prodc
Product management as code: personas, flows, and features traced to evidence, so a product's state is read from the repo instead of chased in meetings.
prodc treats product thinking the way requirements-as-code tools (Doorstop, StrictDoc) treat requirements — plain text, in git, with IDs that link to each other — but for the artifacts a product person actually works with: Persona → Flow → Feature → Assumption, each tied to evidence. Two rules give it teeth:
- No evidence link = flagged as opinion. Every "why" must trace to a user signal (feedback, ticket, interview), or it is marked unsupported.
- Status comes from the repo, not from a standup. "Flow UF-004: 3 of 5 scenarios passing", "Feature F-012: blocked, assumption A-003 has no data" — the report answers "where are we" and "what's blocking" without a meeting.
The lineage is Toulmin's argument model (claim + grounds + warrant): a feature is a claim, its evidence is the grounds, its assumptions are the warrant. The same shape safety engineering uses for a "safety case", pointed at product decisions.
What it's for
prodc does not replace the human conversation with a client — it is scaffolding around it. Going in, the board already shows what is still unknown for that client: needs with no evidence (opinions) and needs with no proof (HOLEs), so the meeting is spent closing real gaps instead of re-deriving them. Coming out, what was learned is captured as evidence on the need, so the next decision is planned from the repo rather than from memory. The board answers "where are we" and "what's blocking" between meetings, without a standup.
Prior art
prodc draws on, and tries to credit, several existing lines of work:
- Continuous discovery / opportunity framing — Teresa Torres (opportunity = need; the interview > artifact > internal evidence ladder; "solutions in disguise").
- The argument/assurance model — Toulmin (claim · grounds · warrant) and the safety-case / GSN tradition it informs.
- Spec-driven development — product/requirements modelled as plain JSON/YAML in git (the ThoughtWorks / Fowler SDD writing).
- Structured requirements — EARS (Easy Approach to Requirements Syntax), the NASA requirement-writing guidance, and the SysML lineage.
- Requirements-as-code tooling — Doorstop and StrictDoc (plain-text, git-tracked, linked IDs), which prodc follows in form but points at product artifacts rather than "shall" statements.
Status: v0.3, on PyPI (
pip install prodc). The model grows from three nouns (Need · Source · Proof) to Goal, Release and a GSN argument graph; with the proof/source adapters and thestatus/show/brief/plan/argument/diff/portfoliocommands in. Seedocs/issues/for what's still open.
Commands
prodc -C DIR <command> runs as if started in DIR (like terraform -chdir), so you can
point any command at another project without passing --config paths.
prodc status— the board: needs vs what the suite proves, with gaps, coverage, and per-goal / per-release / cross-client roll-ups.--widenever truncates the need text; non-product needs (product: false) are listed out-of-scope, not counted in coverage.prodc show <id>— one need as a citable answer (grounds, warrant, sources, proofs).prodc brief <client>— a pre-meeting agenda for one client: its applicable needs that are opinions (no evidence) or HOLEs (no proof), must-haves first.prodc plan— a ranked backlog from recorded inputs only (must + state + evidence kind, refined by an optionalimportanceordinal); ordering, not a fabricated score.prodc argument— the GSN assurance graph and its defects (unsupported claim, dangling assumption);--strictexits 1 on any defect.prodc diff <snapshot>— what changed since a savedstatus --json.prodc portfolio <toml>…— many repos rolled into one scoreboard.
Write-back (the agent records decisions, not just reads them) — each validates the whole
model via the loader before keeping the change, shows the diff, and supports --dry-run:
prodc set <id> key=value(orkey+=valuefor proofs/children) — edit one field.prodc capture <id> --kind … [--who --date --quote --locator]— append a source (evidence).prodc accept <id> --who … --date … [--revoke]— record a human acceptance pass.prodc add need <id> --text … --who … --file …— create a need with agent provenance.
Model
Everything loads into one Model from the prodc files ([prodc].needs glob; any file may
carry a needs:, goals:, releases: or argument: list). The nouns:
- Need — what a named person wants. A shared baseline or a per-client variant
(
variant_of+client); the board counts cross-client recurrence. Carries an optional claim and a human attestation (who/date), so a DELIVERED proof whose warrant no one reviewed is flagged rather than trusted blindly. - Source — provenance (interview > artifact > proxy > internal); no source = opinion.
- Proof — a locator resolved by an adapter to DELIVERED / PENDING / BROKEN / UNRESOLVED.
- Goal — a measurable objective needs trace up to; a need with none is flagged.
- Release — an ordered, releasable slice (
releases:order = ship order). - Argument — a GSN graph (claim / strategy / evidence / context / assumption / justification) that cites the nouns above by id, for a checkable assurance case.
Cross-references (variant_of, children, goal, release, argument links) are string
ids validated at load.
Install
uv sync # dependencies
make install-hooks
make check # lint + typecheck + test
Why not just Doorstop / StrictDoc?
Those track "the system shall…" requirements for auditors in regulated industries. prodc tracks personas, flows and features for product decisions, and derives live status from the project's own tests. It reads a repo it is pointed at — including existing Gherkin/BDD and whatever ID scheme already lives there — rather than asking you to re-author everything in a new grammar.
Layout
src/prodc/— the package (typed;py.typed).docs/issues/— follow-ups, as a plain-markdown issue register.Makefile—format/lint/typecheck/test/check/build/docs.AGENTS.md— build/test/style/gotchas for coding agents.
1 maintainer
Metadata
1 maintainer
Release files for prodc 0.3.32
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | |
|---|---|---|---|
| prodc-0.3.32.tar.gz | 48.2 kB | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| prodc-0.3.32-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 88.5 kB
Release files / prodc-0.3.32.tar.gz
| Download URL | prodc-0.3.32.tar.gz |
|---|---|
| Size | 48.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
a02a03cc36f8343aec45368860b90ec7f59b5b2ca326514a411b49465c3a9961
|
|
BLAKE2b-256 checksum How to use checksums |
1d65bb1c0b7435f17490b41b60962488ad675dae8a2336f92c3e10938a9eef50
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.11.19 {"installer":{"name":"uv","version":"0.11.19","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Debian GNU/Linux","version":"12","id":"bookworm","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
Release files / prodc-0.3.32-py3-none-any.whl
| Download URL | prodc-0.3.32-py3-none-any.whl |
|---|---|
| Size | 40.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
17fa77487a5e998601a28753b38261b1446d9dc7d5884cde72d68f2fa81d7e1b
|
|
BLAKE2b-256 checksum How to use checksums |
5b7ef06906d8133be80da954b80fb3b81a67853c9626b78f06760d87c8e12ea5
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
uv/0.11.19 {"installer":{"name":"uv","version":"0.11.19","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Debian GNU/Linux","version":"12","id":"bookworm","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}
|
