Skip to main content

​
What happened

Following the platform audit that identified 4 critical bugs and mapped all 25 repos, we scoped the complete implementation backlog: 46 GitHub issues organized into 7 phases across 8 repositories. Every issue has clear acceptance criteria and dependency chains. The platform can now be built incrementally by any agent or developer picking up the next available issue.

​
Business impact

The Clari platform converts raw agent work artifacts (TASKSET documents) into polished project documentation — Findings, Architecture Decision Records, Software Design Documents, and Runbooks. This session defined the full product roadmap from broken prototype to shippable product. Revenue model defined: Three Polar subscription tiers scoped with exact feature boundaries.
TierPrice pointWhat it unlocks
FREE$0Rule-based extraction, Findings only, 10 docs/month
PROTBDAI-powered extraction (Claude), Findings + ADRs + SDDs, 100 docs/month
TEAMTBDFull synthesis across multiple documents, all doc types, unlimited
The tier boundaries are designed so FREE demonstrates clear value (automated Findings from agent output), while PRO/TEAM unlock capabilities that require LLM spend — aligning cost structure with revenue.

​
Operational takeaways

  • Execution is now unblocked: 46 issues with dependency chains mean any session can pick up the next logical unit of work. No more re-deriving “what should I build next” — the backlog answers that.
  • Critical path is 4 phases: Fix bugs -> Build ingestion -> Build renderers -> Ship MCP tools. Estimated 6-8 focused sessions to MVP.
  • Shared models prevent drift: A recurring problem was Go and Python implementations diverging on data models. Shared library issues (#2, #3 on clari-shared-lib) ensure all services speak the same language from day one.
  • Each phase is independently deployable: F-0 fixes make sift-service usable today. F-1 makes extraction real. F-2 produces actual documents. No phase requires all others to deliver value.

​
Action items

  1. Next session: Start F-0 execution — clone sift-service, fix the 4 blocking bugs (#3-#6), verify pipeline runs end-to-end
  2. Pricing decision needed: Set PRO and TEAM price points before F-3 (Polar integration). Requires market positioning analysis.
  3. Agent delegation opportunity: F-0 bug fixes are mechanical — ideal for a code-focused agent. F-1 parser work is similarly well-scoped. Consider parallel agent sessions for independent subtasks.
  4. Customer validation: Once F-0 + F-1 + F-2 produce real Findings from real TASKSETs, test with one external user before building F-3 payment gating.
See also: Technical details