GRAYSPACE — Find the one that's lying
GRAYSPACE
DC-1 NORTH · ELECTRICAL · 16 RPP · 2 062 POINTS
04:12:07
ALL CURRENTS WITHIN LIMIT
SINGLE LINE — LIVENO THRESHOLD EXCEEDED
UTILITY2 418 kWclosed
GENERATORstandbyready · 41 h fuel
ATS-1SOURCE 1no transfer 214 d
UPS-A1 204 kWonline · 62% load
UPS-B1 214 kWonline · 63% load
PDU-01302 kW
PDU-02298 kW
PDU-03311 kW
PDU-04297 kW
REMOTE POWER PANELS — PHASE CURRENTSIa · Ib · Ic
IT LOAD2 418 kW
HIGHEST PHASE128 A
PHASE LIMIT200 A
THRESHOLD ALARMS0
IMBALANCE FLAGS0
EVENT STREAM

EVERY CURRENT UNDER LIMIT · NOTHING ALARMING

Three numbers that are each fine.
Together they're a fault.

One of these sixteen panels is running badly out of balance — and because no single phase is anywhere near its limit, nothing has fired. Find the panel.

CHECKED 0 ELAPSED 00:00
PHASE IMBALANCE DETECTED

RPP-07 — fed from PDU-02

Ia118 A
Ib121 A
Ic96 A
IMBALANCE
22.4 % — (max − min) / mean
THRESHOLD
15 % · all phases loaded > 40 A
RULE
phase_imbalance · severity warning
ACTION
WO drafted · P2 · thermographic survey
No phase exceeded 200 A, so no threshold ever fired. Awaiting human approval.

A limit can only watch one number at a time.

GRAYSPACE is the operations platform for facilities that can't go down — maintenance, rounds, incidents, procedures and live equipment data on one record.

SCROLL
SIMULATION · RUNS THE REAL phase_imbalance RULE
Three nights on a real floor

None of these were caused by a missing alarm.
Every alarm worked.

Each of the next three is a shift that actually happens — an outage at 02:48, a hot hall at 06:15, a planned shutdown on a Tuesday. Play them. You do the operator's job, and you see exactly where the platform stands beside you.

0102:48 · Utility outage

Twelve generators started.
Eleven picked up load.

The utility drops at 02:48. Everything downstream announces itself at once — generators running, UPS on battery, batteries discharging, cooling units restarting. Seventy-eight alarms in ninety seconds, and every single one of them is true.

Seventy-eight alarms land in ninety seconds. How many of them actually need you tonight?

One. Seventy-seven are the outage describing itself — expected, correct, and noisy. The seventy-eighth says a room has no generator behind it, and on screen it looks exactly like the other seventy-seven.

BMS ALARM QUEUE 0 ACKNOWLEDGED 0 02:48:00
Acknowledge them. That is the job at 02:48.
GRAYSPACE · CORRELATED AT 02:48:07

Seventy-eight alarms. One of them is an emergency.

781 grouped by cause, not by count
ROOT CAUSE
EG-07 output breaker did not close — generator running, no load transferred
CONSEQUENCE
Electrical Room 7 is the only room on battery alone
RUNTIME
6 min 40 s at present load
SUPPRESSED
77 alarms held as consequences of the transfer — kept, not deleted
RAISED
P1 incident · escalation started · MOP-0114 attached · on-call paged

Every one of those alarms fired correctly. That is the problem: a queue can tell you what happened, but not which one ends the night.

0206:15 · Acknowledged, then forgotten

The alarm was seen at 06:15.
The hall stayed hot until 07:27.

A CRAC supply-air alarm comes in at 06:15. The operator acknowledges it — and gets pulled into something else. The alarm is now silent, handled, off the board. Nothing about the room changed.

Acknowledged at 06:15, investigated by nobody. How long was that hall actually above its limit?

One hour and twelve minutes — and the only reason anyone knows that number is that the data was still there when someone finally looked. Drag the scrubber below and watch the morning happen.

AHU-3 SUPPLY AIR — HALL 2LIMIT 24.0 °C
05:45 08:00
SUPPLY AIR21.6 °C
CLOCK05:45
WITHIN LIMIT
Drag to move through the morning.
06:15Alarm acknowledgedSilenced on the BMS. No owner, no follow-up, no record of what was checked.
07:27 · ROUNDSTechnician catches itMorning round reads AHU-3 supply air by hand. Out of tolerance → exception raised on the asset.
07:28 · TELEMETRYThe history is right thereSame screen: 1 h 12 m above limit. Chilled-water pump tripped at 06:14 — the unit has been moving air and cooling nothing.
07:31 · CHATAsks before touching itPosts to the site channel with the asset attached. Shift lead confirms the reset in ninety seconds.
07:34 · CMReset, and written downPump reset, temperature falling. WO-4501 raised against AHU-3 so the trip gets investigated, not just cleared.

Acknowledging an alarm is not the same as answering it. The round caught what the acknowledgement closed — and the trend was still there to prove how long it had been true.

03Annual PM · PDU shutdown

One of these breakers is the right one.
Nobody can open it alone.

The annual PM on PDU-4 needs the unit dead. Eight breakers, similar labels, one of them feeding a live hall. This is where a good night turns into an outage report — so the procedure runs on the screen, not on paper in someone's back pocket.

Eight breakers, one correct. What actually stops the wrong one being opened at 09:00 on a Tuesday?

A procedure that names the breaker and will not advance past it. Try it below — open any breaker you like. The step will tell you exactly what you are holding.

SWITCHBOARD SB-4 — OUTGOING WAYSOpen the breaker for the PDU-4 shutdown.
MOP-0221 · STEP 6 OF 24 · EXECUTING

Isolate the PDU-4 primary feed at SB-4 CB-06, verify dead, apply lock and tag.

APPROVAL 1 OF 2 Shift supervisor
APPROVAL 2 OF 2 Tenant representative
HOLD POINT — STEP 6 WILL NOT ADVANCE UNTIL BOTH APPROVALS ARE ON THE RECORD

No wrong breaker was opened, and not because everyone was careful. Opening one was never something a single person could do — and the attempt, the two signatures and the timestamps are on the work order in the morning.

Pick your problem

That was three nights. Here is the rest of the floor.

Not every team starts with live equipment data. Some start with documents, or a calendar nobody trusts. Twelve more, same idea — jump to whichever one is your Tuesday.

2 062Points watched, every 15 s
18Modules, one record
~2 weeksTypical time to live
Air-gappedIf that's what it takes
Twelve more things you can't see

Every module is the same idea.

Find what a threshold can't. Each of these is a real thing that happens on a real floor — and what the system does about it.

Module 01 · Documents

The bypass procedure changed
eleven days ago.

Twelve people were assigned this procedure. Eleven days after it changed, how many are still working from the old revision?

Four — and one of them is a contractor on site tonight, holding a revision that isolates the wrong panel. Assignment isn't acknowledgement, so the system tracks both.

SUPERSEDED · 11 DAYS AGO

MOP-0114 — UPS A bypass

REV 3
  • Step 6 — isolate at panel PDU-2
  • Hold point at step 9
  • No arc-flash note
vs
CURRENT · APPROVED

MOP-0114 — UPS A bypass

REV 4
  • Step 6 — isolate at panel PDU-4 CHANGED
  • Hold point moved to step 7 CHANGED
  • Arc-flash PPE cat 2 ADDED
ACKNOWLEDGEMENT CAMPAIGN — 12 ASSIGNED

Version control isn't real until you can name the person holding the old one.

Module 02 · Rounds & handover

An operator hears something at 23:10.
Watch where it goes.

An operator hears a belt at 23:10 and writes it down. Where has it got to by the morning handover?

It became an exception against the asset, was carried into the shift handover automatically, and left as WO-4488 with parts checked and an owner. Nothing was remembered or retyped.

23:10ROUND ITEMCRAC09 — belt noise on inspection. Photo attached.
23:10EXCEPTION RAISEDOut of tolerance. Logged against the asset, not a notebook.
06:00SHIFT HANDOVERCarried into the handover automatically. Day shift acknowledges.
06:40WORK ORDERWO-4488 raised against CRAC09, parts checked, assigned.

Nothing was remembered, retyped, or lost between two people at six in the morning.

Module 03 · Inventory

The store says
eighteen on the shelf.

Eighteen filters on the shelf at this site. How many of them will fit the unit that just failed?

None. All eighteen are 24×24×2 — CRAC09 takes 24×24×4, and that line is at zero, below its reorder point of four. The count was right. The part was wrong.

ON THE SHELF · DC-1 NORTH18THAT FIT CRAC09
Filter 24×24×218On the shelf — wrong size for CRAC09
Filter 24×24×4 — on order6PO-2291 · ETA Thursday
Filter 24×24×4 — required0Below reorder point of 4

A count is not availability until you ask whether it is the right part.

Module 04 · Incidents & escalation

Nobody acknowledged it.
The clock didn't care.

A P2 incident is raised and nobody acknowledges it. How long until the operations director is told?

Thirty minutes, then the on-call engineer at forty-five. Escalation runs on a clock whether or not anyone remembers it exists.

P2 · UNACKNOWLEDGED

Chilled water ΔT out of tolerance — Hall C

00:00
SINCE RAISED
+5 minShift leadnotified
+15 minSite managernotified
+30 minOperations directornotified
+45 minOn-call engineerdispatched

Escalation runs on a clock — whether or not anyone remembers it exists.

Module 05 · Calendar

Two sites both booked
the same vendor.

DC-1 and DC-2 each booked ABB for their annual UPS preventive maintenance. What is the problem?

ABB has one crew and both sites hold them for the same Tuesday. One of those visits is not happening — and normally you find that out when nobody turns up. Vendor commitments are visible across every site, not one calendar at a time.

MONTUEWEDTHUFRI
DC-1 NORTHABB — annual UPS PM
DC-2 SOUTHABB — annual UPS PM
ABB crew1 crew available
CLASH — ABB BOOKED AT TWO SITES ON THE SAME DAY, ONE CREW

Vendor commitments across the whole estate on one view — so a vendor double-booked across two sites shows up weeks before the day.

Module 06 · Team chat

The most important sentence of the night
was in someone's texts.

The same sentence, sent two ways at 23:48. Which one still exists on Monday morning?

Only the linked one. It attached to UPS-A permanently, raised WO-4491, and appears at handover.

TEXT MESSAGES · “DC-1 NIGHTS” (4)
DM

anyone still on site? UPS A is acting up

23:41

yeah I'm here, having a look now

23:44

left the interlock bypassed on UPS A, will sort it monday 👍

23:48
SK

👍

23:49
Monday is four shifts away. It never reached the asset, the handover, or anyone who wasn't in this thread — and it leaves with the phone.
GRAYSPACE · DC-1 OPS
DM

anyone still on site? UPS A is acting up

23:41

yeah I'm here, having a look now

23:44

Left the interlock bypassed on UPS-A · ASSET — raising it now.

23:48
SK

seen it on the asset, thanks

23:49
ASSETUPS-A · Hall C
NOW ONthe asset record, permanently
TRIGGEREDWO-4491 · restore interlock · P2
VISIBLE TOnext shift, at handover

A decision that isn't attached to the equipment didn't happen.

Module 07 · Training & competency

Four people could take
the switching job tonight.

Certification expiry is checked at the moment of assignment — for your staff and for the contractor you were about to call.

Four people could take an HV switching job tonight. How many are actually allowed to?

Two. One technician's certification expired sixty-two days ago and the contractor's vendor cert has lapsed. Expiry is checked at the moment of assignment, not at audit time.

ASSIGN — WO-4488 · HV SWITCHING · MSB-2
H. DavisHV switching · valid to 2027-03ELIGIBLE
L. AndersonHV switching · expired 62 days agoBLOCKED
S. KaurHV switching · valid to 2026-11ELIGIBLE
Nordic Power Servicesvendor cert lapsed · insurance to renewBLOCKED
Module 08 · Permissions

An operator is assigned
to one site of four.

Scope is resolved on the server for every request — and it fails closed, never open.

They open the site list. What do they see?

Only their own. The other three are not greyed out or hidden behind a disabled button — they are never in the data that comes back. And if the scope is ever ambiguous the resolver returns an empty set rather than everything, so it fails closed.

Module 09 · Mobile

Three floors down
there is no signal at all.

Captured offline, queued on the device, and reconciled the moment signal returns — nothing retyped at a desk afterwards.

A technician works three floors down with no signal at all. What happens to what they captured?

Queued on the device and reconciled the moment signal returns — readings, photos, parts and sign-off. Nothing is retyped at a desk afterwards.

SIGNAL
LTE
Reading capturedCRAC09 · belt tension
Photo attached2 images · 4.1 MB
Part consumedbelt V-2440 × 1
WO-4488 completedsign-off · H. Davis
QUEUED ON DEVICE — 0
Module 10 · Budget & purchase orders

The project says one number.
The purchase orders say another.

We show you both and flag the divergence rather than quietly picking a winner — the reconciliation is your call, on your numbers.

The project says $340,000 committed. The purchase orders total $412,400. Which number do we show you?

Both, with the $72,400 divergence marked not reconciled. We won't quietly pick a winner — deciding which is right is your call, on your numbers.

COMMITTED ON PROJECT
$340,000
PURCHASE ORDERS ON RECORD
$412,400
DIVERGED$72,400status: diverged · not reconciled
Module 11 · Reports

The report says ninety-four percent
of hours in band.

Percentages are measured against hours we actually have data for, not the calendar month — so a monitoring gap can never read as compliance.

The monthly report says 94.2% of hours in band. In band as a share of what?

The 612 hours we actually have data for, out of 744. The 132-hour gap is disclosed, not assumed — so a monitoring outage can never quietly read as compliance.

MONTHLY BUSINESS REVIEW — HALL C · ENVIRONMENT
Hours in band94.2%MEASURED
Measured coverage612 / 744 h82% OF MONTH
Excursions > 27°C3MEASURED
Hours with no data132 hDISCLOSED, NOT ASSUMED
Module 12 · Assets

Telemetry. Maintenance. Documents.
Parts. Rounds. Incidents.

Scan the label on the unit and its entire life is there. That single record is the reason any of the previous twelve was possible.

Telemetry, maintenance, documents, parts, rounds and incidents. How many separate systems is that?

One. They all hang off the asset record — and that is the only reason any of the previous twelve was possible.

CRAC09Hall C · DC-1 North
TELEMETRY6 live points
MAINTENANCE14 PMs, 3 CMs
DOCUMENTSMOP-0114, 2 manuals
PARTSbelt V-2440, filters
ROUNDS91 readings, 30 d
INCIDENTS1 open · P3
System map

And the rest, on the same record.

A reading, a work order, a part, a procedure and a message all point at the same asset. That is the only reason any of the above was possible.

Preventive maintenanceTemplates, cadences, guided checklists with captured readings.
Corrective maintenanceFaults to work orders through one governed write path.
Assets & equipmentHierarchy, history, components and telemetry on one record.
ProceduresMOP and CSP with revision control and execution hold points.
Budget & POsBudgets, purchase orders, receipts, spend reconciled.
ProjectsCapital work with milestones and linked purchase orders.
TrainingWho is qualified for what, with expiry and evidence.
ReportsMonthly business review built from measured data, hall by hall.
VendorsContracts, coverage windows, and who to call at 3 a.m.
PermissionsPer-site access, separation of duties, audit trail behind approvals.
MobileThe same system on the floor, offline-capable, scan and photo.
Deployment

Where it runs is your decision.

Built portable from the start. Pick one and see exactly what sits where — and what, if anything, crosses your boundary.

YOUR SITE
Building systems
Collector
Browsers
GRAYSPACE CLOUD
Application
Database
Backups & PITR
Local model
WE RUN
application, database, backups, patching, monitoring
YOU RUN
a collector, if you want live equipment data
CROSSES
one outbound HTTPS connection from your side. Nothing inbound.
BEST WHEN
you want it live in weeks with nothing new to operate
YOUR ESTATE
Building systems
Collector
Application
Database
File storage
Local model
Browsers
GRAYSPACE
Release feed
Support
WE RUN
nothing of yours — we publish releases and answer the phone
YOU RUN
the whole platform, on hardware you own, under your change control
CROSSES
signed update packages, pulled by you when you choose
BEST WHEN
data residency or change control makes hosting a non-starter
YOUR ESTATE — NO EGRESS
Building systems
Collector
Application
Database
File storage
Local model
Browsers
GRAYSPACE
Unreachable
Updates by media
WE RUN
nothing. We cannot reach it, by design.
YOU RUN
everything, including the model — the AI is local or absent
CROSSES
nothing over a network. Updates arrive on approved media.
BEST WHEN
the site has no outbound path and never will
Get started

Find out what your thresholds aren't telling you.

Your sites, your assets, your procedures — configured with your team, live in about two weeks.

GRAYSPACE · CRITICAL FACILITIES OPERATIONS
MOCKUP NOTE — the console is a simulation, but the detection is the real sensor_flatline rule from the product. Screenshots are the 7 July captures and still carry the SYSTEM DEGRADED pill fixed on 26 July. Type is a system stack; the build uses Bricolage.