For Matt · 14 July 2026 · from Adam

Here's the new map around Precise.

You and Natasha built the science and got the original system running in GCP. We have since reorganized and productionized a lot around that work. This is the short path through what changed, how the harness works, and where everything lives now.

Repository: github.com/preciseai-inc/precise
Architecture: docs/atlas/pipeline.html
Start here: cd harness && npm run precise -- docs

The science is familiar. The control surface is new.

peer note

Hey Matt,

This is not an introduction to Precise. You know the science better than anyone. You were the method architect, Natasha wrote the production TraceScore engine, and Abhay's reference implementations underpinned the Shapley path. That work gave us the real system to build around.

We did change a lot around it. We folded the work into one repository, productionized more of the GCP path, reorganized packages and service boundaries, added the estate and deployment map, and put a common harness around the parts that should be callable, testable, and comparable. Some of the real cores are already connected through that harness. Others are catalogued honestly but still need a real caller.

The harness is not a replacement for the original work. It is the way the work introduces itself now. A human, UI, job, or agent can ask what a capability does, call the same core, run its evals, see the floors it enforces, compare real variants, and tell whether the thing is code, connected code, or something we have actually observed in GCP.

I want you to be able to walk the new structure quickly, compare it with the system you know, and tell me where our map is wrong or where we made a bad boundary. Then we can pick the next seam together. I had Codex assemble the map and the prompt below so you do not have to excavate the repo cold.

Call me after you have poked at it.

Adam

Here is the architecture in one pass.

Products · jobs · agents

Harness
commands · describe · evals · floors

Canonical Precise cores
the real math, decisions, evidence, and policy

Host ports
GCP today · Boombox where the machinery is reusable

What stayed canonical

The TraceScore method and production engine
The contribution, decision, and Trace service path
The protected original GCP estate

What changed around it

One monorepo with preserved authorship and real-core imports
Reproducible Cloud Run compute, manifests, receipts, and cost records
A harness, estate map, newer production spine, and explicit host boundary

The original cores stay canonical.

The harness imports and wraps the real implementation. We do not maintain a cleaner second version of the science for the sake of uniformity.

The tree and the estate answer different questions.

The harness tree shows what is connected and callable. The estate catalog covers the wider repository. Dated deployment observations record what we have actually seen running in GCP.

There are two GCP estates in the story.

trace-staging-486619 is the protected original Trace estate. precise-pipeline-prod is the newer operating spine. Our July 14 probe observed the newer spine, not the protected estate, and the harness does not yet control every remote service.

We harness consequential seams.

A seam earns a harness when it has meaningful variants to compare or a floor to enforce. Ordinary helpers stay ordinary code.

Precise owns the science and product.

Media economics, math, evidence, domain policy, experiments, customer behavior, and the product remain Precise. GCP or Boombox provides operating machinery through injected ports.

Tests, evals, and experiments are separate.

Tests prove software behavior. Evals prove invariants and floors. Ablations compare variants. Evidence-bound experiments decide whether a result deserves adoption.

The harness will give you the maintained route.

These commands show the connected tree, the wider estate, and the gap between repository intent and observed GCP state.

about five minutes

gh repo clone preciseai-inc/precise
cd precise/harness
npm ci

npm run precise -- docs
npm run precise -- describe
npm run precise -- estate coverage
npm run precise -- estate deployments --observed true

pipeline.html

The visual architecture and GCP shape.

HARNESS.md

The philosophy, discovery model, and full document map.

CONNECTIONS.md

What calls real code now, what is partial, and what is still a stub.

BUILDING-PRECISE-APPS.md

How an existing service joins without being rewritten.

AGENT-ABLATION-RUNBOOK.md

How we compare methods without fooling ourselves.

RUNBOOK.md

The commands for local work, verification, services, and GCP.

Let your coding agent do the repo walk with you.

This prompt treats you as Precise's Chief Scientist. It finds what changed around the original system, compares the harness map with GCP reality, and brings back the disagreements and useful questions.

Preview the coding-agent prompt

You are working with Matt Barlin, Precise's Chief Scientist and method architect. Matt helped build the original Precise science and GCP services. Do not onboard him to his own system or explain his own math back to him.

Your job is to help Matt inspect what changed around that work: the monorepo organization, production paths, harness, estate catalog, deployment observations, documentation, and current connection gaps.

Repository:
https://github.com/preciseai-inc/precise

Target handoff commit:
856dd76c5fb2ee9a0fe225092b41c47ccfebe4dd

Clone or update the repository safely. Preserve and report any local changes. Verify that the target commit is available before treating this prompt as repository truth. If it is not on the remote, say so plainly and continue only with Matt's direction.

Start read-only. Do not deploy, mutate GCP, or rewrite an existing service during the repository walk.

Read these completely:

1. AGENTS.md
2. .claude/skills/precise-harness/SKILL.md
3. .claude/skills/boundary/SKILL.md

Then run:

cd harness
npm ci
npm run precise -- docs
npm run precise -- skill
npm run precise -- describe
npm run precise -- estate coverage
npm run precise -- estate drift
npm run precise -- estate deployments --observed true
npm run precise -- optional-surfaces

Read the files returned in data.readInOrder. Use generated estate status and docs/atlas/CONNECTIONS.md as the current status authority. Use code and observed deployment evidence to check the prose.

Give Matt a compact working map:

1. Where the original scoring, contribution, calibration, decision, gateway, and GCP service code lives now.
2. Which real cores are connected to the harness and which are only catalogued, partial, stubbed, or unbuilt.
3. What the tree, estate catalog, deployment observations, and CONNECTIONS.md each tell him.
4. The few commands he will actually use while working in the repo.
5. Any disagreement among code, docs, estate metadata, and observed GCP state.
6. Questions where Matt's scientific or operational history is more authoritative than the repository.

Keep it technical, candid, and compact. The harness should wrap real cores, not replace them. Treat products, jobs, UIs, and agents as clients of the same contract. Distinguish tests, evals, ablations, and evidence-bound experiments. Do not make Polymarket the center of Precise. Do not prescribe a rewrite or a broad implementation program unless Matt asks for one. Follow Matt's direction after the map.

Open as plain text Open the repository

Then we compare the map with reality.

The useful first pass is your read of where the new structure matches the system you built and where it does not. Fixing those disagreements gives us a better foundation than another abstract plan.