90b6f02bfa
docs/variational.md describes each call on every surface, every block of it run by the documentation check: a job's parameters, the gradient methods and what the rounding bound counts, the optimisers and the digest chain, the variational eigensolver, QAOA, and the builders with their references. The crate, command-line and service READMEs (and their translations) gain the library, the feature matrix says what a float or angle input now is, the agent docs are regenerated, and wai/CHANGELOG.md acknowledges the one breaking change, engine::schema::FILES growing to 23. Co-Authored-By: Claude Opus 5.5 <[email protected]>
5.2 KiB
WAI Quantum documentation
Every command and code snippet on these pages runs in CI, and its output is checked against the real output or its receipt (below). A page that cannot run one says why.
Start here
- Learn: the circuits course, from a first circuit
through superposition, entanglement, interference, the standard algorithms,
noise and its mitigation, error correction and fault tolerance to resource
estimates. Each exercise is checked by the engine. It runs in the Studio and
on the command line (
wai-quantum learn), in English and in Simplified Chinese. Beside it, Quantum, reconsidered, a seven-module course on interference and learning, runs in the browser at https://wai.transaction.science/quantum-learn. - The command line:
wai-quantum: run, check, draw, compile and estimate programs, verify receipts, and serve the tools to agents. - In the browser and on the desktop: the Studio and the desktop app.
- From Python:
wai_quantum.
The engine and its reference pages
- The crate: the simulator, the operational layer, the research tools and the receipts.
- Verdicts: how far the engine checked an answer, the intents and the repairs.
- The compiler: the passes, the ε budget and the certificate.
- Resource estimates: every number of an estimate and what it assumes.
- The reader's diagnostics and what it does with each construct.
- The job layer's refusal codes.
- Other formats: QIR under its profiles and jeff, read and written, each conversion checked.
- Error correction and mitigation: codes with their certified distances, the decoders, thresholds, and mitigation in a job.
- Pulse and dynamics: pulse programs read into schedules on frames, integrated under a master equation with their error bounded and stated, the same bits everywhere, and what their receipts bind.
- Variational algorithms and the algorithm library: a job's parameters, exact gradients with their rounding bound, the optimisers, the variational eigensolver, QAOA, and programs for phase and amplitude estimation, an adder and QSVT, each checked against a reference.
Other surfaces
- The job service, over HTTP and MCP: running it yourself, as a batch scheduler's resource, and what a result's provenance records.
- Views inside agent hosts: the engine inside them.
- The answer view layer, a draft extension: its specification.
- Editors: the language server, the debugger and the formatter.
- Your own hardware and the provider crate.
- The workbench's specifications.
- For agents: the Agent Skill
and
llms.txt, generated from the servers' own tool lists.
Changes between versions
- The changelog:
wai/CHANGELOG.md. - What is stable, for how long, and how a change is announced, the schemas of every document included: the version policy.
- Each item deprecated, the version that deprecated it and its replacement: the deprecations page.
In other languages
Simplified Chinese: this index, the
command line's README, the
Studio's, the
Python package's, the
crate's, the course, the Studio's interface, and the
command line's own words (--lang zh-CN, or as LANG says). Each is a
machine-assisted translation, marked so, to be reviewed by a fluent reader.
How these pages are checked
wai/tools/quantum-docs/check_docs.py reads every page named in its
docs.json. It runs each shell block marked runnable in a scratch
directory with the command line built from this checkout, runs the Python
blocks against the package and builds the Rust ones, and holds each shown
output to the real one. It verifies every receipt a block writes with
wai-quantum verify. A block that cannot run here (one that needs an
account, a server that does not return, or data the repository does not
hold) says so in its info string, and its commands must still name commands
and flags that exist. Every wai-quantum … in the prose must name a real
command, every relative link must reach a file, and every dependency a page
shows must be one this repository's version satisfies. A translation must
hold the same code as its English page. .github/workflows/wai-quantum-docs.yml
runs it, with the course's tests and the locale checks.