Watermark.
All sites
Switch site 35 sites · 10 basins2 states Group by
OH Ohio 34
BOSC BOSC Lima Ottawa River · Lima, OH Building Under construction #1261 URB Urbana Mad River · Great Miami Live Investigating #1263 DEF Defiance Maumee mainstem Queued Investigating #1264 FIN Findlay Blanchard River Queued Investigating #1265 TOL Toledo Lucas Co WRRF Queued Investigating #1266 VWT Van Wert Town Creek · Little Auglaize Queued Investigating #1267 BRY Bryan Prairie Creek · Tiffin River Queued Investigating #1268 OTW Ottawa Blanchard River (lower) Queued Investigating #1269 SPR Springfield Mad River · Great Miami Queued Investigating #1270 XEN Xenia Little Miami Queued Investigating #1271 WPA Dayton · WPAFB Mad River · Great Miami Queued Investigating #1272 HAM Hamilton · Middletown Great Miami (lower) Queued Investigating #1273 TRP Troy · Piqua Great Miami (upper) Queued Investigating #1274 SID Sidney Great Miami · headwaters Queued Investigating #1275 GRV Greenville · Darke Co Stillwater · basin divide Queued Investigating #1276 WIL Wilmington Todd Fork · Little Miami Queued Investigating #1277 WUN West Union · Adams Co Ohio Brush Creek · Ohio River Queued Investigating #1278 NAL New Albany · Licking Scioto ↔ Muskingum divide Tracking Investigating #1279 COL Columbus Scioto · Olentangy Tracking Investigating #1280 CSH Coshocton Tuscarawas + Walhonding Tracking Investigating #1281 PIK Piketon Scioto River · PORTS Tracking Investigating #1282 SAN Sandusky · Perkins Twp Sandusky Bay · Lake Erie Tracking Investigating #1283 NWK Newark Licking River Tracking Investigating #1284 ZAN Zanesville Muskingum mainstem Tracking Investigating #1285 FRE Fremont · Clyde Lower Sandusky Tracking Investigating #1286 TIF Tiffin Sandusky (mid) Tracking Investigating #1287 BUC Bucyrus Sandusky headwaters Tracking Investigating #1288 CLE Cleveland Lower Cuyahoga Tracking Investigating #1289 AKR Akron Upper Cuyahoga · CVNP Tracking Investigating #1290 LRD Lordstown · Warren Upper Mahoning Tracking Investigating #1291 YNG Youngstown Mahoning mainstem Tracking Investigating #1292 LAN Lancaster Upper Hocking Tracking Investigating #1293 ATH Athens Lower Hocking Tracking Investigating #1294 LOG Logan Hocking Hills Tracking Investigating #1295
MAU Maumee Basin 8 sites
2MI The Two Miamis 9 sites
SE Southeastern Basins 10 sites
NE Northeast Basins 8 sites
Each site is a section of the Watermark platform — pivoted by basinstate. Live & building sites open; queued and tracking sites route to their coming-soon / watch page.
Switch site 35 sites · 10 basins2 states 35 open Group by

Each site is one Watermark investigation. The page color says what opens; the campus line is the build on the ground — two different clocks.

OH Ohio 34
BOSC BOSC Lima Ottawa River · Lima, OH Draft Under construction #1261 URB Urbana Mad River · Great Miami Open Investigating #1263 DEF Defiance Maumee mainstem Queued Investigating #1264 FIN Findlay Blanchard River Queued Investigating #1265 TOL Toledo Lucas Co WRRF Queued Investigating #1266 VWT Van Wert Town Creek · Little Auglaize Queued Investigating #1267 BRY Bryan Prairie Creek · Tiffin River Queued Investigating #1268 OTW Ottawa Blanchard River (lower) Queued Investigating #1269 SPR Springfield Mad River · Great Miami Queued Investigating #1270 XEN Xenia Little Miami Queued Investigating #1271 WPA Dayton · WPAFB Mad River · Great Miami Queued Investigating #1272 HAM Hamilton · Middletown Great Miami (lower) Queued Investigating #1273 TRP Troy · Piqua Great Miami (upper) Queued Investigating #1274 SID Sidney Great Miami · headwaters Queued Investigating #1275 GRV Greenville · Darke Co Stillwater · basin divide Queued Investigating #1276 WIL Wilmington Todd Fork · Little Miami Queued Investigating #1277 WUN West Union · Adams Co Ohio Brush Creek · Ohio River Queued Investigating #1278 NAL New Albany · Licking Scioto ↔ Muskingum divide Watching Investigating #1279 COL Columbus Scioto · Olentangy Watching Investigating #1280 CSH Coshocton Tuscarawas + Walhonding Watching Investigating #1281 PIK Piketon Scioto River · PORTS Watching Investigating #1282 SAN Sandusky · Perkins Twp Sandusky Bay · Lake Erie Watching Investigating #1283 NWK Newark Licking River Watching Investigating #1284 ZAN Zanesville Muskingum mainstem Watching Investigating #1285 FRE Fremont · Clyde Lower Sandusky Watching Investigating #1286 TIF Tiffin Sandusky (mid) Watching Investigating #1287 BUC Bucyrus Sandusky headwaters Watching Investigating #1288 CLE Cleveland Lower Cuyahoga Watching Investigating #1289 AKR Akron Upper Cuyahoga · CVNP Watching Investigating #1290 LRD Lordstown · Warren Upper Mahoning Watching Investigating #1291 YNG Youngstown Mahoning mainstem Watching Investigating #1292 LAN Lancaster Upper Hocking Watching Investigating #1293 ATH Athens Lower Hocking Watching Investigating #1294 LOG Logan Hocking Hills Watching Investigating #1295
MAU Maumee Basin 8 sites
2MI The Two Miamis 9 sites
SE Southeastern Basins 10 sites
NE Northeast Basins 8 sites
The page — what opens on Watermark Open · Draftfull record Queued · Watchingpreview page Restrictedrequest access
The campus — the build on the ground Investigating Confirmed Under construction Operational
← Back to About
About · Sustainability

The build-out has a footprint too. Here's ours.

Watermark exists to put a compute build-out on the public record. Running the platform is itself compute — servers, storage, and a working set of AI models that read documents so you don't have to. We hold ourselves to the same standard we ask of the sites we track: the numbers below are published as a lead, not a verdict, and every one is labeled with how it was derived.

GreenOps

Built to do less, by default

Most of what cuts our footprint isn't a program — it's an architecture decision made once and inherited by every page after. In order of how much they save:

Read once, cite forever

A record's structured read is computed once, at corroboration time — not re-derived every time a page is opened. Every later visit is a cache hit, not a new inference.

Static-first serving

Pages are mostly static HTML built from the corroborated record. No client bundles re-fetching and re-rendering state on every load.

Right-sized models

Extraction and classification run on small, task-tuned models. Large frontier models are reserved for the handful of steps that need judgment.

Batch, not on-demand

Document ingestion runs on a schedule against off-peak grid windows, in batches — not triggered per visitor request.

No tracker sprawl

No analytics scripts, ad pixels, or session replay. Less fetched and less run, on every single page view.

Cold storage for cold records

Sites still in early tracking phases sit in cheaper, lower-power storage tiers until they're active investigations.

Artificial intelligence

Where models touch the record

Models are scoped narrowly — to reading, not to deciding. We don't run recommendation engines, engagement scoring, or ad targeting. Four jobs, each sized to the smallest model that does it well:

extraction
Structured extraction

Turns a scanned permit or filing into typed fields — dates, parties, figures — the record can display and cite.

small task-tuned model
corroboration
Corroboration assist

Surfaces candidate sources that might confirm or contradict an open lead. Never confirms one on its own.

small task-tuned model
search & ask
Search & Ask

Answers a question against the corpus and returns it with citations back to source pages.

mid-size model, retrieval-grounded
drafting
Drafting

Turns dense source excerpts into a readable summary for a profile or timeline entry, held for editorial review.

frontier model, reviewed before publish
[inference]

Everything a model outputs is stamped inference until a human corroborates it against a source. A draft never ships as a finding on its own.

The usage report

Compute, AI, electricity & water

Trailing 12 months · Jul 2025-Jun 2026 figures are inferred, not metered — see methodology below
Computemodeled
6,366vCPU-hrs
hosting, search, CI/CD
[inference] AWS Cost Explorer + GitHub Actions billing
AI inferences runmodeled
1,395calls
extraction, ask, corroboration, drafting
[inference] Anthropic Admin usage (derived call count)
Electricity drawnmodeled
0.0535MWh
compute + cooling, all regions
[inference] power-draw specs x PUE
Water drawnmodeled
52.2gal
direct cooling + generation upstream
[inference] WUE benchmarks
Compute · by function (vCPU-hrs)
Bar · horizontal Compute by function (vCPU-hrs)
0 2500 5000 Hosting 2928 vCPU-hrs Search 2928 vCPU-hrs CI/CD 510 vCPU-hrs

Hosting & serving the record still costs more than reading it with a model — the site is mostly static pages, not live inference.

AI · by task type
Donut · composition AI inferences by task type
1395 inferences
  • Structured extraction 767
  • Search & Ask 418
  • Corroboration assist 140
  • Drafting summaries 70

Structured extraction — turning a filing into typed fields — is most of the volume; drafting is the smallest and most closely reviewed.

Electricity · monthly draw
Line · monthly (MWh) Electricity drawn by month
0 0.25 0.5 0.75 1 Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun 0.0045 MWh

Source mix
  • Grid (regional mix) · 0.0482 MWh · 90.1%
  • Matched renewable (RECs) · 0.0053 MWh · 9.9%
Water · against internal budget
Gauge · against the cap Water drawn vs budget
35% of the internal annual water budget
0 cap · 150 gal 200

Draw breakdown
  • Direct — on-site cooling · 25.4 gal · 48.7%
  • Indirect — grid generation upstream · 26.8 gal · 51.3%
[how these are derived]
Compute

AWS Cost Explorer instance-hours (by service, folded into platform functions) converted to vCPU-hours by instance size, plus GitHub Actions minutes converted by runner core count. Model inference runs on third-party/serverless infra (Anthropic, Bedrock) and is reported as AI volume, not our vCPU-hours; we run no GPU instances.

Electricity & carbon

vCPU-hours x 7 W per allocated vCPU (an average-operational assumption folding utilization) x PUE 1.2. Derived ~0.0145 MTCO2e (electricity x 596.326 lb/MWh, SRVC) vs AWS's location-based estimate 0.045 MTCO2e — ~32% of it. The two agree in order of magnitude; they differ because the vCPU model captures only active instance compute, whereas AWS's estimate spans every service (storage, transfer, managed-service and idle-capacity overhead).

Water

Modeled from electricity using published Water Usage Effectiveness (WUE) benchmarks — a site WUE for direct on-site cooling and a source WUE for the water withdrawn upstream to generate the power drawn. The two bases are reported separately, never summed across bases.

AI inferences

The Anthropic Admin API exposes token aggregates only, so the call count is total tokens ÷ 4000 avg tokens/call (a stated assumption), and the by-task split is modeled until the per-task workspace keys (#1080) are named so the by-workspace usage can be labeled and metered by task.

sources · AWS Cost Explorer + Sustainability/CCFT (Jul 2025-Jun 2026) · GitHub Actions/storage billing (Jul 2025-Jun 2026) · Anthropic Admin usage + cost (Jul 2025-Jun 2026) · EPA eGRID2023 subregion SRVC factors · WUE benchmarks (EPRI 2024 / Uptime Institute 2023 / NREL 2012)

These are estimates, not meter reads
Have real numbers, or found an error?

Same rule as everywhere else on Watermark: every figure links to a correction.

Submit a correction →

maintained by the record team · data catalog · about