What this archive is, what it can and cannot promise, how it is built,
and how to help. If you are new here, this page is the place to start.
What this is
Circuit Codex is an open, structured archive of vintage guitar tube-amplifier
circuits. Every circuit here is independently redrawn from its underlying
facts — component values, topology, published voltage charts — never
scanned or traced from factory drawings, and republished as structured data:
an interactive schematic, a board-layout diagram, a simulation-checked operating
point, a parts list, and a sourced circuit story.
The archive currently documents 10 circuits
(8 verified, 2 draft) with an open-licensed tube SPICE model set.
Everything is free and stays free: circuit data CC-BY-SA 4.0, code MIT,
tube models CC0. The corpus lives in a
public GitHub repository; this site is generated from it on
every change, so the repository — not the site — is the source of truth.
One page per circuit: a pan-and-zoom schematic redrawn in KiCad, a board-layout
diagram with machine-checked wiring, the simulated operating point compared
against the published voltage chart with explicit tolerances, a parts list keyed
to the schematic's reference designators, a circuit story, and the sources every
fact was read from. Lineage links connect derived circuits to their ancestors.
A documented circuit is one moment in a line that ran for decades. The
history tier places each amp in its family — the Champ,
Deluxe, Bassman, Marshall lead and the rest — as a dated, sourced era timeline,
most entries history-tier (placed and cited, circuit not yet drawn) with links into
the documented circuits. The lineage tree draws the whole
corpus as one family tree: succession within each line, documented derivation arrows
between them.
Data pages for every tube the corpus models — basing diagrams, linked period
datasheets, and the SPICE model each simulation uses — plus practical guides
(reading schematics, resistor and capacitor codes, tube pinouts, transformers,
calculators), a glossary, the
high-voltage safety primer, and published
calibration studies that show the reasoning behind model choices.
The corpus itself: per-circuit data files, the tube models, the rendering and
verification pipeline, and the CI gates that everything on this site must pass.
If you want to check how any number on this site was produced, the answer is in
the repository's history.
What “verified” means — and what it doesn't
The ✓ verified badge is earned through a chain of
checks that runs on every change:
Facts read, never traced. Values and chart voltages are read
from published drawings and cited; the schematic is drawn independently.
Simulation vs. the published chart. The circuit's netlist is
simulated in ngspice and the DC operating point must land within stated tolerance
of the printed chart — tube pins within the ±20% measurement convention the era's
charts themselves printed, power rails within tighter internal targets.
Structural cross-checks. The schematic round-trips through
KiCad tooling; the parts list and schematic must agree on every reference
designator, in both directions.
Layout wiring equivalence. The board layout's drawn wiring is
proven electrically equivalent to the simulated netlist — same net structure,
terminal for terminal — by a checker that is itself tested with planted faults.
Human sign-off. A maintainer reviews before any badge is
granted. CI can block a claim; it can never grant one.
Where a printed factory chart is internally impossible — it happens; two circuits
here document cases — the affected node is excluded as
chart disputed with the arithmetic shown, never silently
forced to fit. Circuits that haven't completed the chain carry a
draft badge.
Known limitations
An honest archive states its edges. These are the current ones:
Layouts are connectivity references, not dimensioned templates.
Board and part positions are drawn at diagram scale. Lead dress, grounding paths,
and physical spacing — which genuinely matter in a high-voltage chassis — are not
captured. Parts outside the DC model (control-network internals, the heater chain,
the AC side) are drawn from sources but not machine-checked; the CI report
enumerates every such terminal so the boundary is explicit.
Tube models are DC operating-point fits. They are fitted to
published datasheet anchor points and validated in ngspice for bias-point work.
They are not full AC or distortion models and will not predict tone.
Charts are single published sources. Most circuits have exactly
one printed voltage chart, measured to the era's ±20% convention on
1950s-tolerance components. Real amplifiers of the same model vary.
Simulation checks DC conditions, not sound. A verified
operating point says the circuit's documented numbers are self-consistent — not
how it behaves with signal.
Coverage is small and growing. 10 circuits today,
against hundreds of historically significant designs. Draft circuits
(currently 2) are published early with their status stated rather than
held back.
Built with AI — and audited like it
This archive is researched, drawn, and maintained by AI agents working under human
direction. That is how a corpus this dense gets built with this much cross-checking,
and we state it plainly rather than leave you to wonder.
The engineering assumption is that every author — human or AI — makes
mistakes, so nothing merges on trust. Every change passes the full CI gate
suite; adversarial review passes deliberately try to break the site's claims; and a
human maintainer signs off on every verified badge. When errors are found, they are
fixed in public — the repository history is the audit trail. (Example: the layout
equivalence gate, on its first full run, found thirteen errors — eleven wiring
errors in the drawings and two netlist naming errors — in work that had already
passed both automated and human review. They are fixed, and the commits say
exactly what was wrong.)
Treat this site like any secondary source. The citations are
there — check them. Do not trust any single document, this one included, with your
money, your build, or your safety.
For AI agents working with this project: /llms.txt orients
you to the site, and the repository's
AGENTS.md carries the binding
conventions — the intellectual-property rules, the verification gates, and the
editorial voice. The same rules bind every contributor, human or machine.
How to contribute
The project is community-driven by design and curator-controlled in practice:
anyone can propose, CI and maintainers gate what merges, and only the pipeline
grants badges. Ways in, easiest first:
Report an error. Spot a wrong value, a broken link, a claim
that doesn't match its source? Open an issue
and cite what you checked against. Corrections are the most valuable small
contributions this project gets.
Share bench measurements. Voltages measured on a real,
era-correct amplifier — with the meter, conditions, and mains voltage noted —
become primary data the charts can be cross-referenced against.
Only if you have high-voltage training; read the
safety primer first, and never open a chassis
for our benefit.
Find period sources. Dated schematics, factory bulletins, tube
manuals. Link finds in an issue; the project links to archives and rehosts only
documents whose public-domain status can be established.
Fix or extend a circuit. Small correction PRs (a wrong value, a
missing lineage edge) need only a cited source and a
DCO sign-off.
Add a whole circuit. Copy the template directory, fill in the
data files (metadata, netlist, chart voltages, parts list, schematic, layout),
and open a PR — CI runs every gate described above and a maintainer reviews. New
circuits land as draft; verified is earned, not granted on merge. Full
instructions: CONTRIBUTING.md.
Not sure where to start? Check the
wanted-circuits list.
Three hard rules bind every contribution, stated here because PRs that break them
are closed regardless of quality: redraw from facts (never scan or
trace a factory drawing), circuit-number-first naming (maker names
appear only descriptively), and DCO sign-off on every commit.
Safety, always
⚡ Tube amplifiers carry lethal voltages — several hundred volts DC
that filter capacitors hold long after power-off. Everything on this site can be
studied freely; open a chassis only with
high-voltage safety training, and verify every
connection and every voltage yourself as you work. No document — printed in 1957 or
generated in 2026 — substitutes for your own meter and judgment.