project-stateby Atomic47 Labs

Backstory

Born from a project that demanded more than the tools could give.

The pressure

A real program. Real stakes. Too many tools.

The Protein Industries Canada (PIC) consortium brings together industry, academia, and government to advance plant-protein innovation. The programs are ambitious — multi-year, multi-organisation, with quarterly claims, steering-committee packs, mandated publication-review windows, and a funder who expects precise, templated reporting on exact dates.

Cross-organisation communication, collaboration, and reporting demands were extraordinarily high, compressed into aggressive timelines. Every partner had their own tools, their own workflows, their own cadences.

The integration burden
Partner AJira + Confluence
Partner BSlack + Google Docs
Partner CEmail + Spreadsheets
FunderTemplated claims
Steering cmteQuarterly packs
you — the human integration layer
chasing context across everything
The problem

The information existed. It was everywhere.

Too many places

Slack threads, email chains, shared drives, Jira boards, Confluence pages, verbal hallway decisions. The same fact lived in six formats.

Reconstructed every time

Assembling the weekly status, the SC pack, the quarterly claim — the same story rebuilt from scratch, for every audience, every cycle.

Reporting as a second job

The project lead spent as much time assembling reports as doing the actual work. The reporting load nearly outweighed the program itself.

The shift

From reconstruction to rendering.

Before
Slack thread
Email chain
Meeting notes
Jira board
Shared drive
each report = reconstruct from scratch
After
One substrateplain files, shared drive
Facts recorded onceby whoever is closest
Skills render reportson cadence, per audience
Packs encode rulesswappable YAML, not code
You review, not authordrafts, never auto-sent
each report = render from state
The answer

project-state was the answer.

A single substrate — plain files on a shared drive — where facts are recorded once, by whoever is closest to them. Skills that generate every report from that state, on cadence, in the voice each stakeholder expects.

Compliance packs encode each funder's rules in swappable YAML, not hardcoded logic. The same engine that writes your weekly status also assembles the quarterly claim, the steering-committee pack, and the publication-review dossier.

The model
project-state/
the single source of truth
status
funder
review
Email draft
.docx pack
Blog post
fix the state → every report regenerates correct
In production

The proof is live.

pic-pcaisPRODUCTION

The PIC compliance pack is in production today — driving real quarterly claims, steering-committee lifecycles, and a mandated 30/14-day publication-review window. It's the proof that the model works: the reporting became a byproduct of the work, not a second job.

4
Quarterly claims filed
12
SC packs generated
30/14
Day pub-review window
427+
Activity log entries
Timeline

How we got here.

2024
PIC consortium begins

Multi-organisation plant-protein program launches. Reporting demands immediately exceed tooling capacity.

2025 Q1
First substrate prototype

Plain-file architecture. Milestones, risks, decisions as YAML. The activity log as append-only NDJSON.

2025 Q2
Skills go live

35 skills — status reporter, funder reporting, review meetings, document curator, harvester. All stateless verbs reading one substrate.

2025 Q3
pic-pcais pack ships

First compliance pack in production. Quarterly claims, SC lifecycle, 30/14-day publication-review window — all automated.

2026
keep-state-app + pilot

Desktop operator UI, local-first + GitHub hub sync, and the pilot programme opens to external teams.

Principles

What we believe.

Reporting is a byproduct

You do the work; the reports fall out of it.

State is the source of truth

When a report is wrong, fix the state — not the report.

Review, not author

Every artifact stops at human review. Nothing leaves without approval.

Packs configure, code stays generic

Your compliance regime is YAML, not hardcoded logic.

Own your data

Plain files, git history, runs offline. No database, no SaaS lock-in, ever.

Secrets stay local

0600 files on your machine. Not a keychain. Not a cloud.

The team
Atomic 47 Labs

We build operational software for programs that outlast any single team — local-first, own-your-data, no lock-in. project-state is our first product.

David Olsson

Founder & lead. Designed and built project-state end to end — the substrate, the 35 skills, the compliance packs, and the operator app — while running the PIC consortium program that demanded it.

Built under pressure. Ready for yours.

If your program has more stakeholders than hours, we should talk.

▸ project-stateATOMIC47 LABS