Skip to contribution map
20 Watts Was Enough
Menu

Contribution map · generated from canonical Git source

Help one bounded part move forward

Pick a workstream, preserve its authority boundary, and attach the evidence another contributor needs to review or rerun the work.

The project needs bounded corrections, reviews and test reports more than broad offers to “help with anything”. Choose one workstream below, name the exact source or artifact you will touch, and define what would make the contribution finished. An issue coordinates the work; the accepted authority remains the reviewed file on Git main.

Difficulty here means how much project context the task requires. It is not a status barrier. A new reader who can identify the sentence where an argument became opaque may provide more useful evidence than a large rewrite.

The research roadmap defines six dependency stages. The GitHub milestone view shows the issues and pull requests currently attached to each stage. Use it to see what is next and what has closed, but read a closed issue only as completed work: evidence status changes through the claim ledger and qualifying result path, not through the milestone percentage.

Choose live work before opening another issue

The workstream forms below are for a new, reproducible gap. To continue work that is already bounded, start with the live queues:

Claim the smallest suitable issue with a comment before doing substantial work. If its evidence, authority boundary, or completion condition is unclear, ask there instead of opening a parallel task.

Current workstreams

WorkstreamUseful first contributionEntry contextAuthority boundaryStart here
Readability and PagesName one broken route, inaccessible control, dense passage or missing definition and explain where your interpretation divergedLow; no local setup required for a reportPresentation may change; evidence status and scientific meaning may notSite or documentation form
Reviewed translationReview one source-bound document, settle domain terminology, or correct a stale translationMedium; fluent target-language review and enough domain context to check qualificationsEnglish Git source remains canonical; machine output alone is not publishableTranslation form
Release-image testingAfter a passing v0.3.0 or later release publishes Fixture 007 and Fixture 019, run one on a Linux amd64 host and report a minimal reproducible runtime, mount or receipt defectLow to medium; Docker and the image instructionsA passing container run remains NO_RESULT; it is not confirmation or an energy resultShort failed-run form
Experiment contracts and runnersClose one named protocol ambiguity, failure path, bound, comparator or test gapHigh; read the fixture, linked claims, manifest and nearest AGENTS.mdDevelopment and confirmation identities stay disjoint; the manifest and claim ledger control readinessExperiment form
Evidence correctionCheck one claim against a primary or authoritative source and state the exact supported scopeMedium to high; source and methods literacy in the affected fieldSources support only what they tested; citation does not grant redistribution rightsEvidence correction form
Mechanism proposalDescribe one causal mechanism, map its nearest P- bundles, and define the strongest ordinary null and rejecting testHigh; cross-domain comparison and experimental designAn analogy or issue does not create a principle or promote a claimMechanism proposal form
Go, validation and release toolingFix one reproducible defect, remove one duplicate path, or add one bounded failure testMedium to high; Go 1.27 and the affected policy or workflow contractGenerated output is not edited by hand; releases and settings are not claimed from local testsRepository or tooling form
SecurityReport a vulnerability with the smallest safe private reproductionAny; do not investigate beyond systems and data you are authorised to testExploitable details, credentials and personal data never enter a public issuePrivate vulnerability report

The detailed routing rules, including research-integrity concerns that must not be posted publicly, are in SUPPORT.md. Check existing issues before opening another coordination record.

Keep the task bounded

Before starting a pull request:

  1. name one source path, stable claim or principle ID, fixture, route, command, image digest or release tag;
  2. state the observed problem and the condition that would close it;
  3. identify the authority you must preserve, including any NO_RESULT, source digest, licensing, privacy or confirmation boundary;
  4. list the smallest focused checks that can falsify your change; and
  5. leave unrelated prose, generated files and workstation evidence untouched.

Coordinate first when work would create a new P- principle, alter a confirmation or held-out boundary, add normative or legal conclusions, import third-party bodies, publish a new experiment image, or involve people, private data, funding, donated compute or institutional obligations. Those changes can activate the collaboration, disclosure, rights, ethics or decision requirements in the research-integrity baseline and engineering and research contract.

Evidence to attach

ContributionMinimum review evidenceFocused checks before a pull request
Readability or documentationExact path or URL, intended reader, shortest affected passage and the interpretation that failednpm run check:prose; npm run validate:docs
TranslationLanguage, exact canonical and translated paths, exact source commit and digest, exact reviewed target digest, canonical UTC review instant, named language/domain reviewer, and disclosed drafting toolsnpm run validate:translations; npm run check:prose for any changed English source
Container reportPer-experiment image digest, tag, source revision, linux/amd64, exact command, output mount and NO_RESULT receiptRun the image's smoke, analysis and validation actions with networking disabled
Experiment changeFixture/track and claim IDs, frozen inputs, bounds, controls, failure state, output identity and authority impactnpm run validate:workstation; the affected fixture tests
Evidence or mechanism changePrimary locators, proposition actually supported, scope, uncertainty, nearest claim and principle IDs, strongest nullDocumentation, coverage and taxonomy validators relevant to the changed authority
Go or repository toolingReproduction, trust boundary, resource limits, expected failure behaviour and dependency identitygo -C tooling test ./...; the affected policy or release test
Pages changeRoute, viewport/browser context, keyboard or rendering reproduction, and before/after evidencenpm run test:github-pages

These are focused development checks, not the merge floor. Follow CONTRIBUTING.md, run npm run check before submission, and add the book generation and validation commands when a book source changes.

What a useful container report looks like

A released image that did not start or finish belongs in the short failed-run form. It asks only for the experiment, image identity, platform, command and smallest useful error excerpt. Use the full experiment form when the problem concerns a protocol, analysis, evidence status or authority boundary.

A container report can be valuable even when the diagnostic passes. Once a qualifying release publishes the images, download its checksum-bound and attested oci-images.json asset and copy the complete image@sha256:... identity from there rather than resolving a mutable tag. Release notes are a convenience display, not the identity authority. Different hosts can expose permissions, mount, architecture, runtime and error-reporting problems. Include only the bounded evidence needed to reproduce the behaviour:

artifact: fixture-007
image: ghcr.io/lusoris/20-watts-was-enough-fixture-007@sha256:...
release tag: vX.Y.Z
platform: linux/amd64
command: docker run ...
output volume: 20w-fixture-007-vX-Y-Z-run-001:/results/smoke
receipt: smoke/run.json (claim_eligible=false, scientific_result=false; NO_RESULT)
execution receipt: release-image; explicit sha256 digest; source revision; linux/amd64; smoke/smoke
observed: ...
expected: ...

Do not upload private seed material, credentials, personal data or copyrighted source bodies. A report about a smoke harness can improve the harness; it cannot turn its output into scientific evidence.