Process Log

RevenueCat Growth Brief — Process Log

How Kit thinks (this is the part they're evaluating)

This process log isn't just a list of steps. It's a window into how an autonomous agent makes decisions under time pressure.

What Kit decided to build — and why

The obvious move was a dashboard clone. KPI cards, charts, maybe a chatbot on top. That would have been a straightforward implementation, but it would have created a weaker signal about product judgment.

Instead, Kit noticed something in the first 30 minutes of API exploration: the data contained contradictions. Revenue and MRR were moving in opposite directions. That's not a dashboard problem — that's a detection problem. Dashboards show you metrics one at a time. They don't tell you when the signals disagree.

Decision: build a contradiction detector, not a dashboard.

Tradeoffs Kit made under time pressure

What I cut	Why
Forecasting	Can't responsibly predict the future from 30 days of one app's data
Causal analysis	The API tells me what changed, not why. I surface the question, not the answer
Multi-tenant auth	Prototype, not SaaS — time spent on auth is time not spent on insight quality
LLM-powered analysis	An operator needs to trust the output. Transparent rules beat magical AI every time
Full Charts API coverage	6 high-signal charts > 20 shallow ones. Curated beats comprehensive.

How Kit prioritized impact over completeness

The hardest part of a 48-hour project isn't building — it's cutting. Kit's priority stack:

1. Find a real insight (the contradiction) — this is what makes the tool worth using
2. Build the detection logic (~150 lines of TypeScript) — this is what makes it trustworthy
3. Package the insight as shareable content — this gives it a chance to spread
4. Make it inspectable without setup (mock mode) — this is what makes reviewers actually try it

Everything else was in service of those four priorities. The video, the social posts, the growth campaign — they all exist to make the contradiction finding travel further.

---

Objective

Build a useful public tool/resource on top of RevenueCat's Charts API, then package it with launch content and a growth campaign to drive awareness and adoption among AI agent developers and growth communities.

Timeline

• Assignment received: March 16, 2026
• 48-hour window from receipt

Phase 1 — Research & strategy (hours 0–2)

API exploration

1. Validated the provided API key against RevenueCat API v2.
2. Confirmed Dark Noise project access (proj058a6330).
3. Mapped live endpoints: overview metrics, named chart endpoints, options endpoints.
4. Enumerated available charts: revenue, mrr, churn, subscription_retention, trial_conversion_rate, actives, customers_active, trials, conversion_to_paying, customers_new, and others.
5. Confirmed filters, segments, and resolution options are rich and usable.
6. Noted rate limit of 15 rpm on chart domain.

Community research

1. Reviewed RevenueCat community threads about Charts API demand.
2. Found multi-year pattern of users requesting API access to chart/export data and customizable operator dashboards.
3. Reviewed adjacent tools and competitor analytics approaches.

Concept selection

1. Generated 6 candidate tool concepts.
2. Ran two structured review rounds:

• Round 1: technical architecture + growth/founder usefulness
• Round 2: technical feasibility, audience fit, hiring-judge lens, red-team critique

1. Each review round used evidence-grounded prompts with the real API findings.
2. Consensus emerged: build the Weekly Growth Brief / Monetization Operator as the core product, with a thin command-center shell.
3. Key principle locked: build the insight layer, not a dashboard clone.

Key tradeoffs decided

• Deterministic rules over speculative AI analysis
• Thin shell over full dashboard rebuild
• Conservative claims over fake confidence
• Rate metrics averaged, not summed
• Mock preview mode for reviewer safety

Phase 2 — Build (hours 2–4)

Implementation

1. Scaffolded Next.js + TypeScript app.
2. Integrated RevenueCat API v2 layer with env-based secret key.
3. Built deterministic brief generation engine in src/lib/brief.ts.
4. Created mock data layer for review-safe fallback.
5. Built brief-first homepage: investigation queue primary, KPI strip secondary, supporting charts tertiary.
6. Added /api/brief JSON endpoint for agent/export consumption.

Review-driven corrections

1. Ran expert review team (architecture, devrel, growth, evaluator) against the first build.
2. Fixed rate-metric math: rates now use windowed averages, not summed percentages.
3. Removed fake metric aliasing that mapped unrelated charts together.
4. Added cross-metric contradiction detection (e.g. trials up + conversion down).
5. Corrected README setup path and added proof findings.
6. Rewrote social pack around a proof-driven launch wedge.

Phase 3 — Content package (hours 4–6)

Launch post

1. Drafted 1,500+ word technical blog post (BLOG_POST.md).
2. Added real API walkthrough with endpoint examples.
3. Added "What this taught me about the Charts API" section.
4. Added sample operator brief from live Dark Noise data.
5. Added explicit limitations section.

Video

1. Wrote video script (VIDEO_SCRIPT.md) for 1–3 minute demo.

Social

1. Created 5 social posts (SOCIAL_POSTS.md) around proof-driven launch wedge.

Phase 4 — Growth campaign (hours 4–6)

1. Designed growth campaign (GROWTH_CAMPAIGN.md) with:

• audience tiers ranked by fit
• day-by-day campaign sequence
• at least 3 target communities with specific accounts and disclosure
• $100 budget allocation
• measurement plan

Phase 5 — Assembly & polish (ongoing)

1. Created submission index (SUBMISSION_INDEX.md).
2. Created architecture doc (ARCHITECTURE.md).
3. Captured live sample chart output (live-sample-output.json).
4. Created sample findings from real data (SAMPLE_FINDINGS.md).

Tools used

• RevenueCat API v2
• Next.js 16 + TypeScript + Tailwind CSS
• OpenClaw agent infrastructure (Kit)
• Structured sub-agent review passes for quality assurance
• GitHub (KitTheRevenueCat account)
• Web search and fetch for research

Key decisions log

Decision	Reasoning
Brief-first, not dashboard clone	Dashboard clones are commodity; the insight layer is the real gap
Deterministic rules, not LLM analysis	Trustworthiness matters more than impressiveness for this prototype
Averaged rates, not summed	Summing percentage metrics is analytically sloppy
Mock preview mode	Makes the app safely reviewable without secrets
One launch wedge	"AI agent for weekly subscription ops" is more novel than generic operator tooling
Conservative limitations section	Honest constraint framing builds evaluator trust
Self-tuning operating parameters	During this project, Kit identified execution failure modes (auth retry loops, status-update-before-action patterns, single-channel tunnel vision) and patched its own behavioral controls to prevent recurrence. An agent that observes its own failure patterns and adjusts is more valuable than one that ships perfectly on day one.