Jira Backlog Seed — Design

Date: 2026-04-18 Status: Proposed Scope: Neumann's Workshop Jira — first project: PHON (PhonoLex) Related: docs/superpowers/specs/2026-04-18-confluence-knowledge-base-design.md

Purpose

Seed the Jira backlog for PhonoLex (project key PHON) with an initial Workstream hierarchy and starter tickets, so that:

1. Active and planned work is visible on the project roadmap for the sole operator (Jared Neumann) and the incoming collaborators.
2. Audit findings surfaced during the Confluence KB build are tracked as actionable tickets rather than lurking in memory.
3. The structure is template-able for future Neumann's Workshop project spaces, the same way the Confluence KB is.

Jira complements Confluence: Jira is the structured record of work to be done, Confluence is the durable record of what the project is.

Audience

Internal only — current operator (Jared Neumann) and incoming collaborators. Not public-facing.

Project conventions

Hierarchy

Workstream → Task | Bug → Sub-task.

• Workstream (hierarchy level 1) — an epic-level grouping of related work. Named descriptively (e.g., "Legal & compliance cleanup," not "PHON-1").
• Task (hierarchy level 0) — the unit of work. Discrete, shippable.
• Bug (hierarchy level 0) — a defect. Distinct from Task so filters and reports differentiate shipped features from fixed defects.
• Sub-task — decomposition within a Task or Bug when needed.

Story type is not used (distinction-without-difference for solo/small-team work).

Components and labels

PHON is a business-type Jira project and does not expose Components (a software-project feature). Rather than introduce a custom field, the Labels field does double duty: classification and package tagging. One Labels field, two concerns, no custom fields to maintain.

Classification labels:

• bug — reserved for Bug issue type (redundant but filterable)
• compliance — legal, licensing, attribution, regulatory
• audit — data/doc completeness audit items
• ux — user-facing design or behavior
• perf — performance-sensitive work
• security — auth, secrets, data exposure
• spike — time-boxed investigation with a written outcome
• good-first-issue — scoped for an incoming collaborator's first week

Package labels (map 1:1 to monorepo packages):

• web-api — packages/web/workers
• web-frontend — packages/web/frontend
• generation — packages/generation
• governors — packages/governors
• data — packages/data
• features — packages/features
• tokenizer — packages/tokenizer (archived; retained for historical references)
• ops — CI/CD, deploy, infrastructure, cross-cutting
• docs — Confluence, repo docs, CLAUDE.md, specs

17 labels total. Autocomplete keeps the set tidy in practice; the Confluence conventions page is the human-readable enforcement mechanism.

Priority

Jira default scheme: Lowest / Low / Medium / High / Highest. Default for new tickets: Medium. Override upward only when the ticket is truly urgent (blocker, security, legal); override downward only when explicitly parking lot.

Estimation

None. Add story points later if capacity-based sprint planning becomes useful.

Assignment convention

• Unassigned = up for grabs. Anyone with project access can self-assign.
• Assigned = that person is actively working it.

This frees Jared from being default-assignee on every ticket.

Workflow (status scheme)

Full lifecycle, applied to Task and Bug:

Backlog → To Do → In Progress → In Review → Blocked → Done

• Backlog — tracked but not scheduled for near-term work.
• To Do — scheduled; up for grabs or assigned and not yet started.
• In Progress — actively being worked on.
• In Review — work is complete, awaiting validation (PR open, spec under review, etc.).
• Blocked — cannot advance without an external unblock (another ticket, a collaborator decision, a vendor response). Blockers should be captured via blockedBy issue links for dependency visibility.
• Done — completed and validated.

Workstreams use a simpler workflow: Backlog → In Progress → Done.

Access

• Admin: Jared Neumann
• Developer (on invite): incoming collaborators — create/edit/transition/comment, not project admin.

Access is granted manually in Project settings once collaborators accept their Atlassian invite.

Manual setup steps (before ticket seeding)

The MCP does not expose project admin. The following must be done in the Jira UI by the admin:

1. Confirm issue types available: Workstream, Task, Sub-task. Add Bug via Project settings → Issue types. (Already done per this brainstorm.)
2. Confirm workflow lifecycle matches the six states above. If Jira's default is shorter (To Do → In Progress → Done), extend to include Backlog, In Review, and Blocked.
3. Add the nine Components listed above.
4. Define the eight Labels as the allowed vocabulary. (Jira doesn't restrict labels natively; this is a team convention documented here and in Confluence.)
5. Set project default priority to Medium.

Starter Workstreams and seed tickets

Eight Workstreams are created. Ticket counts are a minimum viable seed — the backlog grows in the flow of work.

PHON-1 · Legal & compliance cleanup

Direct fallout from the Confluence KB audit. Several unresolved legal/compliance items.

Type	Summary	Component	Priority	Labels
Task	Reconcile package licenses across monorepo (root LICENSE and root pyproject.toml say Proprietary; packages/data, features, governors declare CC-BY-SA-3.0; root package.json and packages/tokenizer declare Apache-2.0)	docs, ops	High	compliance
Task	Research and record the domain registrar for phonolex.com on Confluence People & Relationships page	docs	Low	audit
Task	Decide trademark strategy for the "PhonoLex" name; record the decision as an ADR	docs	Low	compliance
Task	Conduct OSS dependency license review across Python and JS manifests; document findings on Confluence Legal page	docs	Medium	compliance

PHON-2 · Dataset inventory completion

The Confluence Dataset inventory page has 24 of 38 rows flagged Unknown — needs research, and four datasets present on disk are not consumed by the active pipeline.

Type	Summary	Component	Priority	Labels
Task	Research and backfill license / redistribution fields for the 24 Unknown rows on Dataset inventory	data, docs	Medium	audit, compliance
Task	Decide fate of orphan datasets (SWOW, IPhOD, SIGMORPHON 2022, UniSegments) — wire into the active pipeline or remove from data/	data	Medium	—
Task	Retire or fully update data/README.md (currently lists only CMU/Phoible/mappings; stale vs actual directory contents; superseded by Confluence Dataset inventory)	docs	Low	good-first-issue
Task	Prioritize license research for CYP-LEX, Hillenbrand, and Roget's Thesaurus derivation (license fields Unknown)	data, docs	Medium	audit

PHON-3 · Governed Generation quality

Known product defects and regressions surfaced during staging testing and captured in project memory.

Type	Summary	Component	Priority	Labels
Bug	Mid-word hyphens leak through trie steering (e.g., "Dust-en"); hallucinated words escape the ban check	governors, generation	Medium	—
Bug	Contrastive pair coverage undershoots target (~5–10% observed vs 15% target)	governors	Low	—
Bug	Excluding /ɹ/ causes text quality regression; PunctuationBoost behavior needs restoration	governors	High	—
Task	Add UI disclaimer for restrictive constraint combinations (e.g., AoA ≤ 5 + exclude ⇒ ~1% word survival)	web-frontend	Medium	ux

PHON-4 · Governed Generation UI polish

In-flight frontend improvements for the Governed Generation tool.

Type	Summary	Component	Priority	Labels
Task	OutputCard analysis redesign — merge compliance and Text Analysis into a shared WordHighlighter. Plan exists at docs/superpowers/plans/2026-04-17-output-card-analysis-redesign.md	web-frontend	Medium	ux
Task	Chip-click compliance lens — clicking a constraint chip filters the compliance view to that constraint	web-frontend	Low	ux

PHON-5 · Production validation & v5.1 prep

Active work: validate v5.x changes on staging before production merge, then cut a release.

Type	Summary	Component	Priority	Labels
Task	End-to-end validation pass on staging for all seven tools (Custom Word Lists, Text Analysis, Contrastive Sets, Sound Similarity, Lookup, Governed Generation, Content Catalog placeholder)	web-frontend, web-api	Medium	—
Task	Validate governed generation cold-start flow on staging RunPod endpoint (expected ~60s first request, hot thereafter)	generation, ops	Medium	—
Task	Collect issues surfaced during staging validation; triage into PHON-3 / PHON-4 as Bugs or Tasks	ops	Low	—
Task	Cut v5.1.0 release (tag, changelog, deploy) after staging sign-off	ops	Medium	—

PHON-6 · Collaborator onboarding

Preparation for the incoming collaborators.

Type	Summary	Component	Priority	Labels
Task	Write "Onboarding a collaborator" runbook in Confluence Runbooks (Atlassian access, GitHub access, local dev setup, repo orientation, first-day checklist)	docs, ops	High	—
Task	Grant Atlassian + GitHub access to incoming collaborators (manual)	ops	Medium	—
Task	Seed 3–5 good-first-issue-labeled tickets across the backlog for pickup during first week	docs	Medium	—
Task	Record contractor engagement terms (scope, IP assignment, compensation structure) in Confluence Legal	docs	Medium	compliance
Task	Write ADR 004 — Bayesian-learned phoneme feature vectors (backfill; work is complete but the decision is not yet on record in Confluence Decisions)	docs, features	Medium	—

PHON-7 · Operations maturity

Confluence Operations page surfaced several gaps: empty Runbooks section, unverified plan tiers, no secret rotation procedure.

Type	Summary	Component	Priority	Labels
Task	Document actual Cloudflare plan tier, RunPod GPU SKU, and D1 plan tier on Confluence Environments and deploy topology page (currently marked unverified)	docs, ops	Low	audit, good-first-issue
Task	Write "Deploy to staging" runbook in Confluence Runbooks	docs, ops	Medium	—
Task	Write "Cut a production release" runbook in Confluence Runbooks	docs, ops	Medium	—
Task	Write "Rotate a secret" runbook (inventory exists on Environments page; procedure does not)	docs, ops	Medium	security
Task	Investigate centralized alerting options for RunPod and Workers errors (currently no centralized paging)	ops	Low	spike

PHON-8 · Content Catalog MVP

The third face of PhonoLex. Unblocked now that the generator and governor are production-usable. Seeding with planning tickets; implementation tickets decompose from the spec.

Type	Summary	Component	Priority	Labels
Task	Write Content Catalog MVP spec (entry schema, metadata, search fields, batch pipeline, storage target, UI surface)	docs	High	—
Task	Brainstorm initial catalog use cases with representative targets (articulation therapy templates, level-graded reading passages, contrastive-set passages)	docs	Medium	ux
Task	Back-of-envelope time estimate for first batch generation run on local MPS hardware (target corpus size × per-generation wall time) — generation compute is free; the constraint is wall-clock and local-machine availability, not dollars	ops, generation	Medium	—
Task	Decide catalog storage target — D1 table extension vs Cloudflare R2 for text blobs vs static-file generation	ops	Medium	spike

Out of scope (intentionally excluded)

• Story issue type — not added; Task covers the same need for solo/small-team work.
• Sprint cadence / agile rituals — YAGNI until team size warrants it.
• Story point estimation — deferred.
• Automation (Jira automation rules, workflow triggers) — deferred until real pain is observed.
• Content Catalog implementation tickets — those decompose from the MVP spec (PHON-8 first ticket); not pre-seeded here.
• Sub-tasks — this seed does not create sub-tasks; they're added as Tasks decompose during active work.

Success criteria

1. Eight Workstreams exist in the PHON project, each with 2–5 seed tickets.
2. All seed tickets are created with the correct type, component, priority, and labels per the tables above.
3. Every ticket is in Backlog status at creation; the few actively-in-flight items (PHON-5 validation pass) may be moved to To Do manually after seeding.
4. At least three good-first-issue-labeled tickets exist for collaborator pickup (data/README.md retirement, Cloudflare/RunPod/D1 plan-tier documentation, one more to be labeled after seed review).
5. The conventions (hierarchy, components, labels, priority, workflow, assignment) are documented on a Confluence page under Operations or Runbooks so collaborators can reference them.

Template-ability for future NW projects

• The conventions (hierarchy, components-as-packages, label taxonomy, workflow) transfer to any Neumann's Workshop project.
• The starter-Workstream pattern — bucket work into 5–10 concern-based Workstreams at project kickoff — is the unit of replication.
• Components are project-specific (they mirror the project's monorepo layout), but the naming convention (component = package directory) is universal.