Waylight · design notes

A plan you cannot argue with is a plan you cannot trust.

Why the scheduler shows its constraints, and what happens on the fortnightly morning when it cannot explain itself.

We publish the scheduler's assumptions because a plan you cannot argue with is a plan you cannot trust. Every board Waylight cuts carries the constraints it was cut under, in plain language, at the top of the sheet.

When the plan changes, it says why. A reshuffle that cannot name its cause is withheld and flagged for a human instead of published, which happens about once a fortnight and is the feature working, not failing.

None of this makes the scheduler smarter. It makes it accountable, which in a yard is worth considerably more — a crew will work around a constraint they understand and will quietly ignore an instruction they do not.

That is the whole design brief, and it is why the product looks like an instrument rather than an assistant. Instruments show their reading and let you decide.

M. Aldana — Founder

Waylight · who is behind it

Two of us, and what each of us answers for.

We started Waylight because we had both spent years watching good crews lose their mornings to an argument about what had happened overnight. Not to hard work, not to weather — to the absence of a shared record.

We are two people and a small team, and we intend to stay small enough that you can find out who made a decision. That is why the list below says what each of us is answerable for rather than giving us titles.

If something here is wrong, one of us wrote it, and one of us will answer for it.

  • Mara AldanaFounder

    Answerable for

    What the scheduler decides, and every plan it refuses to explain.

  • Tobias BrekkeFounder

    Answerable for

    Whether it stays up, and what we publish when it does not.

Both addresses are first name at waylight, and they reach us rather than a queue.

Switchyard · the year in routing

What a year of better routing added up to.

Twelve months ago every request waited in one queue. This is what moved once requests started landing with the right desk on the first try — measured, not estimated.

Figures from the routing ledger, full-year window, all desks.

  • Requests routed

    412,000412,000+18%

    Requests routed: 12 points, from 60 to 95

  • First-touch resolution

    7171%+9pts

    First-touch resolution: 12 points, from 52 to 71

  • Median wait, minutes

    66−11m

    Median wait, minutes: 12 points, from 6 to 17

Waylight · what went wrong

Every incident of the last year.

Written by the person who fixed it, published whether or not anyone noticed. If this list were empty we would not expect you to believe it.

33
incidents, last 12 months
433433
minutes affected, total
  1. 14 Jan

    Boards slow to cut for 41 minutes

    degraded
    What happened
    A migration held a lock on the planning tables longer than we had tested. Boards still cut, but some took over a minute.
    What changed
    Migrations now run against a copy first, and the planner reports its own latency to the status page rather than to us.
  2. 2 Nov

    Gate view unavailable, 12 minutes

    outage
    What happened
    A bad config reached production because the deploy gate only checked the primary region. Two yards saw an error page at shift change, which is the worst possible minute.
    What changed
    Config changes now roll region by region with a five minute soak, and the gate view falls back to the last cut board offline.
  3. 8 Aug

    Exports delayed overnight

    degraded
    What happened
    A queue backed up behind one very large export and nothing shed load. Nightly exports landed at 06:40 instead of 23:00.
    What changed
    Large exports run on their own lane, and anything late now tells the customer before we notice.

P.S. — We built Waylight because our own mornings were arguments. If yours are too, the first yard is free, and it takes an afternoon. Worst case, you go back to the radio and you have lost a Tuesday.

M. AldanaFounder, Waylight

or read how a rollout goes