Eleven calls to make before the doors open.

CrossVenue reads a building and turns it into the calls somebody has to make today, each one answered before it is explained. What to run tonight, what is broken, what it costs in electricity and in water, and what the last thirty nights would have cost if they had been run differently. A concert hall, a hospital, an office tower, an arena: anywhere a plant runs against a calendar.

Read-only, and stricter about it than the rest of this family has to be. It never writes a setpoint, and every source it reads is asked for an account that cannot write one.

Click through the demo 1 min Contact us

The roof call, at four in the afternoon

Call: Closed. It failed 4 of the checks in the club’s own rule: humidity, too warm, wet floor, low sun on the seats.
The one that decides it: The concourse floor is 8.8 F below the temperature at which the air above it starts to sweat. That is condensation, and it is a slip hazard on a night with 29,900 people walking on it.
What it costs: Holding it closed costs roughly 32,760 kWh of cooling over the 4 hours it would otherwise have been open.
Do this: Keep it closed. The reason to give is humidity.

Nobody wrote that. The engine did, and the demo below opens the arithmetic under the third line.

The problem

Every system in the building has a screen. None of them has an answer.

The controls have 1,750 objects and 430 trend logs. The meter has an interval file. The maintenance system has a work queue, the calendar has tonight’s event, and the weather service has an hourly forecast. All of it is true, all of it is on a different screen, and none of it says what to do at four in the afternoon with a full house coming in at five.

So the answer comes from whoever has been there longest. That works, right up until they retire, or until somebody upstairs asks what the building costs and why, and the honest reply is that it costs what it costs.

The gap is not measurement. Buildings have been measured for thirty years. The gap is that measurement was never turned into a decision anybody could defend.

See it work

Four in the afternoon, and there is a full house tonight.

A real run: eleven calls, seventeen things broken, a number in dollars, and a search across thirty recorded nights. The building here is a ballpark, because that is the venue the engine ships a sample for. It is a real one and its name is withheld, so the physics are somebody’s and the identity is nobody’s.

CrossVenue Demo data

Start here. You will take a decision, open the arithmetic under it, read what is broken, look at the money, and see what the search refused to recommend. It takes about a minute.

CrossVenue venue operations Sample

Sample data, from a real venue whose name is withheld. Every figure is the engine’s own output.

Your first month

Almost nothing was connected, and it still gave you a number. Why is that not a problem?

Because the number arrives already marked with what it rests on. A venue on day one has a controls network, a meter nobody exports, and a utility statement in somebody’s inbox. CrossVenue runs on that, and it is loud about which half it is standing on.

It counts the live feeds. It does not claim them. The top of every screen carries how many of the nine sources are really connected, and this sample says zero. A source that is standing in is labelled on its own row with what it is standing in for. The weather is the one feed that needs nothing from the site, so it is the one that connects itself, and until it answers the sample sky stands in and says so.

A figure resting on an assumed rate grades itself down. Of the $1,565,888 on the money screen, $1,556,398 rests on a rate nobody has confirmed, and the screen is graded Indicative until a statement arrives. Twelve of the sixty findings carry a figure at all. The other forty-eight say they cannot be priced yet, and the product is clear that unpriced is what that makes them. Small is a different word.

Every gap names the answer it is blocking. Seven things are missing, and each one is written as a job with a beneficiary: three temperature loggers at seat height for one event turns every comfort figure from modeled to measured. That is an ask a facilities manager can act on this week. It is not a roadmap.

What the search refused to recommend

Tried: 7,140 settings, against each of 30 recorded nights. 18 of those nights had a crowd in the building and 12 did not.
Answer: One standing setting for every event night, worth $47,295 across the window. It holds the seats inside the band on all 18 of them, and the warmest any seat gets is 77.0 F.
Refused: 4,223 of the 7,140 left a seat too warm and were dropped. The cheapest setting in the whole search would have saved a further $3,505, and on at least one of those nights it leaves the warmest seat at 84.8 F with people sitting in it.
Printed anyway: “That figure is here so nobody has to wonder what was hidden, and it is not a recommendation.”

A comfort constraint whose cost is never shown is a constraint nobody can argue with. This one shows what it cost, and then holds.

Connect one source and the grade moves. One interval export from the delivery company replaces every demand figure in the product with a measured one. One utility statement moves the money screen off assumed rates. The product names both asks, in the same words, on the screen that is waiting for them.

How it works

One report, and every screen is a view of it.

The answers hang together because they are the same answer looked at from different sides. Each screen calls one builder, and that builder is a pure function of the feeds, so two screens cannot quietly disagree about what tonight is.

It writes the record itself Recording runs on a timer, on render, and from the command line for a machine that is switched off out of hours. All three are the same idempotent call. A day already on record is rewritten when the feeds now describe it differently or when the arithmetic has moved on since it was written, and that last case is stamped on the row. Run it twice and it writes nothing the second time.
The search is a full grid, not a nudge Four levers across their whole range, against every night on the record. Energy adds up across the window; the demand charge and the summer peak charge are each billed once, on the highest interval anywhere in it. Days with nothing on are held exactly as they ran, because a quiet Tuesday sets a floor under the peak charge that no amount of managing the busy nights gets under.
Comfort is a constraint and never a term in the cost The search drops anything that leaves the seats above the band and names what it refused. Holding demand down is priced against the cooling it costs, which sounds obvious and was wrong for a while: the shed came off the billed demand and never off the cooling, so the search shed more than the building was drawing and reported a saving with no consequence attached.
One pot of money is counted once Four findings reach the same four billing intervals. Added up they say $2,234,830, and the site can bank that pot once, so the total is $1,565,888 and the excluded rows are shown at full size with the reason they are excluded.
A comparison is matched before it is offered Two nights differ in everything at once. The second night is chosen on what the building was asked to do: the crowd first, then the air, the humidity, the sky and the time of day. A full house is never matched against an empty day, and where nothing on record is really comparable it says so rather than picking the least bad.
The sky is read, and it is kept Forward, the hourly gridpoint forecast for the square the building sits in. Backward, what the nearest station recorded. Each day’s sky is written onto the record as that day happens, because it cannot be fetched again: the service publishes about a week either way and nothing at all about last April. A night replayed without its sky is a night replayed under a sky somebody assumed.

The honesty rule

It stores readings, and works the conclusions out when you open the page.

Every figure on a CrossVenue screen is worked out at the moment you open it, from the record and the feeds. So a number is always as fresh as the reading under it, and the question of when the dashboard last ran has an answer on the screen.

Saved plans are the one exception, and they earn it. Opening one runs the same search again and the page says whether the two answers still agree. A plan saved on Tuesday that no longer holds on Thursday is the most useful thing this product can tell you, and it can only tell you because it kept the question instead of the answer.

The limits

It runs one building, and it needs that building to answer some questions.

This is a venue product, sized for one site at a time. A concert hall, a hospital, an office tower, a convention center, an arena, a ballpark, a campus with a central plant: what it needs is a building with a plant big enough to argue about and a calendar that says when people will be in it. It is a single-site tool, and a portfolio stays a set of separate installs. Nine data sources are described in it and the weather is the only one that can be reached without asking the operator for anything, so the first month is a series of requests, and the dashboard fills in behind them.

Some figures wait on a person. The energy and demand rates stand as placeholders until the site supplies a statement, and every screen that prices anything says which of the two it is using. Seat temperatures are modeled until somebody puts three loggers at seat height for one event. The product asks for both by name.

And it advises. It reads the controls, the meters and the calendar, works out what to do, and hands it to a person who does it. Every path through the software ends at that handover: there is no code in it that writes a setpoint, opens a roof or starts a chiller, and the accounts it asks for on the controls network are read-only ones.

Bring us a night that cost too much.

Whatever the building is, give it a calendar, a controls network and one utility statement, and it will tell you what that night should have been run at and what the difference was worth.