When the mixture hits a live bus, is the first event a fire? The fire and life-safety basis files a glycol release as a release scenario rather than a fire scenario, then grades its electrical consequence above its combustion consequence. The propylene-glycol mixture conducts, so contact with an energised busbar, distribution unit or rack incomer produces a short circuit and arcing. The grading is bounded on one side, and the same table says why: the release upper bound cannot be computed at all, because the system fill volume has no fire-design input, so what is ranked above the combustion case is a combustion severity nobody has yet been able to assess.
What follows from the grading is a detection-placement rule rather than a suppression requirement. If the value of leak detection is to isolate the liquid before it reaches live parts, detection belongs upstream on the liquid path and at low points, and not spread evenly across a floor. The same basis counts the places a release can start: roughly 186 rack branches, 24 distribution-unit injection points and 48 ring segments, with the whole set moving about 40 per cent either way as the rack count moves [TARGET]. Its escalation chain runs point leak, local pooling, spread along under-floor and tray routes, contact with live parts, electrical fault, and only then possibly a fire.
A Leak Can Fault the Bus; a Fault Can Open the Same Joints
The bonding basis reaches the identical interface from the electrical side and reverses the arrow. Direct liquid cooling puts a metal fluid network through every rack, the distribution units, the primary side and the outdoor heat-rejection equipment, which ties all of them into one electrical body. That network is a useful auxiliary equipotential path. It is also an unintended path for fault and lightning current, and the points where such a current would be forced through pipework and joints never designed to carry it are the same points that hold the liquid seals. One document says a leak can cause an electrical fault; the other says an electrical fault can cause a leak; the geometry involved is one set of joints.
The grounding chapter states the same geometry as a connectivity requirement. The grounding system is the one system that must, by design, connect the plant. Electrical, A/B, feeder and cooling failure domains collapse onto one earth. Because the fluid itself conducts, the assumption the bonding basis calls common, that fitting an insulating joint achieves electrical separation, is not sufficient on this system, and it refers the point to the record engineer. That is the consequence with the largest effect on layout. No conductivity figure is published anywhere, because the safety data sheet has not been obtained.
Cooling Gave a Fill Band; Fire Still Lacks a Release Bound
The fire chapter still writes that the cooling basis never gave a system volume, and on that ground grades fill as empty. The cooling basis did give 25 to 45 m³ at about 40 per cent uncertainty under an assumption flag [MODELED], enough to size an expansion vessel at the upper bound and not enough to purchase fluid, and not the maximum single-break release the fire chapter needs. Citing the fire chapter as "volume unknown" and the cooling chapter as a frozen inventory are both wrong. The five-term sum the fire chapter writes in order to lift its own empty grade is a formula. Every term in it is still empty.
The chapter divides its own subject on this line. The physical release consequences do not depend on the flammability judgement and can be delivered in full now; the judgement itself cannot, and neither can the release bound. The flammability classification carries no obtained input.
Leak-Detection Placement Does Not Wait on Fill or Flammability
Detection layout waits on none of that. Neither does the cost of getting it wrong in the other direction, which the cooling basis has already priced. A false detector action that closes a rack branch valve leaves a fully loaded 130 kW rack without cooling within tens of seconds [TARGET], and whether that action is automatic at all is an owner decision still open, with both branches carried in parallel until it is taken. The facility must not default.
Two disciplines have now asked the cooling side for two different missing fields on the same pipework: a five-level volume table, and the material of every segment together with the position and electrical continuity of every insulating joint, hose and quick disconnect. The bonding basis asks that the two be answered together, which costs the cooling side one reply instead of two. Neither request carries a date.
Should the sheet return a conductivity too low to sustain fault current at the distribution voltages present in the space, the placement rule loses its electrical justification and reverts to a housekeeping rule.
Limits and open items
The 186 / 24 / 48 leak-point counts and the 40 per cent swing remain topology figures [TARGET]. The 25 to 45 m³ fill band is an assumption, not a fire-design value and not a buy quantity [MODELED]. The 130 kW rack and the tens-of-seconds thermal clock remain the cooling chapter's false-trip price [TARGET].
Open: the safety data sheet; flammability classification; the five-term fill sum as a maximum-release figure; conductivity at the voltages in the space; the owner fork on auto-close versus alarm-only. Nothing here has been sealed. Nothing is a suppression scheme. [HOLD]
© 2026 K&K Data Service Inc. All rights reserved. Reproduction or republication is permitted only with clear attribution to K&K Data Service Inc. and a working hyperlink to the canonical URL of this article. Excerpts must preserve the technical context, maturity labels, assumptions, and limitations. No excerpt may imply project approval, field validation, certification, or endorsement that the original article does not state.
Source: K&K Data Service Inc., “A PG25 Release Is Graded an Electrical Fault Before a Fire,” https://www.kkdatasvc.com/lab/fire-structure-site-and-compliance/a-pg25-release-is-graded-an-electrical-fault-before-a-fire/.
Discuss This Constraint. If your project record shows a different result, or the same failure domain under another name, we want to see it. Email inquiry@kkdatasvc.com or use the contact page.