The factory

Built with itself, on the record.

Fulltrace's own development runs through Fulltrace. Every slice of the product ships through the same governed ceremony, the engine keeps the receipts, and the dated output is the roadmap on this site. This page is the method.

01 · The difference

Specs are promises. Traces are receipts.

Spec-driven development is the industry's current answer to governing AI work: write the intent down, hand it to an agent, review what comes back. Fulltrace starts where that stops, because a spec has no way to check itself at run time. Here, the governance runs while the work does.

What a spec holds

Intent. A spec captures what the agent should build and trusts the run to comply. Nothing on the execution path checks the promise while it is being kept or broken.

What the run holds

Enforcement. Gates in code refuse bad work mid-run, an approval is tied to the exact change it reviewed, and every model call is capped, priced and recorded.

What the replay proves

Everything. Who proposed the change, what approved it, how it was verified and what it cost, on a real clock. Any claim about the work can be checked against the record.

02 · Three kinds of work

Verify. Build. Pursue.

The engine and the factory are different machines under one discipline. One verifies work that already exists, the other ships new work, and a third, loops that pursue a goal under hard caps, is being designed in the open. All three leave the same receipts.

Verify what exists

Shipped today. Studio's workflow runs sweep the work you already have: deterministic graphs audit a project, findings are graded and challenged, and no fix lands without a person at the gate. That engine is the features story.

Build what is next

Shipped today, as a method. New features move as scoped slices through the ceremony below: multi-agent build, adversarial review, an owner ruling at every gate. Since August low-risk slices can also be queued and built unattended overnight, one at a time under a time cap, and a slice that meets an open question stops and waits for a ruling. Studio's Build mode is the factory's cockpit.

Pursue a goal

In design, in the open. You set a goal and hard caps; the loop reruns a workflow until the goal is met or a cap trips: iterations, spend or wall clock. A tripped cap stops the loop, never loosens it. The design is public before the build.

03 · The ceremony

A software factory that shows its working.

Work enters as a scoped slice and leaves as a dated record, one slice at a time. The ceremony is the same every time, whoever, and whatever, did the building.

  • KickoffThe slice's row on the steering board flips to in-progress before any code moves. The record leads the work, and the scope is pinned in writing.
  • Decision gatesChoices that shape the slice are put to the owner as named gates and recorded as rulings. Nothing structural is decided silently.
  • Adversarial reviewA second pass argues against the work. New checks are first proven able to fail, claims are re-read against the code, and the reviewer's brief is to refute, not confirm.
  • CloseoutA dated closeout records what shipped, what was refused, and what it cost. The changelog and the board flip in the same pass, and the record is append-only. Since September every build is numbered and stamped with the slices in it, and every deploy leaves a click-pass page listing what to check by hand.

04 · In Studio

It runs its own roadmap.

Studio's Build mode reads the live steering board, tracks this ceremony for the slice in flight, and drafts the next one. Its launch buttons pick which assistant does the building, from a short list of verified ones. Today the factory builds Fulltrace itself: it is the method behind the product, shown live in Studio. The roadmap page on this site is hand-carried from that same board, dated on every carry.