roadmap: automate the proven application-grade CV workflow

[akoita]

Product outcome

Deliver the proven application-grade CV workflow through three evidence-backed increments rather than one oversized release:

1. v0.7 produces a complete factual, source-traceable draft.
2. v0.8 adds independent critique, adjudication, readiness gates, and professional rendering.
3. v0.9 demonstrates parity with the consented private manual baseline and publishes the release evidence.

The target remains workflow parity, not merely successful agent execution or file export.

Product constraints

• One default CKB is expected; additional isolated CKBs are explicitly selected per application.
• Candidate sources remain maintainable, local by default, provenance-bearing, and untrusted as instructions.
• Candidate facts, chronology, employers, titles, dates, credentials, links, and metrics remain source-grounded.
• Job requirements and optional research never become candidate experience.
• Anthropic author and OpenAI critic remain the default independent pair.
• Human approval remains the final gate.
• Private reference material stays local; repository fixtures and reports remain sanitized.

Release sequence

v0.7 — Evidence-backed CV drafting

• feat(knowledge): manage reusable candidate knowledge bases and source lifecycle #78 — CKB tracking parent
• feat(knowledge): complete CKB source intake and incremental refresh #110 — source intake and refresh
• feat(knowledge): bind applications to explicit CKB and source-version snapshots #111 — application CKB selection
• feat(knowledge): expose CKB lifecycle through shared CLI and desktop contracts #112 — CLI and desktop CKB controls
• feat(storage): enforce CKB deletion, retention, backup, and restore safety #113 — lifecycle storage safety
• feat(profile): build a persistent canonical candidate career profile #66 — canonical candidate profile
• feat(opportunity): assemble a reviewed opportunity brief from URLs and instructions #67 — reviewed opportunity brief
• feat(retrieval): build knowledge-base-scoped lexical RAG with lifecycle evidence #80 — CKB-scoped lexical RAG
• feat(planning): map opportunity requirements to the strongest candidate achievements #68 — requirement-to-achievement planning
• feat(artifacts): compose a complete source-grounded application CV #69 — complete source-grounded CV
• feat(style): persist and enforce the candidate writing policy #70 — candidate writing policy

v0.8 — Independent review and readiness

• feat(critic): produce a structured independent application-readiness report #71 — structured independent critic report
• feat(orchestration): add explicit author adjudication and traceable revision #72 — author adjudication and traceable revision
• feat(readiness): define agreement and application-readiness stopping rules #73 — readiness stopping rules
• feat(rendering): deliver a professional ATS-readable CV template and visual QA #74 — professional ATS-readable rendering

v0.9 — Workflow parity and release

• eval(pilot): demonstrate parity with the proven manual CV workflow #75 — consented parity evaluation
• release: publish v0.9.0 application-grade CV workflow evidence #76 — v0.9 release evidence

Deferred from the critical path

• feat(research): add scoped user-approved opportunity and evidence research #79 — optional scoped research
• eval(retrieval): enable local vector or hybrid RAG only after measured gains #114 — local vector/hybrid retrieval evaluation

Program exit

• A complete factual CV preserves required sections and chronology with source traceability.
• Independent critique and author adjudication produce a traceable revision.
• Readiness and exact-artifact human approval gates prevent factual or structural regressions.
• DOCX/PDF output is ATS-readable and professionally usable.
• A consented private comparison records factuality, completeness, achievement recall, requirement coverage, editing effort, confidence, and limitations.
• v0.9 publishes cross-platform artifacts and traceable release evidence.