Open the read-only demo → — a fixed fictional portfolio as of 2026-07-01. Every page is a real driftless screen with the write controls removed.
277 pages: the portfolio dashboard, one project’s hub with its schedule, board, RAID log and twelve assistants, and the method reference. RAID is risks, assumptions, issues and decisions. Each part of the method reference opens on its own:
Status is computed, never typed
Most project tools assume a person at the screen. Take the person away and most of what remains is theater: status fields somebody types, dashboards somebody patches up, documents somebody forgets to update. The plan says one thing, the tracker another, the weekly deck a third. Nobody decided that; three hand-kept copies of the truth drifted apart.
driftless starts from the opposite idea. Information is entered once and everything else is computed. Dashboards, rollups and documents are queries over one store — never stored copies.
What cannot drift
These states are not just banned. They are hard to even write down.
- Percent complete cannot be typed. A status snapshot takes its figure from the calculation core. The create request has no field for it, so a screen and a document cannot disagree.
- Plan history is additive. Re-baselining adds a new version; it never edits the old one. An approved baseline — the plan of record — cannot be changed or deleted.
- Signing off a threat does not hide it for good. It stays hidden only while its score is no worse than at sign-off; a regression brings it back.
- Reports are not the artifact. Every generated document takes a stated as-of date and never reads the clock, so building it again gives the same bytes. That is tested, so a report can be deleted at will; the store was always the artifact.
- Process state is computed, not tracked. The predictive processes of PMBOK (the Project Management Institute’s standard method) live as a versioned catalog in code. A project’s state on each is read from which artifacts exist plus the sign-off ledger, so no process table can fall out of step.
What it tracks
One store holds the whole chain: business → portfolio → program → project → workstream → task. Per project it holds baselines, milestones, sprints, risks, issues, change requests, budget lines and dated costs; departments and people carry capacity and cost rate.
Earned value (what the work done so far is worth against plan), forecasts and rollups are pure functions over those rows. The web pages and the report engine share those functions, so a number on a screen and in a document come from the same code.
The method is a map you can check
- The method map draws every process, technique and artifact kind once, joined by lines for what feeds, uses, produces or reads what. Point it at a project and each process node shows that project’s own state; focus one node and its neighbors light up.
- Every technique says how much help it gives. A guide only explains; a worksheet fills in a page of related techniques; a calculator runs its own formula. Each result shows its formula beside it, computed from rows already stored — nothing is typed twice.
- The proof page is the service checking itself. It walks the whole catalog and names any process, technique or artifact kind left unreachable, unexplained or unlaunchable without a stated reason.
Runs without a person at the screen
A wizard reports the next unfinished process as structured output. It produces that process’s output through the same checked write path the web pages use. A loop can stand a project up end to end, and every write lands in the change log. That is what headless means here: the interface is a reader rather than a privileged writer. The human and the agent go through the same gate, and the gate is the only door.
Who it is for
An operator running real delivery work who wants PMBOK-grade discipline without a PMO (project management office) to keep it up. The week the numbers are bad is the week somebody smooths the deck; an agent under pressure summarizes on the bright side. Here the bad week looks like the good one, because there is nowhere to type.
Where it stands
driftless runs Back Road Creative’s live project delivery; posts are waypost’s job, a different product. The source is public under Apache-2.0 at Back-Road-Creative/driftless, so every claim on this page can be read against the code. The public snapshot is re-taken on Back Road Creative’s cadence and may trail the version the demo shows. There is no installer and no self-service release; driftless is available through a scoped engagement. To run your own delivery work on it, get in touch.
Documentation
driftless’s documentation lives in its repository, next to the code it describes — the README is the manual.