Marketing Wants to Call It Tier III

The claims are not defensible as written. Concurrently maintainable means every component - power path, cooling path, and the distribution serving the IT load - can be removed from service for planned maintenance without the IT load being affected. Test each system against that definition:

• Utility supply: two feeders from the same busbar share a single upstream failure domain. Diverse feeds must originate from independent substations or at least independent busbar sections. Fails.
• Chillers: N+1 capacity is fine, but common headers with manual isolation valves mean isolating a chiller or a header for maintenance requires a partial or full shutdown sequence. Fails - this is the classic gap between N+1 and concurrent maintainability.
• UPS: a maintenance bypass that transfers to raw mains exposes the load to utility disturbance during UPS maintenance. Fails.
• Dual cords to one output board: the cords are diverse at the rack but converge immediately; a board or upstream failure takes both paths. Fails the fault-isolation intent even though it looks compliant.
• Generators N+1 with ATS: likely acceptable, but only if the ATS and fuel system are also maintainable without load impact - verify, do not assume.

Also note the standards framing. "Tier" is a proprietary certification mark of a specific institute and cannot be self-declared; EN 50600-1 and ISO/IEC 22237-1 define availability classes 1–4 on a different basis. The 2019 consultant report is not a certificate and is now stale. Defensible language today: "N+1 with partial fault tolerance; concurrent maintainability available for generator and chiller plant only, subject to documented procedures."

Prioritisation principle: fix what changes the tenant's actual risk of unplanned downtime first, then what unlocks the claim, then what is merely presentational. Cost per unit of risk removed is the ranking metric.

Tier 1 - within three weeks, low capex, high credibility:

1. Re-cord or re-patch racks onto two independent output boards where the topology allows, or formally document which racks cannot be dual-fed and disclose them.
2. Write a maintenance philosophy document: for each system, state whether maintenance is online, requires a window, or requires shutdown, with the procedure reference. Disclosure beats an unsupportable claim.
3. Correct all customer-facing material to "N+1, partial concurrent maintainability" and withdraw the 2019 language from circulation.

Tier 2 - 3 to 9 months, unlocks the claim for cooling: motorised isolation valves and a loop layout that lets a chiller be isolated without dropping the header. This is the highest-value capital item because cooling is the most common planned-maintenance driver in a colo.

Tier 3 - 9 to 24 months, structural: a second UPS block with independent output distribution (A/B), and utility diversity to a separate substation. Both are major capital and need board approval; promising them in a tender is not acceptable.

Interim wording for the tender: describe the actual topology, name the maintenance windows required for the UPS and chiller headers, state availability as a historical measured figure with the measurement period and method, and offer the tenant a walkthrough plus the maintenance philosophy. A tenant's engineer will find these gaps anyway; disclosing them converts a disqualifying discovery into a credible roadmap, whereas an unsupported Tier claim is a contractual misrepresentation exposure that survives long after the deal closes.

What the incident reveals. The outage was not a technical surprise - it was a predictable consequence of the gap that was disclosed or should have been. The governance failures are:

1. Claim drift: if any contract schedule, SLA annex or sales deck still described the site as concurrently maintainable, the disclosure in the tender was overridden by later paperwork. Contractual documents must be reconciled against the corrected technical position.
2. No maintenance notification regime: the tenant appears not to have been told which works affect their load. A maintenance calendar with defined notice periods, affected systems, and expected risk is a basic colo obligation.
3. Procedure not validated: a maintenance activity that can affect customer load should be rehearsed or peer-reviewed, with a go/no-go checklist and abort criteria.
4. Change control boundary: planned works on shared infrastructure need customer-facing change records, not just internal work orders.

Controls to insist on now:

• A contractual schedule listing, system by system, whether maintenance is online or requires a window, with maximum notice and annual window limits.
• Joint change advisory sign-off for any work on shared cooling or power paths, with the customer's right to defer once.
• Availability measured and reported monthly with a defined method, plus root-cause reports within a stated period for any event affecting the customer.
• A remediation plan with dates for the Tier 2 valves and Tier 3 electrical diversity, contractual but with relief if the customer declines the associated rate increase.

The general lesson: compliance claims are governance artefacts, not marketing copy. When the technical position changes, every document that repeats the old claim must be found and corrected.