SHAELA
Custom decision systems · built to ship · available for contract

Know the real number before you bet on it.

Your spend, your revenue, and your results live in separate systems, so the number you decide on is late and inflated. I build the system that reconciles them into one honest view, and the ranked moves that follow.

Here is one I shipped, in the open: Noon.

The problem

The dashboards are lying to you

Whether you steer a company, run operations, or decide where a client's money goes, you are acting on numbers that over-claim, arrive late, and never quite agree with each other. Every platform reports its own flattering version. The truth is scattered and slow. So money moves the wrong way, and the picture only clears after the spend is gone.

This is not one industry's problem. Anywhere real money meets messy data, the same gap opens. Below is what closing it looks like, built and shipped.

FoundersOperatorsAgency & media ownersPartners who resell to their own clients
The shift

From guessing to knowing, every morning

One honest number. One ranked list of moves: where to add, and where to pull. You stop optimizing against a distorted picture and start deciding on the truth, before the money is committed instead of after.

Proof · a system I shipped

Noon: the real number for media buying

Noon is the built, working version of everything above, for one concrete world: media buying. It pulls ad spend from every platform and the real revenue from the affiliate and lead system, reconciles them into one honest profit-per-offer view, and every morning shows the buyer exactly where to move money to grow profit: in one number and one ranked list. The daily digest is The Noon Report.

Picture the morning it changes. An offer that looked like a winner all week gets its leads scrubbed, and the real payout lands far under what every dashboard promised. Meanwhile a genuinely strong offer sits starved, its true payout invisible. Noon surfaces both: the hidden winner to fuel, and the drainer to pull budget off and redeploy. The morning inverts from guessing to knowing.

The Noon Report digest (synthetic demo data): platforms claim $154,141.63, you actually made $67,967.35, and $204,851.26 is still yours to reclaim this week, with a ranked Fuel list of offers to feed.

Every figure in the digest is from synthetic data shaped like real platform exports. Nothing here is a real client's result.

Why you can trust the number

Built like production, proven against known truth

A number is only worth deciding on if it is right. So Noon is built the way production software is built: a pure core holds the money math in plain Python with no framework inside it, so it can be tested in complete isolation. A single point does all the database work and hands plain records to that core. A test enforces the boundary, so the math stays independent and verifiable. And a golden-dataset evaluation runs the whole pipeline against a known-truth answer key, so the decisions are proven, not asserted.

raw inputsAd spend rows + conversion/revenue rows
infra (touches the database)Reader maps rows to plain records
↓  boundary enforced by a test
pure core (no framework)reconciliation → profit fact → report builder
views + templatesThe Noon Report digest

See it for yourself. Each link points straight at the artifact in the public repo.

Reconciliation core domain/reconciliation.py

Joins ad spend to real revenue on the tracking token, with an offer, geo, and date fallback. Pure Python, no framework.

The money-math tests tests/unit/test_reconciliation.py

Written test-first (red-green): the join, idempotency, reversals, de-dup, and the edge cases. 15 tests in the reconciliation file, part of 87 across the project.

Golden-dataset eval tests/integration/test_golden_eval.py

Runs the full pipeline on the seeded fixtures and asserts profit-per-offer, the flagged cases, and the ranked moves all match known truth.

Architecture guard tests/unit/test_architecture_boundary.py

Fails the build if anything in the pure core ever imports the framework. Keeps the money math independent.

Money as Decimal
Decimal throughout, in the domain and the database. Never float.
Real export shapes
Fixtures built on each platform's real export quirks: Google, Meta, Taboola, TikTok, and affiliate status.
Seeded data
12 months of synthetic, real-shaped data.
Designed for scale
Indexed canonical fact table, aggregate-then-join, retention tiers up to 5 years. Designed for scale, not benchmarked at it.
Test suite
87 tests.

Scale is a design target, not a demonstrated benchmark: a high-volume run was not executed, so nothing here claims "demonstrated at" a row count. Built with Python 3.12, Django, PostgreSQL, and a light JS layer for the live Refresh. The demo runs on synthetic data; a 2-minute walkthrough is linked above, and a live URL follows once yeenom.com is deployed.

Work with me

Want one built for you?

Noon is proof of what I build: production-grade systems that turn messy, lying data into one number you can act on. If your business runs on decisions like that, I will build you the version for your world.

Two ways to work

For your company

Founders and operators: I build the reconciliation and decision system for your data, your platforms, your moves. Built once, built right, and built to keep working.

To offer your clients

Agencies and partners: I build a system you put your own name on and offer to your clients. You keep the relationship and the upside. I get paid what the build is worth.

My own shipped work is built to pass the very audit I sell.

What is next

The roadmap

Today, Noon is deterministic: reconciliation, fuel and free-up moves, and a live Refresh. What follows is the plan, not shipped:

  • A learning engine that ranks moves from outcomes over time.
  • A licensable reconciliation API partners can build on and resell.