# Prescription method memo: basin one (Lake Erie drainage, Michigan portion, plus the GLWA service territory)

**Date:** 2026-07-29. **Seat:** Fable (builder seat for the wave, operator ruling
2026-07-29). **Status: PROPOSED, NOT RATIFIED.** No value in this memo reaches
`standard_values.py` or the prescription registry until the operator ratifies it
(wave contract, sequencing rule). Design authority:
`docs/AQUADRIO_PRESCRIPTION_WAVE_CONTRACT_2026-07-13.md`, as amended by
`docs/AQUADRIO_PRESCRIPTION_WAVE_BRIEF_2026-07-28.md` and
`docs/AQUADRIO_PRESCRIPTION_WAVE_SOW_2026-07-29.md`.

**The one-sentence result.** Basin one's governing regime ratifies process gates
and thresholds, not a numeric consumptive-use ceiling, so the prescription's
TRANSLATED operand is honestly absent; the DIAL operand is derivable and this memo
proposes it at **0.30 L/kWh (strict-under)**; the schema accepts a one-operand
prescription by design, and the day Michigan enacts a numeric consumptive cap
(SB 1046 would be one), revision trigger C(a) forces a version event that adds the
translated operand.

**Evidentiary posture, stated per leg (kickoff amendment, 2026-07-29):**
- *Translated leg:* would claim judicial notice through the cited ratified
  instrument. For basin one v1 the leg is absent; the instruments are documented
  in §3 so the absence itself is verifiable against primary text.
- *Dial leg:* documented to the five Daubert factors in §5, not just conversion
  arithmetic. Its parent (CAA §112(d)(3), the MACT floor) is litigated public
  methodology.

---

## 1. The method, fully parameterized

Every constant below is a memo-time proposal with its argument. Ratification
adopts or amends them as a set.

### P, the achievability anchor

- **X = 12** (contract's starting proposal, confirmed). With the climate band's
  demonstrated-configuration count at N=9 (§2.3), the best-performing 12% selects
  exactly one configuration, so any X at or below ~22% yields the same anchor.
  X=12 is retained for its MACT lineage rather than re-argued: at this N it is
  not doing independent work.
- **K = 5** [PROPOSED]. Below 5 distinct demonstrated configurations in a climate
  band, fall back to the best single demonstrated configuration. Argument: MACT
  practice uses a 5-source floor for small categories (CAA §112(d)(3)(B)); the
  parallel is the point of the parent.
- **The distribution is over demonstrated cooling-system configurations with
  nonzero design water use** [PROPOSED, the memo's one genuinely contestable
  definition]. Waterless configurations (direct expansion, air-cooled chiller,
  dry cooler without assist; LBNL simulates all at effectively zero site WUE)
  are documented in the record and trivially satisfy any ceiling, but they are
  excluded from the anchor's distribution. Argument: including them collapses P
  to zero, which converts a performance floor into a technology mandate
  (evaporative cooling banned per se). MACT floors operate within the emitting
  category; a source with nothing to emit does not set the floor. The
  water-energy tradeoff this exclusion touches is carried in daylight in §6.
- **Anchor value: P = 0.30 L/kWh** [PROPOSED]. The best demonstrated water-using
  configuration in this climate band is the airside economizer with adiabatic
  cooling, the hyperscale-prevalent system (LBNL 2024 p. 45). LBNL's own
  simulation places it near zero but the report flags that result as likely low
  and cites operator-reported WUE of 0.1 to 0.3 L/kWh for similar systems
  (p. 47, verbatim: "Some hyperscale facilities report WUE values for similar
  systems of 0.1-0.3 L/kWh, which is much higher than the values simulated
  here"). P takes the TOP of the operator-reported band, which is the
  conservative choice against a simulation the source itself distrusts, and it
  is a reported, demonstrated number rather than a modeled one.

### S, the stress tightening

- **Classification source precedence** (contract, unchanged): the basin's own
  statutory or compact designation first; a named public screen (WRI Aqueduct
  baseline water stress class) as fallback where no instrument exists.
- **Multiplier table** [PROPOSED]:
  | External classification | S |
  |---|---|
  | No quantity-stress designation | 1.00 |
  | Designated water-quantity stress (statutory or compact instrument) | 0.75 |
  | Ratified basin budget in deficit (the budget is the citation) | per budget, below 0.75 |
- **Basin one classification: S = 1.00.** Michigan's portion of the western Lake
  Erie basin carries a ratified impairment designation (CWA §303(d) listing,
  MDEQ 2016) and a ratified phosphorus reduction target (40% by 2025, Executive
  Directive 2019-14, implementing GLWQA Annex 4). Both are WATER-QUALITY
  instruments: nutrient loading, algal blooms. Neither designates water-QUANTITY
  stress, and the WWAT's zone machinery (§3.2) is site-specific rather than
  basin-wide. Mapping a nutrient impairment onto a quantity multiplier would be
  the body making a basin-condition judgment across domains, which is the act
  the ratified posture forswears. The impairment is therefore documented as
  basin context and S stays 1.00. [OPEN] if the operator wants a ratified
  cross-domain mapping, it is a new dial constant, argued separately.

### C, revision triggers

- **C(a), superseded input: obligation, unchanged.** Every input to this memo is
  listed in §7; a superseded input forces re-derivation as a version event. The
  standing armed instance: **enactment of SB 1046 or any successor creating a
  numeric consumptive-use cap** supersedes §3.4's "no numeric ceiling" finding
  and adds the translated operand.
- **C(b), N = 3** [PROPOSED]: at 3 metered facilities in the climate band, the
  modeled distribution upgrades to the measured one. Argument: 1 facility is an
  anecdote, 2 is a coincidence, 3 is the smallest cohort where a percentile
  means anything; and the flywheel wants the smallest honest N.
- **C(c), cadence = 8 years** [PROPOSED], MACT's technology-review parent,
  because C(a) carries the velocity (it fired within a day on our own 1.8
  framing when LBNL primary text became readable); the cadence is the backstop,
  not the mechanism.

### The map

- **HUC level 8** [PROPOSED]. Basin one is not a single HUC4: it is the
  Michigan-draining portion of two subregions. Level 8 makes the selection
  explicit and auditable.
- **The HUC8 set** (USGS WBD, subregion boundary descriptions):
  - 0409 St. Clair-Detroit, all five units, all Michigan: 04090001 St. Clair,
    04090002 Lake St. Clair, 04090003 Clinton, 04090004 Detroit, 04090005 Huron.
  - 0410 Western Lake Erie, the Michigan-touching units: 04100001 Ottawa-Stony
    (MI/OH), 04100002 Raisin (MI/OH), 04100003 St. Joseph (IN/MI/OH),
    04100006 Tiffin (MI/OH).
- **Multi-state units**: the registry cannot split a HUC8 at a state line, and
  boundaries are never lines we draw. The Michigan-portion scope is enforced
  where it already lives: facility basin assignment is hand-verified and
  unassigned-by-default (registry seam, shipped 2026-07-28), so only Michigan
  facilities are assigned into these units. An Ohio facility in 04100002 is
  simply never assigned by the basin-one table. Documented here so the choice is
  contestable.
- GLWA's service territory rides the same mechanism: it is a UTILITY territory,
  not a hydrologic unit, and its facilities are assigned by the same
  hand-verified table. Its instruments enter §3.3 as ratified numbers to
  translate when applicable.

## 2. The descriptive layer (what efficient facilities demonstrably run)

### 2.1 LBNL 2024, primary, page-cited

Source: Shehabi et al., *2024 United States Data Center Energy Usage Report*,
LBNL-2001637, December 2024. PDF on disk
(`~/aquadrio-data/calibration-water/`), extracted with pypdf; figure graphics
read directly from the extracted images (Figure 4.4 p. 46, Figure 4.5 p. 47).

- Method basis: TMY climate data from 965 weather stations; 50 sampled
  operational scenarios per space type and cooling system (p. 43).
- **Figure 4.4 site-WUE ranges by cooling system** (box ranges read from the
  figure graphic, approximate by nature and labeled so):
  | Configuration (large-scale unless noted) | Site WUE, L/kWh |
  |---|---|
  | Water-cooled chiller, no economizer (small/midsize) | ~2.1 to 4.1 |
  | Waterside economizer (water-cooled chiller) | ~1.8 to 2.8 |
  | IT liquid cooling: waterside economizer | ~1.7 to 2.5 |
  | Airside economizer + adiabatic cooling (water-cooled chiller) | ~0.25 to 1.55 |
  | Dry cooler with adiabatic assist | ~0.1 to 0.65 |
  | Airside economizer + adiabatic cooling (air-cooled chiller) | ~0 (flagged low by the report itself, p. 47) |
  | Direct expansion; air-cooled chiller; dry cooler; IT liquid dry cooler | ~0 |
- **Figure 4.5 printed medians, exact** (2023 aggregates): Small 0.32,
  Midsize/Colo 0.67, Hyperscale 0.32, AI Specialized 0.61 L/kWh.
- Text anchors: hyperscale aggregate median 0.32, rising to 0.40 under the
  report's own 0.2-median sensitivity (p. 47); fleet average site WUE "just over
  0.36 L/kWh through 2023," projected 0.45 to 0.48 by 2028 (p. 48); operators
  report 0.1 to 0.3 L/kWh for airside-economizer-plus-adiabatic systems (p. 47).

### 2.2 Ren dataset, corroboration only, coverage-checked

Source: Gupta, Hossen, Li, Ren, Islam, Water-Sustainability-Dataset (MIT
license; SOURCES.md artifact 9). Detroit is basin one's climate city.
Coverage check (2026-07-29, committed extract regenerated by
`derive_ren_city_stats.py`): weather availability 99.2% of 43,824 hours;
the uncovered 0.8% carry a hard 0.0 fill and are excluded from means.
Detroit annual means, weather-covered hours only, MODELED (physics model):
fixed-approach cooling-tower config 1.211 L/kWh; fixed-coldwater config
0.644 L/kWh.

**Status: corroboration, not anchor.** The dataset's hourly weather carries an
availability flag and city-differentiated series (the prior suspicion of a
data-fill artifact was our own extract's parse offset, resolved 2026-07-29,
MAPPING.md mismatch #6), but its weather lineage is not yet pinned to named
measured station records [OPEN]. Under the contract's rule (named lineage or
drop), Ren does not anchor P. It corroborates: both Detroit configs sit inside
LBNL's simulated ranges for their system classes.

### 2.3 The climate band's demonstrated-configuration count

Distinct demonstrated configurations informing the band: 7 LBNL large-scale
configurations (§2.1 table) plus 2 Ren Detroit configs = **N = 9**, of which
**6 have nonzero design water use**. Against K=5 the small-N clause does not
trigger; against X=12 the anchor selects the single best water-using
configuration either way (§1).

## 3. The normative layer (what the basin's regime has ratified), primary text only

### 3.1 The Great Lakes Compact, as enacted in Michigan law (MCL 324.34201)

- §1.2 definition, verbatim: "Consumptive Use means that portion of the Water
  Withdrawn or withheld from the Basin that is lost or otherwise not returned to
  the Basin due to evaporation, incorporation into Products, or other
  processes." (The Standard's own frame.)
- §4.5: a proposal resulting in a New or Increased Consumptive Use of
  **5,000,000 gallons per day or greater average over any 90-day period** is
  subject to Regional Review.
- §4.10-4.11: each Party manages and regulates new or increased withdrawals; the
  decade-default management floor is withdrawals of **100,000 gallons per day or
  greater average in any 90-day period**.
- §4.2: water conservation and efficiency program obligations (the parent of
  Michigan's §32708a sector measures).

### 3.2 Michigan Part 327 (NREPA, 1994 PA 451)

- MCL 324.32701(aa), verbatim: large quantity withdrawal is "1 or more
  cumulative total withdrawals of over 100,000 gallons of water per day average
  in any consecutive 30-day period that supply a common distribution system."
- MCL 324.32701(a): adverse resource impact (ARI) standards, keyed to stream
  thermal class and fish population response (for example, cold streams at 3%+
  reduction in thriving fish density; warm streams at 5%+ characteristic fish
  abundance; 25%+ index flow reduction).
- MCL 324.32701(k): consumptive use, the evaporation/incorporation frame,
  matching Compact §1.2.
- MCL 324.32723(1): permit required for (a) new withdrawal capacity over
  **2,000,000 GPD** to a common distribution system; (b) increased capacity
  beyond baseline over 2,000,000 GPD; (c) new or increased LQW over
  **1,000,000 GPD** determined a zone C withdrawal; (d) intrabasin transfer over
  **100,000 GPD average over any 90-day period**.
- MCL 324.32723(6), the permit standard: all water withdrawn, less any
  consumptive use, returned to the source watershed; **no individual or
  cumulative adverse resource impacts**; compliance with legally binding
  regional agreements including the Boundary Waters Treaty of 1909.
- The WWAT (MCL 324.32706a et seq.) is the screening instrument that assigns
  withdrawal zones; zone C triggers site-specific review. It is a per-withdrawal
  instrument, not a basin-wide classification.

### 3.3 GLWA, ratified numbers (the operator's non-negotiable inclusion)

From GLWA's own published system documents (glwater.org, facilities pages and
CIP Appendix D, FY 2025-2029): five water treatment plants with rated
capacities Lake Huron 400 MGD, Springwells 540 MGD, Northeast 300 MGD,
Southwest 240 MGD, Water Works Park 240 MGD; **system total 1,720 MGD rated**.
Raw-water intake locations remain [UNVERIFIED] pending a primary intake
document; not load-bearing for v1 since no volume is translated (§3.4).
Wastewater side: the Water Resource Recovery Facility serves Detroit and 76
other communities across more than 946 square miles (GLWA facilities page).

### 3.4 The finding: no ratified numeric consumptive-use ceiling exists today

Every ratified number in §3.1-3.3 is a THRESHOLD (where registration, screening,
permitting, or regional review begins) or a CAPACITY (what infrastructure is
rated to deliver). None is a ceiling: a facility may lawfully exceed every
threshold by obtaining the permit whose standard is site-specific ARI review,
and no instrument caps a facility's consumptive volume at a number. Translating
a scrutiny trigger into a certification ceiling would misrepresent the
instrument, which is exactly what the translation posture exists to prevent.
Therefore: **translated_ceiling = ABSENT for basin one v1**, no allocation
basis is required (nothing is converted), and the prescription proceeds on the
dial leg alone, which the registry schema accepts by design (one operand
suffices; both are recorded when both exist).

### 3.5 SB 1046, the watched instrument (candidate, not translatable)

Michigan SB 1046 (introduced 2026-06-18, Bayer; referred to Energy and
Environment; amends 32705, 32708a, 32713, 32723 and adds 32723a). Verbatim from
the introduced text (read from the legislature's PDF): beginning July 1, 2027, a
person using or expected to use **550,000 gallons of water or more per day for
consumptive use, based on a yearly average**, shall "[n]ot withdraw from the
waters of the state" (community-supply partnership required) and shall apply for
a permit; a permit holder "shall not use more than 2,000,000 gallons of water
per day for consumptive use." Scope: **all persons; the bill does not define or
name data centers.** Status: a bill is not a ratified limit; it translates to
nothing today. On enactment, C(a) fires: the 2,000,000 GPD consumptive cap is a
true numeric ceiling and becomes basin one's (and Michigan's) translated
operand through a ratified allocation basis at that version event.

### 3.6 Stress instruments (documented; not mapped to S, per §1)

- MDEQ (now EGLE) declared Michigan's western Lake Erie basin waters impaired
  under CWA §303(d) (2016), driver: phosphorus and algal blooms.
- Executive Directive 2019-14: Michigan's Domestic Action Plan under GLWQA
  Annex 4; 40% total phosphorus reduction to the western basin by 2025 (interim
  20% by 2020).
- Both are quality instruments; neither designates quantity stress; S stays 1.00
  and the mapping question is [OPEN] for the operator (§1).

## 4. The first derivation, worked

The contract's fill-in-the-blanks sentence, every blank cited, adapted to the
two-operand record (2026-07-28 ruling: two ceilings, two verdicts):

> Aquadrio prescribes a direct water ceiling of **0.30 L/kWh (strict-under)**
> [PROPOSED] for **the Lake Erie drainage, Michigan portion, plus the GLWA
> service territory (HUC8 set: 04090001-04090005, 04100001, 04100002, 04100003,
> 04100006; USGS WBD)**, effective **on ratification**, version **1**, because:
> (1) efficient facilities in this climate demonstrably run **0.1 to 0.3 L/kWh**
> (operator-reported WUE for airside-economizer-plus-adiabatic systems, LBNL
> 2024 p. 47; the hyperscale-prevalent configuration, p. 45; Ren Detroit configs
> corroborate at 0.64 and 1.21 L/kWh, MODELED, §2.2); (2) the basin's
> quantity-stress posture is **unclassified** (the ratified Erie instruments are
> water-quality designations, §3.6), so S = 1.00; (3) the derivation rule maps
> (1) and (2) to 0.30 x 1.00 = **0.30**.
>
> The basin's governing regime ratifies **no numeric consumptive-use ceiling**
> (§3.4): its instruments gate by thresholds (100,000 GPD registration;
> 1,000,000 GPD zone C permit; 2,000,000 GPD capacity permit; 5,000,000 GPD
> Compact regional review) and a site-specific adverse-resource-impact standard
> (MCL 324.32723(6)). The record states this in place of a translated operand.

The two-verdict rendering for basin one therefore reads:

> The basin's governing regime sets no numeric consumptive ceiling; it gates by
> permit and adverse-resource-impact review at cited thresholds (MCL 324.32723;
> Compact §4.5). Facility permit standing: [reported per facility].
> Aquadrio certifies at 0.30 L/kWh (strict-under), per the basin one
> prescription v1 [when ratified]. Facility meets this bar: [per record].
> The mark requires the facility's lawful standing AND Aquadrio's bar.

Boundary rule: strict_under, carried from Standard v0.3.

Registry entry this ratifies into (shape only, values PROPOSED):
`Prescription(basin=<each HUC8 above, level 8>, boundary="strict_under",
effective_date=<ratification date>, version="1", ratified_by="operator",
derivation_ref="calibration/water/PRESCRIPTION_METHOD_MEMO_2026-07-29.md",
translated_ceiling_l_per_kwh=None, dial_ceiling_l_per_kwh=0.30,
allocation_basis_id=None, dial_ref="this memo §1, §4")`.

## 5. The dial leg against the five factors

1. **Testability.** Every input is public and every step recomputable: LBNL
   PDF page cites, the committed `derive_ren_city_stats.py`, this memo's
   arithmetic (one multiplication). A challenger can rerun all of it.
2. **Peer review and publication.** LBNL-2001637 is a reviewed federal lab
   report; the Ren dataset is published under an archived methodology paper.
3. **Known or potential error rate.** The honest weak factor, stated rather
   than dressed: the anchor is an operator-REPORTED band atop a simulation the
   source itself flags low, and no facility-specific error rate exists for a
   published benchmark. This is the standing corollary recorded 2026-07-28:
   only a meter on a real building creates the error rate, which is why C(b)
   upgrades the distribution at N=3 metered facilities and why the pilot
   outranks further modeling.
4. **Standards controlling the technique's operation.** The parameterization is
   MACT-parented throughout: X=12 (§112(d)(3) best-performing 12%), K=5 (the
   small-category floor), 8-year review cadence, stress tightening shaped like
   residual-risk review.
5. **General acceptance.** The MACT floor method has three decades of
   litigation; percentile-of-demonstrated-performance is the accepted shape of
   technology-forcing floors in US environmental law.

## 6. The water-energy tradeoff, named (adaptation 2 obligation)

Site WUE can be bought with energy: dry cooling approaches zero site water by
burning more electricity, raising carbon and indirect (source) water, and no
per-kWh site-water gate can see it because the penalty lives in the kWh
quantity (LBNL 2024 p. 45 states the same tradeoff). This memo's exclusion of
waterless configurations from the P anchor (§1) is a facet of the same
tradeoff: a waterless facility passes the water gate trivially AND pays its
penalty where the carbon gate and the indirect-water record already look. The
register carries both water axes plus energy on every entry, which is where the
tradeoff stays visible. One dial; the floor stays MACT-simple on site WUE.

## 7. Inputs (the C(a) watch list)

| Input | Vintage | Supersession consequence |
|---|---|---|
| LBNL-2001637 (2024 report) | Dec 2024; the 2025 Update carries no water figures | New water figures re-derive P |
| Ren Water-Sustainability-Dataset | v 05.14.24, 2019-2023 | Lineage pinned or new vintage re-runs corroboration |
| MCL 324.32701, .32723 (Part 327) | Current through fetch of 2026-07-29 | Amendment re-derives §3 |
| MCL 324.34201 (Compact) | Current through fetch of 2026-07-29 | Amendment re-derives §3 |
| SB 1046 | Introduced 2026-06-18, in committee | **Enactment adds the translated operand (armed)** |
| GLWA CIP Appendix D / facilities pages | FY 2025-2029 cycle | New capacities update §3.3 |
| CWA §303(d) MI listing; ED 2019-14 | 2016; 2019 | A quantity-stress designation re-opens S |
| USGS WBD HUC8 delineation | Fetched 2026-07-29 | Boundary revision re-maps the set |

## 8. [OPEN], honestly listed

1. Ratify or amend: P = 0.30; K = 5; N = 3; cadence = 8y; HUC level 8 and the
   nine-unit set; the S table values; S = 1.00 for basin one.
2. The waterless-exclusion rule (§1) is the memo's most contestable definition;
   it needs an explicit operator yes.
3. Cross-domain stress mapping (quality impairment to quantity multiplier):
   refused here; operator may commission it as a separate ratified constant.
4. Ren weather lineage: pin to named station records or Ren stays
   corroboration-only.
5. GLWA intake locations [UNVERIFIED]; becomes load-bearing only when a volume
   is translated.
6. WRI Aqueduct class for the Detroit/Erie region: document as fallback
   cross-check (not fetched this pass; precedence resolved without it).
7. Whether the basin-one artifact also lists GLWA territory as a named overlay
   in the registry `notes` field (cosmetic, WP-3).
8. Counsel confirmations outstanding (the 2026-07-28 packet): MRE 703
   business-record posture; judicial-notice expectations for translated
   instruments; nothing here reaches a public surface before that.

## 9. Sources

- LBNL-2001637: eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report_1.pdf (PDF on disk; pages as printed)
- MCL 324.32701: legislature.mi.gov/Laws/MCL?objectName=mcl-324-32701
- MCL 324.32723: legislature.mi.gov/Laws/MCL?objectName=mcl-324-32723
- MCL 324.34201 (Compact): legislature.mi.gov/Laws/MCL?objectName=mcl-324-34201
- SB 1046 status page: legislature.mi.gov/Bills/Bill?ObjectName=2026-SB-1046; introduced text PDF: legislature.mi.gov/documents/2025-2026/billintroduced/Senate/pdf/2026-SIB-1046.pdf
- GLWA: glwater.org/our-system/facilities/; CIP Appendix D FY 2025-2029 (glwater.org/wp-content/uploads/2023/10/Appendix-D_System-Background-Information-1.pdf)
- USGS WBD boundary descriptions: water.usgs.gov/GIS/huc_name.html; HUC8 04090004 lookup: water.usgs.gov/lookup/getwatershed?04090004
- MDEQ WLEB impairment: michigan.gov/deq (2016 announcement); ED 2019-14: michigan.gov/whitmer (2019-06-20)
- Ren dataset: github.com/Ren-Research/Water-Sustainability-Dataset (SOURCES.md artifact 9)
- MultiState data-center legislation tracker (secondary, for SB 1046 context): multistate.us/insider/2026/7/15