How Elrond Plans the Build: Sprints, Budgets, and a Plan That Learns
When an autonomous build system works serially — one initiative at a time — a backlog is not a pile; it's a sequence waiting to be ordered. Elrond takes every planner-filed initiative awaiting work, assigns each a human-supplied URGENCY level (critical/high/normal/low), and orders them into one-day sprints (SPR-2026-06-22, SPR-2026-06-23, ...) where dependencies are respected and the result is a row of dated, ordered tasks. A human never says "work on this on Tuesday"; Elrond figures out which day each initiative lands based on cost and prerequisites. The backlog becomes visible — no more guessing which of 140 waiting tasks comes next.
The build team has a monthly budget of EUR 200. Divide by the number of days in the month, and each day gets ~EUR 6.67. Elrond estimates the cost of each initiative (effort and resource usage derived from sizing, never guessed by the requester), and packs initiatives into days until the daily budget is full — then the next initiative rolls to the next day. That roll-to-the-next-day IS the ETA: an initiative's promised finish date is simply which day it lands on once everything ahead of it is paid for. Cost becomes the constraint that produces honesty: overruns are expensive and visible; estimates are learned, not guessed.
Every morning, Elrond's Sprint Builder rebuilds the entire forward schedule from scratch. Any initiative that didn't finish yesterday simply re-packs into the next available slot with a fresh ETA — no manual rollover, no status meeting. The plan self-heals because it doesn't cling to old estimates. Elrond tracks the INTERNAL ETA (raw schedule) and the OFFICIAL ETA (outside promise), which pads the internal date by a historic-delay factor — if past work ran 20% over, next sprint's promises pad by 20%. The system gets faster because it replans; it gets more honest because it learns how much it tends to slip.
This all applies to the planner pipeline — the backlog of normal, filed initiatives that the autonomous build system works through in order. But Orion's bypass work and Albus's out-of-band fixes don't enter a sprint; they ship directly because they're emergencies or architectural patches. The sprint plan is the autonomous team's fair, budgeted queue. Bypass is the operator's emergency lane. Keeping them separate is how the autonomous system stays predictable while the team can still move fast when it matters.
