RUN 2026-08-08 21:00 → 2026-08-09 07:00 ET TALOS + NAMI UNATTENDED

While you slept

The two machines were taught to take turns, then measured properly for the first time: 200 jobs, no failures, 4,000 an hour each. Along the way they turned up a stowaway nobody had logged, and a spelling accident that has been mislabeling your live job board.

21:00 00:00 04:00 07:00 TALOSNAMIDSF ××× idle until 04:00 → RESERVED 123 4567 fighting for thegraphics chip clean · taking turns 21:00 22:00 23:00 00:00 01:00 02:00 03:00 04:00 05:00 06:00 07:00 TALOS NAMI DSF FOUND ×××× RESERVED 04:00–07:00 · FIRST LIVE NIGHT nothing scheduled until 04:00 → 123 4567 both machines fighting an uninvited program for the graphics chip clean — everyone taking turns

Solid = working cleanly · dashed red = starved or crashing · hatched = measurements that turned out to be lies

TALOS NAMI DSF BLOCK FAULT

Needs you

  1. Your live job board is mislabeling. Approve the fix. Over 1,500 jobs confirmed wrong, and the cause is understood. The repair is small and I have it staged, but it touches production, so it waits for your word (or hand it to the JobHinge chat).
  2. Confirm three assumptions I had to make while you slept. Written down in PRODUCT.md: that these morning reports become a recurring thing, that operational detail is fine on an unlisted page, and that decisions belong at the top.
Per machine, per hour. Measured on the clock, not projected: 100 jobs in 89.9 seconds. Both machines together clear about 8,000.
4,004jobs
Of each second, this much is actual work: reading and checking the job. The other two-thirds is fetching and filing. Speeding that plumbing up is the next easy win.
0.33of 0.90 sec
Jobs run, failures. 200 across both machines, and the twins produced identical results on identical work.
200 / 0
Live jobs wearing the wrong label: 503 nursing, 1,027 equipment, 31 aerospace. That is three keywords out of nineteen suspects, so the real figure is higher.
1,561
Live jobs with no category at all, a quarter of the board. This is where the unusual jobs are hiding.
79,759

What happened

Seven moments, matching the numbered flags on the chart.

  1. 21:50 ET
    1
    21:50 ET

    The first numbers were lies

    First real test: 20 jobs on each machine. Talos said 4.7 seconds a job, Nami said 11.5, and Nami's graphics driver kept crashing outright. Two identical machines shouldn't disagree by that much. That gap was the clue.

  2. 22:20 ET
    2
    22:20 ET

    There was a stowaway on both machines

    A production job-classifying service, running a model roughly five times heavier than ours, started up on both machines and never stopped. It was deployed months ago, appears in no schedule, and answered to nothing. Each machine's graphics chip can really only serve one program at a time, so ours stood in line behind it: Talos timed out politely, Nami's driver fell over.

    Your instinct, one model per machine and one job at a time, is now enforced in the machines' settings rather than just in policy.

  3. 00:46 ET
    3
    00:46 ET

    Measured properly, it's fast

    With the machines to ourselves: 100 jobs each, zero failures, 89.9 seconds on the clock, or 4,004 jobs an hour. Nami's "hardware problem" was never hardware, it was the roommate: the twins finished within a tenth of a second of each other and produced identical results, column for column.

    Only 0.33s of each job is real work; the rest is fetching and filing. That plumbing is the next easy win.

    TALOS 100 jobs 89.9s → 4,004/hr walmart 50 · u-haul 50
    NAMI 100 jobs 89.8s → 4,008/hr walmart 50 · nvidia 50
    work per job 0.33s · trust skipped 18-24 of every 50 ai checks
  4. 01:03 ET
    4
    01:03 ET

    Everyone answers to the Baton now

    The stowaway turned out to have been built with a yield switch already inside it, waiting years for something to connect to. We connected it. It now takes the machine, does a batch, hands it back, and waits its turn like everything else. Proof from the ledger: our test asked for a machine while the worker held it, waited, took over 5.3 seconds after the worker let go, finished, and handed it straight back.

    01:38:30 worker releases talos
    01:38:35 test claims talos waited its turn
    01:38:41 test releases talos 5 jobs done
    01:39:08 worker reclaims talos
  5. 01:50 ET
    5
    01:50 ET

    Two letters have been mislabeling your site

    An NVIDIA networking job came back categorized as nursing. The cause: the category matcher looks for keywords anywhere inside a title, and "RN", the abbreviation for Registered Nurse, hides inside ordinary words. Intern. Journeyman. Harness. Government.

    It isn't only "RN". "PM" hides inside equipment. "AE" hides inside aerospace.

    Heavy Equipment Operator — Landfill → Product, Design & Creative
    Journeyman NetOps Engineer → Healthcare, Clinical
    Investment Banking Associate II → Healthcare, Clinical
    ElectroMechanical Harness Engineer V → Healthcare, Clinical
  6. 02:05 ET
    6
    02:05 ET

    You called it: the live site is affected

    You said you felt the live site was mislabeling. It is. Counted against production, read-only: 503 jobs wrongly in nursing, 1,027 equipment jobs wrongly in Product & Design, 31 aerospace jobs wrongly in sales. That's from three keywords out of nineteen suspects, so the true figure is higher.

    Fixed in the test system tonight and verified: the fakes stopped, while genuine nursing titles still land in nursing. The production repair is staged and waiting on you.

  7. 02:30 ET
    7
    02:30 ET

    The bigger truth: a quarter of the board has no category at all

    Fixing the fakes exposed something larger. 79,759 jobs, 24.5% of your board, carry no category, and some of what looked like coverage was noise wearing a label. That's the pool your marine biologist lives in.

    The design that answers it, from your own question: let the system say "I don't know" instead of guessing. A blank is honest; a wrong label is damage. Unknown jobs pool up, similar ones cluster, and when enough of them agree they graduate into a proposed new category for you to approve. Research found the free government job dictionaries (they already contain "marine biologist") and a national statistics agency running this exact three-step design in production, so most of this is adoption rather than invention.

Built tonight

All of it in a sandbox. Nothing on your live site was changed.

  1. B

    The Baton

    A shared calendar plus a pass-along lease, so the machines take turns instead of colliding. DSF owns 4:00–7:00 every morning on both; outside that they belong to job parsing. Hands back early when it finishes early, keeps going past 7 if a render is mid-flight, exactly as you described it. A new project gets a block by adding one row. If anything crashes holding the machine, it's released automatically after ten minutes.

  2. P

    The parse lab

    Per-company extraction rather than per-platform. That was your call, and clustering proved you right: Walmart and NVIDIA are both "Workday" and look nothing alike. Alongside it, a trust system watches each company's results and quietly stops double-checking once a company has earned it. That's where the speed came from.

  3. A

    The AI-versus-regex face-off you asked for

    Built and loaded: 100 jobs per machine where the AI reads each job cold and extracts everything itself, with no pattern-matching hints, scored field by field against the current method. It runs this morning after the DSF block.

Still ahead while you sleep