Sustainability Interoperability Layer for Operations (SILO)

One language for the whole data centre.

The problem. A rack inlet temperature, a pump flow rate and a node power draw arrive from three vendors, in three schemas, with inconsistent units and sampling rates, and with three different ideas of which asset they describe. The existing standards aren’t the problem. Redfish and Swordfish manage IT equipment and, increasingly, power and cooling. BACnet and Modbus monitor and control facility plant. Project Haystack and Brick Schema model building assets. Each works within its own domain, but none defines what its values mean relative to the others. As a result, every cross-domain optimisation needs bespoke integration, vendor by vendor and site by site.

The pressure. Regulation is outpacing the data. Under the EU Energy Efficiency Directive (2023/1791), every data centre with at least 500 kW of IT power demand must report sustainability data every year. In September 2026 the Commission published its assessment of that data under Article 12(5) and opened a consultation on minimum performance standards, with a legislative proposal planned for 2027. The assessment found significant data quality problems: only 70.1% of first-period data was deemed reliable. Operators will be asked to compare and improve numbers that are hard to trust, and consistent, machine-readable definitions are part of the fix.

The answer. SILO is a semantic layer built on GSF’s WDPC Data Bus. It defines a canonical data model, giving every metric a fixed meaning, unit, metadata and asset context. It translates and validates telemetry from existing protocols into that model, and it exposes the result through standard APIs for real-time and historical access. Existing protocols stay in place; SILO maps across them, so you integrate once against a common model rather than once per vendor.

What this Assembly will examine

Five topics, run in order. Each closes before the next opens and does not reopen.

1. Conformance boundary
2. Where normalisation happens
3. Minimum viable metric set, units and resolution
4. Trust, provenance and validation (CSRD / EED)
5. Gaps and priorities

To keep this finishable, the first blueprint is constrained to the minimum set of definitions supporting three use cases: adaptive cooling, dynamic workload placement, and automated PUE, WUE and CUE reporting. Everything else routes through governance afterwards. The data centre to grid interface sits at a different layer and is out of scope, covered separately in Bridging the Sustainability Gap.

The Assembly does not define the specification. Its output is a Specification Baseline Report: a map of the problems and open questions the Hardware Standards Working Group will then take on.

Key dates. Applications open Wednesday 15 October and close Friday 6 November. Participants are confirmed, and the question set frozen, on Wednesday 11 November, so round topics may be refined before then. The working group kicks off in the week of 19 January 2027.

What we ask, and what you get back

This is boundary-drawing work across six domains that have never shared a vocabulary. A clear position on where you would draw a line is worth more than agreement, above all when it differs from the group.

In return:

• Your responses stay private. Only anonymised synthesis is shared back, so you can be candid about what your systems can and cannot do without disclosing proprietary internals.
• Your voice carries equal weight, regardless of seniority, organisation, or whether you are a GSF member.
• You see what your input changed. Every synthesis opens with the reasoning that actually moved a position.
• Named contributor credit in the Specification Baseline Report, opt in.
• A review before submission, to confirm the draft reflects what the group agreed.

Decide rounds record the group's position and any reservations. They do not bind the working group. If you do not respond within the window, we record an abstention.

Who should be in the room

• Facility engineers and infrastructure managers running mixed-vendor estates
• Cooling, power, rack and sensing manufacturers, particularly firmware and telemetry teams
• DCIM, BMS and optimisation platform builders paying the per-vendor integration cost today
• Practitioners from Redfish, Haystack, Brick Schema, ASHRAE TC 9.9 or OCP who know where the model boundaries sit
• Sustainability leads responsible for EED and CSRD reporting across a fleet
• Grid operators and energy platforms working on data centre flexibility
• Data centre operators and co-location providers
• Electrical design engineers and EPC firms who design power distribution
• Telemetry providers who supply the instrumentation, schemas and data feeds
• ESG and policy specialists working on how reported data is classified

You do not need to span the whole problem. Nobody does. Contribute where you are strong and stay open where someone else is.

This Assembly is open to non-members. If you are not from a GSF member organisation, you will need to complete a participation agreement before the first round.

How an Assembly works

This assembly runs primarily over email. There are no required meetings. The work happens asynchronously, on your own schedule. Optional drop-in sessions run throughout the process for anyone who wants to connect with the group. Nothing to install, no new tools to learn. You just reply to emails.

Questions go out, you reply, and Harmony (GSF's AI-assisted consensus engine) synthesises everyone's input into shared positions that the group refines and reaches consensus on. The process is designed so no one gets left behind. It specifically seeks solutions that the least-satisfied person can live with.

DISCOVER: What does everyone think?

Each round starts with a question. You reply by email with your honest view. No turn-taking, no live discussion, no one drowning anyone else out. Harmony reads every response, anonymises it, and produces a synthesis: key themes, where people agree, where they diverge. You see what the group thinks, not just what you said.

DELIBERATE: Which approach works best?

Harmony drafts candidate positions based on the synthesis. You rate each one from 0–10. A candidate only becomes viable when even the lowest score reaches at least 5. The process is explicitly designed to find answers everyone can live with, not just what the majority prefers. Rounds continue until a viable candidate emerges.

DECIDE: All aboard?

The strongest candidate goes to a final consensus check. You choose one of three responses: ENDORSE (full support), CONSENT (you can live with it), or OBJECT (a concern that must be resolved before the process moves forward). A single OBJECT from any participant halts consensus. Your voice carries regardless of which organisation you represent.