2 source reread retained from research/assay-line/cited-1014/cordic/N2-codex.md: [194 added-nearby-frontier] At the pin, the linked portal classifies CORDIC as an alphabet at lines 274-287 and says its Newton table/constant frontier is "simply not available to it" at lines 356-361. This is true within the portal stated no-multiplier comparison. Positive control for the omission search: the source record contains "gain 1/K = 1.6468" and "Nothing here is a stored answer." but no "The alphabet end of the spine". Add the missing navigation-caption entry. ## Findings - [fixed] X1 (calculator-never-multiplies): Re-read Cochran's accuracy paragraph: the HP-35 sine routine multiplies before and after its iterative core. Replaced the opening and markdown dek with the narrower no-general-hardware-multiplier claim. The matching Corrected 2026-10-14 line appears in the page and markdown body. The matching text in public/og/cordic.png and public/og-thumbs/cordic.webp comes from that dek; share card rebuilt at landing from the corrected dek. - [declined] X2 (angle-ladder-exact-halving): The tangents halve exactly; atan(1/2)/atan(1)=0.590334470601733. The body already says half-ish. Directing triage keeps this precision slip at MINOR, tier C, in sentences separate from the WRONG fixes; left unchanged. - [declined] X3 (development-publication-date-blur): Walther p. 379 dates Volder's development to 1956; the paper appeared in 1959. Directing triage treats the sentence as a chronology-precision blur, not an explicit first-development claim; tier C, left unchanged. - [fixed] X4 (binary-point-direction): Re-read Walther p. 381, Use of Shifters, and the local recurrence. Binary 1.0 times 2^-1 is 0.1. Preserved the original illustrative quotation and identified its reversed point direction; the explanation now distinguishes rightward digits from leftward point motion. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [declined] X5 (universal-arctangent-table): Walther Table I gives arctangent, linear and inverse-hyperbolic-tangent constants for the three modes. Directing triage keeps the circular-table scope overstatement MINOR; tier C, left unchanged. - [declined] X6 (partial-angle-table-called-whole): The core allocates 64 table entries but renders eight rows; the preceding sentence calls them the first rungs. Directing triage keeps this caption imprecision MINOR; tier C, left unchanged. - [declined] X7 (finite-endpoint-exactness): The geometry uses 28 iterations and displays a residual; its finite endpoint is approximate. Directing triage keeps the limiting-endpoint wording MINOR because the displayed error discloses the approximation; tier C, separate from the conserved-path error, left unchanged. - [declined] X8 (fallback-error-readout): At theta=40 degrees and N=14 the maximum error is 6.517392226701446e-5, about 14 bits; the initial fallback says 1.2e-4 and 13 bits. Directing triage keeps this JavaScript-replaced fallback mismatch MINOR; tier C, left unchanged. - [declined] X9 (sign-direction-caption): The code uses d=(z>=0?1:-1); at a negative target the first negative step turns toward the target. Directing triage keeps the default-example caption ambiguity MINOR; tier C, left unchanged. - [declined] X10 (operation-types-as-instruction-count): Each step updates x, y and z; shift, add/subtract and sign selection name operation types, not three individual instructions. Directing triage keeps the loose count MINOR; tier C, left unchanged. - [declined] X11 (universal-gain-storage): Walther says initial compensation may be used, and the checker's FAQ reread permits omitting it for phase-only results. Directing triage keeps the universal storage overstatement MINOR while preserving the valid running product; tier C, left unchanged. - [declined] X12 (displayed-gain-last-bit): The calculation gives 0.6072529350088813 but the display uses toFixed(14). Directing triage keeps the displayed-precision overstatement MINOR because the rounded constant is valid; tier C, left unchanged. - [fixed] X13 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X14 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X15 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [declined] X16 (guard-bit-guarantee): AN5325 p. 6 describes a 24-bit engine with 19-bit result precision and quantisation in both the table and arithmetic. It does not prove the stated universal guarantee. Directing triage keeps this overconfident rule of thumb MINOR; tier C, left unchanged. - [fixed] X17 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X18 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X19 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X20 (logarithm-transcendental-conflation): Re-read the circular ATAN table and recurrence, and Lomont pp. 1-3. Circular CORDIC accumulates arctangent rotations. Inverse square root is algebraic (x*y^2=1). Corrected the section heading, comparison and related note without denying hyperbolic CORDIC logarithms. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [declined] X21 (nothing-stored-overclaim): The page contains the literal gain string 1.6468 although other panels calculate it. Directing triage explicitly keeps the live-recomputation overclaim MINOR; tier C, this standalone sentence is left unchanged. - [declined] X22 (universal-hardware-optimality): The checker's FAQ reread supports attractiveness without multipliers, not universal optimality across workloads. Directing triage keeps the comparative overstatement MINOR; tier C, left unchanged. - [fixed] X23 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [declined] X24 (survey-current-scope): A 2009 survey does not establish an exhaustive 2026 use/non-use account. Directing triage keeps this citation-scope overstatement MINOR, rather than STALE, because no change in the underlying facts was established; tier C, left unchanged. - [declined] X25 (every-figure-recomputed): The verifier checks selected ladder rows and a different gain-proof vector from the live page. Directing triage explicitly keeps the checking overclaim MINOR; tier C, the standalone footer sentence is left unchanged. X33 fixes the separate sentence containing the WRONG attribution. - [fixed] X27 (calculator-never-multiplies): Re-read Cochran's accuracy paragraph: the HP-35 sine routine multiplies before and after its iterative core. Replaced the opening and markdown dek with the narrower no-general-hardware-multiplier claim. The matching Corrected 2026-10-14 line appears in the page and markdown body. The matching text in public/og/cordic.png and public/og-thumbs/cordic.webp comes from that dek; share card rebuilt at landing from the corrected dek. - [declined] X28 (angle-ladder-exact-halving): The tangents halve exactly; atan(1/2)/atan(1)=0.590334470601733. The body already says half-ish. Directing triage keeps this precision slip at MINOR, tier C, in sentences separate from the WRONG fixes; left unchanged. - [fixed] X29 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X30 (logarithm-transcendental-conflation): Re-read the circular ATAN table and recurrence, and Lomont pp. 1-3. Circular CORDIC accumulates arctangent rotations. Inverse square root is algebraic (x*y^2=1). Corrected the section heading, comparison and related note without denying hyperbolic CORDIC logarithms. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X31 (inverse-sqrt-seed-as-full-algorithm): Re-read Lomont pp. 1, 3 and 10: bit manipulation generates an initial guess, then the code takes a Newton step. The table distinguishes the 3.43756 percent seed error from 0.175228 percent after one refinement. Corrected the related note and clarified the comparison. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X32 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X33 (verifier-corrections-misattributed): Re-read the complete verifier and README. The historical/CPU qualifications occur in the page and README, not the numerical verifier. Corrected the source attribution and narrowed the every-number claim within this same sentence to selected values and modes. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [declined] X34 (angle-ladder-exact-halving): The tangents halve exactly; atan(1/2)/atan(1)=0.590334470601733. The body already says half-ish. Directing triage keeps this precision slip at MINOR, tier C, in sentences separate from the WRONG fixes; left unchanged. - [fixed] X35 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X36 (unit-starting-vector): Re-read the structured description and circTrace initialization: x=K_INF, y=0. Recomputed K_INF=0.6072529350088813. Corrected the starting length and the roughly-halving qualification within this same sentence. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [declined] X37 (fallback-cordic-sine): The default CORDIC sine is 0.6428527836088063, rendered 0.6429, but the fallback says 0.6428. Directing triage keeps this last-digit fallback mismatch MINOR because JavaScript replaces it; tier C, left unchanged. - [declined] X38 (fallback-error-readout): At theta=40 degrees and N=14 the maximum error is 6.517392226701446e-5, about 14 bits; the initial fallback says 1.2e-4 and 13 bits. Directing triage keeps this JavaScript-replaced fallback mismatch MINOR; tier C, left unchanged. - [declined] X39 (fallback-scale-as-length): The proof vector has length 0.62, so its next length at i=1 is about 0.6932; 1.1180 is the scale factor. Directing triage keeps this live-replaced fallback mismatch MINOR; tier C, left unchanged. - [declined] X40 (fallback-scale-as-length): The proof vector has length 0.62, so its next length at i=1 is about 0.6932; 1.1180 is the scale factor. Directing triage keeps this live-replaced fallback mismatch MINOR; tier C, left unchanged. - [fixed] X41 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X42 (hyperbolic-error-mislabel): Re-read drawGeo and evaluated uniTrace at theta=-0.8: cosh error 1.6054886309291305e-9, sinh error 2.4177708723982505e-9, exp error 8.122821304468175e-10. The readout takes their maximum. Corrected its label and kept the arithmetic. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X43 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X44 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X45 (false-double-precision-floor): Re-ran the actual circTrace core at theta=0.7: sine errors against Math.sin are 1.56621604574525e-9 (24 steps), 4.132300834847058e-9 (28), 1.0726974863928262e-12 (40), and 1.1102230246251565e-16 (56). Kept the 28-step plot and numbers; corrected prose, metadata, accessibility text, labels, README and verifier to describe a rebound followed by further improvement. AN5325 p. 6 concerns a different, 24-bit engine. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X46 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] X47 (pseudo-rotation-conserved-curve): Re-read Walther p. 379 equations 3-8 and uniTrace. Expanding the displayed update gives Q_next=(1+m*2^(-2i))*Q. Circular Q starts 0.36875612707690064 then 0.7375122541538013; hyperbolic Q starts 1.4580491686564172 then 1.0935368764923128. Corrected circular/hyperbolic target descriptions and captions, retaining the true linear x=a conservation. The matching Corrected 2026-10-14 line appears in the page and markdown body. - [fixed] N2:024 (copy of false-double-precision-floor): changed the slider hint from inviting readers to feel a floor to comparing iteration counts. This is another copy of X13 and shares its dated correction.