Hi maintainers — I would like to gauge interest in contributing a new reference
trace before doing the build work.
Background
I have been building PaperSwarm
on top of Flowtrace — a multi-role academic peer-review workflow for research
drafts. A paper is parsed, a reviewer set is selected by paper type, independent
reviewer roles run in parallel, conflicts are detected, a revision plan is
produced, and a Worker/Verifier loop verifies the revision.
What I would contribute
A new builder under scripts/examples/ (e.g. paper-review/) that walks the
full CLI lifecycle and demonstrates Flowtrace capabilities that the current
six reference traces do not cover well together:
- fan-out / fan-in — 4 reviewer roles (editor, methodologist, domain
expert, devil's advocate) run as independent steps, then fan into a single
conflict step. None of the existing data-analysis traces have a wide
many-to-one fan-in like this.
- blocked → resume — the
conflict step blocks awaiting author
adjudication between disagreeing reviewers, then resumes (analogous to
assumption_check in iris-analysis, but for multi-source disagreement).
- error → retry — the
revise step's Verifier rejects an incomplete
revision, triggering a retry loop.
Questions before I start
- Is a non-data-analysis reference trace welcome? Most existing traces
are data pipelines; only tailored-resume and spring-demo are not. A
multi-agent review DAG would extend that non-data coverage.
- Fixture domain — I am leaning toward a short, fictional
social-science / education paper so only the 4 base roles are needed and
the fixture stays small. Any preference?
- Scope — I have a 9-step working version (already validated end-to-end
with a real run). Does that size fit the "land in an hour or two" spirit
of the existing examples?
Happy to adjust the theme/scope to whatever fits the project best.
Hi maintainers — I would like to gauge interest in contributing a new reference
trace before doing the build work.
Background
I have been building PaperSwarm
on top of Flowtrace — a multi-role academic peer-review workflow for research
drafts. A paper is parsed, a reviewer set is selected by paper type, independent
reviewer roles run in parallel, conflicts are detected, a revision plan is
produced, and a Worker/Verifier loop verifies the revision.
What I would contribute
A new builder under
scripts/examples/(e.g.paper-review/) that walks thefull CLI lifecycle and demonstrates Flowtrace capabilities that the current
six reference traces do not cover well together:
expert, devil's advocate) run as independent steps, then fan into a single
conflictstep. None of the existing data-analysis traces have a widemany-to-one fan-in like this.
conflictstep blocks awaiting authoradjudication between disagreeing reviewers, then resumes (analogous to
assumption_checkin iris-analysis, but for multi-source disagreement).revisestep's Verifier rejects an incompleterevision, triggering a retry loop.
Questions before I start
are data pipelines; only
tailored-resumeandspring-demoare not. Amulti-agent review DAG would extend that non-data coverage.
social-science / education paper so only the 4 base roles are needed and
the fixture stays small. Any preference?
with a real run). Does that size fit the "land in an hour or two" spirit
of the existing examples?
Happy to adjust the theme/scope to whatever fits the project best.