1. Executive Summary

Data centers cannot prove their carbon numbers. That is true, and it matters, but it is not the argument that gets a project built.

The argument that gets a project built is this: a data center's largest near-term risk is not its emissions accounting. It is its license to operate. Water, noise, and air are what stop projects, void rezonings, and generate lawsuits. In 2026 that stopped being a forecast and became a docket.

The problem in every one of those disputes is the same, and it is a measurement problem. When a neighbor's well goes bad, when the hum starts, when the air smells different, the operator has no defensible record of what conditions were like before the facility existed. Without a before, there is no after. Causation becomes a battle of retained experts, and the operator, who chose the site and built the thing, loses the narrative by default.

Mālama's position is narrow and specific:

Deploy the sensor network before you break ground. Sign every reading in hardware. Anchor it publicly. Then keep measuring for the life of the asset.

The baseline is not a study you commission and file. It is the first phase of a continuous instrument that runs from pre-construction through decommissioning, producing one unbroken, tamper-evident record. A consultant's sound study is a snapshot an opposing expert can attack. A continuously anchored record is a different category of evidence.

Four sensing domains, one trust architecture:

The same secure element signs all four. The same pipeline anchors all four. The incremental cost of adding a domain is the sensor, not the infrastructure.

---

2. The Risk Surface Moved in 2026

This is not a projection. These are filed cases.

Two observations that matter more than the case list.

First, the plaintiffs' bar has found the sector. The claims are old tools, nuisance and negligence and trespass, pointed at new infrastructure. Crypto mining litigation is the precedent, and Texas courts have already declined to dismiss noise nuisance claims at the pleading stage. The theories work.

Second, and this is the part that should interest an operator: the defense bar's own recommended mitigation is a Mālama product. Counsel advising data center developers list, among the top defensive measures, baseline and post-construction sound studies, confirmation that on-site generation is fully permitted with monitored emissions controls, and transparent disclosure of resource use to the community.

That is not Mālama's marketing claim. That is what defense counsel is telling clients to do. The gap is that the recommended practice is currently executed as a pair of consultant snapshots, which is the weakest possible version of it.

---

3. The Baseline Problem

Here is the structural failure, stated once.

A nuisance or contamination claim turns on causation. Did the facility cause the change? Answering that requires knowing the prior condition. Almost no data center has a rigorous record of the prior condition, because nobody instruments a green field.

What exists instead is a pre-construction study: a consultant visits for a few days, takes measurements under whatever conditions prevailed, and issues a report. Then the facility is built. Then a complaint arrives. Then a second consultant measures again.

Every part of that is attackable. Sampling was too short to capture seasonal, diurnal, or weather variation. The equipment class was insufficient. The locations were chosen by the developer. The report was commissioned by the party it exonerates. The two studies used different methods. And critically, nothing about either report is tamper-evident, so an opposing expert can question not only the methodology but the integrity of the underlying data.

Compare a continuously anchored record:

• Duration. Twelve months of pre-construction data across every season, weather condition, and time of day, rather than a three-day snapshot.
• Continuity. The same instruments in the same locations before, during, and after construction. No methodology discontinuity to attack.
• Independence. Every reading signed at the point of measurement by silicon the operator does not control and cannot alter, with a public timestamp. The operator cannot have adjusted it, and neither can Mālama.
• Falsifiability. This cuts both ways and that is the point. If the facility did cause the change, the record shows it. An operator who deploys this is making a credible commitment, which is precisely why it carries weight with a regulator, a judge, or a community group. Evidence you could not have faked is worth more than evidence you merely say you did not fake.

The honest framing for a community meeting is not "our data proves we are clean." It is "we instrumented the site before we built, we cannot alter the record, and you can check it yourself." That is a fundamentally different conversation than the one most operators are currently having.

---

4. The Four Sensing Domains

4.1 Power and carbon

Two devices. The MRAA-01 is a 2U rack-mount attestation appliance that replaces or sits upstream of the rack PDU. The MGM, the Mālama Green-Grid Meter, sits at the service entrance and at any behind-the-meter generation tie-in.

The metering class is the point. IEC 62053-22 Class 0.5S is revenue grade, the tier utilities bill against. It is the difference between a number you can transact on and a number you can only report.

Outputs, computed continuously and signed: per-rack kWh, continuous measured PUE, hourly location-based and market-based Scope 2, per-workload attribution where the utilization feed is available, and the hourly carbon-free energy gap.

That last one is the one that becomes a contract line item. The regulatory driver is the GHG Protocol Scope 2 revision, which is moving the market-based method to hourly matching and deliverability, with a revised standard anticipated in late 2027 and phased adoption after. Hourly matching is a data problem before it is a procurement problem. Under that regime the CFE gap is both the number you procure against and the number you are exposed on. Estimated, it is a liability. Measured, it is a hedgeable position.

4.2 Water use and quality

Water is the fastest-moving siting constraint in the sector, and unlike carbon it is locally felt, immediately visible, and politically potent.

Scale, for context. US data centers used roughly 449 million gallons per day as of 2021. Average water usage effectiveness sits near 1.9 litres per kWh. Roughly 80% of withdrawn freshwater evaporates rather than returning to the local system, which is why "withdrawal" and "consumption" are different numbers and why using the former in community materials reads as evasion. Loudoun County's roughly 200 facilities consumed about 900 million gallons in 2023.

MWN-Water sensing stack:

The groundwater wells are the item most operators skip and the one that matters most in litigation. A drawdown or contamination claim from a neighboring residential well is unanswerable without a monitoring well network established before pumping began, with off-site control points that isolate site effects from regional trends.

Note on the regulatory picture: state-level water rules for data centers are proliferating quickly, including reclaimed water mandates, groundwater certificate restrictions, and public water-use dashboards for larger facilities. The specifics vary by state, move fast, and should be verified against the current rule for the jurisdiction in question rather than any summary, including this one.

4.3 Air quality

The air issue is not the data hall. It is combustion: backup generators, and increasingly, on-site turbines installed to bypass interconnection queues.

That shift is what created the NAACP citizen suit against xAI over unpermitted turbines. Under the Clean Air Act, stationary combustion turbines and reciprocating engines are subject to NSPS and NESHAP requirements, and the permitting question turns on potential to emit, on whether emergency-use limits are respected in practice, and on whether adjacent units aggregate into a single major source. EPA maintains guidance specific to data centers and has stood up a dedicated contact for permitting authorities and operators.

The operational trap is the emergency-use limitation. A generator permitted as emergency-only carries hour limits. Run it for peak shaving, grid services, or extended commissioning, and the permit basis can fail. Most operators track generator runtime in a maintenance system that was never designed to be evidence.

MWN-Air sensing stack:

Two design notes a technical reader will want.

Sensor tier honesty. Low-cost electrochemical and optical sensors are non-regulatory supplemental and informational monitoring. EPA publishes performance testing protocols and target values for this tier. They are appropriate for fenceline networks, trend detection, complaint response, and community transparency. They are not federal reference or equivalent methods and cannot substitute for FRM/FEM instruments where a permit or NAAQS determination requires one. Any Mālama material that blurs this will be caught, and should be. The correct architecture is a dense NSIM-tier fenceline network anchored by one or two reference-grade instruments for calibration transfer and for the measurements that carry regulatory weight.

The met station is not optional. Air quality data without co-located, synchronized wind data cannot support or refute a source attribution claim. It is the cheapest component and the one that makes the rest admissible.

4.4 Acoustic

Noise is generating the most litigation, and it is also the domain where continuous measurement most decisively outperforms the consultant-study status quo.

Data center noise has a specific character that makes it unusually actionable. It is continuous rather than intermittent, so there is no respite. It is tonal, carrying discrete frequency components from fans and transformers that the ear locks onto. And it is weighted toward low frequency, which propagates further, penetrates building envelopes more readily, and is systematically understated by A-weighted measurement. Plaintiffs in the 2026 cases describe exactly this: high-pitched squealing, continuous roaring, low-frequency rumbling, and tonal humming, at distances up to about a mile.

An ordinance written as "55 dBA at the property line" does not capture any of that. An operator can comply with the ordinance and still generate a viable nuisance claim, because nuisance turns on unreasonable interference, not on the ordinance.

MWN-Acoustic sensing stack:

What the baseline produces: a full year of L_90 residual background at each receptor, across every season and weather condition, before anything is built. When a complaint arrives, the question "how much did the facility change the sound environment at this location" has an arithmetic answer instead of an expert opinion.

Data handling is addressed separately in the security brief. Briefly: these are sound level meters, not recording devices. Computed metrics are transmitted; raw audio is not retained or transmitted, and the optional event-classification clip feature is off by default.

---

5. MSB-01: The Site Baseline Package

The pre-construction deployment. A bundle, not a study.

5.1 Composition

Typical configuration for a single campus. Scales with parcel size, receptor count, and hydrogeology.

5.2 Timeline

The twelve-month requirement is the hard part of this product. It means the baseline package has to be sold and deployed roughly a year before groundbreaking, which is earlier than most developers are thinking about environmental monitoring, and often before land is fully controlled. That is a real constraint, not a detail. It also means an operator who calls after a complaint has already arrived cannot be helped in the way that matters, and we will say so rather than sell a weaker substitute.

A shortened deployment of six months captures two seasons and is meaningfully better than nothing. It should be positioned as a compromise, with the limitation stated plainly rather than papered over.

5.3 What the baseline report contains

1. Statistical characterization by domain, by season, by hour of day, by weather condition.
2. Residual background L_90 at each acoustic receptor, with tonality and low-frequency characterization.
3. Air quality distributions with wind-sector conditioning, so downwind and upwind conditions are separable.
4. Groundwater elevation trends, seasonal variation, and baseline chemistry with detection limits stated.
5. Verification instructions: how any third party independently confirms every figure against the on-chain record without access to Mālama or the operator.
6. Explicit statement of limitations, including sensor tier, detection limits, and any gaps in coverage.

Item 6 matters more than it looks. A report that overstates its own certainty is worse than no report, because it hands the other side an impeachment.

---

6. Full Lifecycle

The same physical network, the same signing keys, the same anchored chain, across six phases. Continuity is the product.

Phase 1 · Pre-construction baseline

Duration: 12 months minimum, before ground disturbance. Deployed: MSB-01 environmental package. No power infrastructure yet. Produces: The reference condition. Every later claim is measured against this. Buyer: Development, land-use counsel, community relations.

Phase 2 · Construction

Duration: 18 to 36 months. Deployed: Baseline network continues, uninterrupted. Add construction-phase nodes: dust at active earthwork, vibration at the nearest structures, stormwater discharge quality and volume. Produces: A signed record of construction-phase impacts, separable from operational impacts. When a complaint arrives during construction, and it will, the record distinguishes pile driving from permanent operations. That distinction is worth a great deal, because construction impacts are temporary and legally treated very differently. Buyer: EPC contractor, construction manager, counsel.

Phase 3 · Commissioning

Duration: 3 to 9 months. Deployed: MRAA-01 and MGM come online. Environmental network continues. Generator testing is instrumented for both runtime and emissions. Produces: The single most valuable dataset in the lifecycle. Commissioning is the only period when equipment is operated in controlled, isolated configurations. Run the chillers alone, measure. Run the generators alone, measure. Run at partial IT load, measure. This produces a per-source acoustic and emissions signature that makes later source attribution a lookup rather than an argument. Buyer: Commissioning agent, operations, counsel.

This phase is under-appreciated and worth building the product around. No other point in the facility's life allows clean source separation. If commissioning passes without it, that signature can never be recovered.

Phase 4 · Operations

Duration: 15 to 25 years. Deployed: Everything, continuously. Produces: Hourly Scope 2 and CFE gap for reporting and procurement. WUE and consumption for permits and community disclosure. Continuous fenceline air and acoustic monitoring against the baseline. Generator runtime against permit limits. Automatic flagging when any parameter exits its baseline envelope. Buyer: Operations, sustainability, compliance, legal.

The operational value that is easiest to underrate: complaint response time. A complaint arrives naming a date and time. With this system the operator answers within the hour, with signed data, showing exactly what conditions were at that receptor at that moment and how they compared to baseline. Today that same question takes weeks, a consultant, and produces a contestable answer. Fast, verifiable, good-faith response is also the single most effective way to keep a complaint from becoming a class action.

Phase 5 · Expansion

Duration: Recurring. Deployed: Existing network becomes the baseline for the next phase. New nodes for the expansion footprint. Produces: A pre-existing, unbroken, multi-year record when the expansion goes before the planning commission. Compounding advantage: an operator who instrumented phase one walks into the phase two hearing with years of verified data while a competitor arrives with modeling. Given that the Prince William County rezoning was voided on procedural grounds and that permitting friction is now a primary constraint on capacity growth, this may be the largest commercial value in the whole system. Buyer: Development, planning, community relations.

Phase 6 · Decommissioning

Duration: Terminal. Deployed: Environmental network continues past shutdown. Produces: Post-closure monitoring demonstrating return to or divergence from baseline. Closes out permit obligations, supports remediation scoping if needed, and bounds long-tail liability at asset transfer. Devices are cryptographically retired, with the DID revoked on-chain so no post-retirement reading can be attributed to them. Buyer: Asset owner, counsel, acquirer diligence.

---

7. The Trust Chain

One chain, identical across all four sensing domains.

1. Provision. Keypair generated inside the ATECC608B at manufacture. Private key non-exportable, unreadable by anyone including Mālama. Public key becomes the device DID, bound to serial, manufacturing attestation, and calibration record.
2. Measure and sign. Sensor produces a sample. The payload, measurement plus monotonic counter plus timestamp, is hashed and signed by the secure element before touching any network interface or filesystem.
3. Transport. Outbound mutually authenticated TLS. Transport is not the trust boundary. A compromised network can drop or delay a reading, which is detectable, but cannot alter one.
4. Anchor. Merkle root written to Cardano every 60 seconds. Removes Mālama from the trust path.
5. Verify and export. Any third party with a reading and its inclusion proof verifies independently, with no access to Mālama systems.

The precise claim, stated narrowly: this proves a specific piece of silicon produced a specific reading at a specific time and that it has not been altered since. It does not prove the sensor was correctly sited, correctly calibrated, or appropriate for the measurement. Those are handled by the controls in section 8. Anyone presenting the cryptography as a complete answer to data quality is overselling, and a competent opposing expert will say so.

---

8. Measurement Integrity and Honest Limits

8.1 Instrument tiers

Different domains carry different evidentiary weight, and conflating them is the fastest way to lose credibility.

Design the network with the tier stated for every parameter, and say so in the report. An operator who claims regulatory-grade air monitoring from a fenceline sensor network will be impeached on it, and the impeachment will contaminate the domains where the instrumentation genuinely is reference grade.

8.2 Controls

8.3 Stated uncertainty

Every reported figure carries its uncertainty. Power at roughly ±0.5%. Acoustic at Class 1 tolerances. Grid carbon intensity at roughly ±5% locational, which dominates any location-based Scope 2 figure. Supplemental air sensors carry substantially wider bounds that vary by pollutant, concentration range, humidity, and calibration age, and should always be reported as ranges.

---

9. Compliance and Evidentiary Mapping

The read: the carbon drivers are slipping while the local environmental drivers are hardening. Water, air, and noise are where the near-term exposure sits. Carbon is the longer arc, and the Scope 2 revision will matter, but it is not what stops a project in 2026.

---

10. Economics

10.1 Indicative pricing

Environmental node pricing below is a first-pass internal estimate and is not validated. Power pricing carries forward from the existing model.

10.2 Illustrative campus

MSB-01 baseline package, mid-size campus, eight acoustic nodes, five air nodes plus one reference anchor, three surface water, six groundwater, one met mast, two gateways:

• Hardware and installation, excluding wells: roughly $200K to $300K
• Well installation, six wells: roughly $90K to $240K
• Twelve-month baseline SaaS: roughly $12K to $18K
• Total pre-construction: roughly $300K to $550K

Operational phase adds MRAA-01 at scale. A 1,000-rack facility is roughly $1.5M hardware and $200K per year recurring, on top of the environmental network at roughly $15K to $20K per year.

10.3 The comparison that matters

Do not benchmark this against a consultant's baseline study, which costs $30K to $80K. That comparison loses, and it is the wrong frame anyway.

Benchmark it against:

• A single nuisance class action. Defense costs alone run well into seven figures before any judgment or settlement. The Oregon groundwater matter settled at $20.5M.
• A rezoning delay. A voided or delayed approval on a project of this scale costs more per month of delay than the entire monitoring program.
• A permit condition triggered by absent data. Regulators facing uncertainty impose conservative conditions. Conservative conditions on cooling capacity or generator hours have permanent operating cost consequences.
• A stalled expansion. The compounding value in phase five is probably the largest number in this document and the hardest to put a figure on.

The framing for a developer: this is siting insurance with an operational dividend, not an environmental monitoring line item. Price it and sell it against the risk budget, not the facilities budget.

---

11. Risks and How We Handle Them

Honest about what can go wrong with a deployment, and what we do about it.

---

---

12. Next Step

If you are 12 to 24 months from groundbreaking, the useful conversation is a scoping call. One hour, no cost. We establish where you are in the lifecycle, what is actually contested at your site, and whether we can help. You will get a straight answer, including no.

Pricing is illustrative throughout this document. For full scope and pricing based on your needs, book a scoping call.

pilots@malamalabs.com

---