Skip to content
GRPNR.

Dual-run parity harness — product research (“the quality-control bench”)

Goal: Full critical product research for the program’s flagship build — the dual-run parity harness. Decide (per angle, decisively) between build / limited MVP / OSS-foundation / buy / integrate / partner / process-change / pivot / abandon, and produce the PRD. The BUILD-flagship verdict from plans/001 is the input to attack, not the conclusion to defend.

Status: Research complete (2026-07-19) — executive-recommendation.md landed: BUILD, limited MVP on OSS foundation, pivoted shape (money gate partnered to Zaruba’s oracle plane; recorded replay re-scoped to functional/long-tail/DARK parity; two verdict tiers). Awaiting Robert’s acceptance; next action = E1 (Zaruba boundary conversation) + E2 (forex-ng boot PoC).

Trigger: Robert, 2026-07-18: comprehensive product research for the harness, pm:brainstorm → pm:discovery → pm:write-prd → pm:pre-mortem sequence, internal-product scoping allowed.

Read order

# Doc What it does
0 00-context-brief Shared ground truth + terminology for all docs
1 initial-recommendation Pre-research expert opinion (Fable) — the hypothesis under attack
2 research Brainstorm, problem discovery, evidence classes, advisory board, steelman
3 competitive-analysis Direct/indirect/substitute solutions, comparison matrix
4 open-source-analysis OSS foundation candidates, classified
5 journey Stakeholders + journeys (agent, builder, auditor paths)
6 user-stories Stories by persona and journey stage
7 ux-ui CLI/report/evidence-bundle experience spec
8 made-to-stick Internal narrative (SUCCESs), scoped for internal product
9 north-star Strategy, North Star metric + metric tree
10 architecture Kernel/lab/adapter architecture, ADRs, diagrams
11 build-vs-buy 8-option comparison, specific recommendation
12 prd The PRD (MVP = smallest credible assumption-testing product)
13 pre-mortem Failure modes, kill/pivot triggers
14 validation-plan Riskiest assumptions + cheapest disproving tests
15 executive-recommendation Final verdict (Fable), five-advisor summary, founder questions

Out of scope

  • Implementation (follows as its own plan once the PRD is accepted).
  • Zaruba-lane decisions: Orders scope, oracle internals, traffic-shifting machinery ownership (researched as boundary questions only).
  • Fleet monitor / work-ledger / triage products (plans of their own; interfaces referenced here).

Cross-references

Path What
plans/001-factory-apps-validation/ Portfolio decision that promoted the harness to flagship
analysis/legacy/SYNTHESIS.md Legacy estate evidence (EOL stacks, MBUS, hubs)
analysis/monorepo-development/ASSESSMENT.md Gap list naming the missing harness
research/02-web-research.md Verified agentic-refactoring case evidence

Run evidence

Produced by a Fable-orchestrated agent workflow (discovery + writers + audit), spawned-agents lane ($0, Claude Code session), 2026-07-18/19. Workflow wf_d5136910-64b: 28 agents (6 discovery, 13 writers, 3 adversarial auditors, 6 fixers), 27 done / 1 error (greenfield-recon agent died on structured-output return AFTER writing its notes; a follow-up enrichment agent folded its findings back into architecture/prd/journey), ~3.07M subagent tokens, 493 tool calls, ~39 min wall clock. Models: Sonnet (discovery, most writers, fixers), Opus (research/architecture/build-vs-buy/prd writers + auditors), Fable (orchestration, initial + executive recommendations, final audit). Audit pass: 9 findings fixed (DONE-row numbering convergence, citation mismatches, threshold consistency); fabrication auditor returned zero findings after web-spot-checking the highest-risk external claims.