Public Bid Data · Dataset 05

Water and Wastewater Equipment Bid Benchmark

We tested how well the reading step of Mavlon's AI Quoting Engine reads water and wastewater equipment bid packages, using ten public packages from the United States, Canada and Spain. It matched 294 of 298 checks against answer keys written before the runs and found 99 of the 104 contradictions in the keys.

Version 1.0 · updated 18 November 2026 · compiled by Mavlon, Inc. · CC BY 4.0
294 / 298checks matched against answer keys written before the runs
10 of 10public packages run, one run each, no reruns
99 / 104contradictions in the keys found
175 of 228contradictions it flagged that held up when we checked the pages

Water and wastewater equipment makers spend their days inside other people's spec books. A city hires a consulting engineer, the engineer writes a specification around one maker's product "or equal", and every other maker has to read hundreds of pages to find out what it takes to qualify, what the guarantee is, and what the bid form really asks for. So we built a test that anyone can check. We took ten public bid packages for water and wastewater equipment from the United States, Canada and Spain. We wrote down what each one says before any run, and then we let the reading step of Mavlon's AI Quoting Engine read each package once.

This page shows every score, every miss and every answer key, so you can judge the result for yourself.

Can AI read a water equipment bid package? What this test measures

The test checks the reading that an applications engineer does before anyone prices a bid. For each package, the answer key holds 22 to 35 checks, which comes to 298 in all. Every check is something that changes the price, and every one is tied to a short quote and the page it came from. For example, we check these items:

  • We check the equipment, the quantity, the basis of design and the rules for an "or equal" offer.
  • We check the design criteria, the performance guarantee and how the performance test is run.
  • We check how offers are evaluated, including power cost over the life of the equipment.
  • We check the materials, motors, enclosures and hazardous area ratings.
  • We check the controls: the PLC, the screens, the signals to the plant SCADA system and who writes the programs.
  • We check field service days, spare parts, warranty, delivery, damages and the bid form items.

The test also checks the contradictions, which are the places where one document says something different from another. Every key lists the contradictions we confirmed before the runs, coming to 104 in total across the ten packages.

How the test was run

We wrote every answer key first. Mavlon prepared the keys with AI help before any run, quoting the exact words and page for each check, and a script then compared every quote with the page it cites.

Then the system read each package once, with no reruns and no hints. The reading step worked inside a closed folder that held only that package, with no internet and no access to the keys. It had to find the equipment on its own, read every document, and write down what it found with the file and page for each value.

A separate reader then scored each run against its key. The scoring is strict: an answer that was only partly right counts as a miss. That reader also opened the pages behind every contradiction the run flagged and marked each one as holding up, not holding up, or unclear.

The results

Across the ten packages, the reading step matched 294 of 298 checks. It found 99 of the 104 contradictions in the keys. The packages ran to 3,692 pages in total, and each one took between 9.0 and 28.2 minutes to read.

PackageLanguagePagesChecks rightKey contradictions foundFlagged contradictions that held upMinutes
Henniker, New Hampshire, United StatesEnglish1822 / 226 / 613 / 219.0
Ann Arbor, Michigan, United StatesEnglish23232 / 328 / 822 / 2617.7
Valdosta, Georgia, United StatesEnglish51632 / 3211 / 1126 / 3021.4
Camas, Washington, United StatesEnglish18230 / 3010 / 1017 / 2215.8
Tampa, Florida, United StatesEnglish24130 / 3010 / 1218 / 2418.8
Essex Junction, Vermont, United StatesEnglish7530 / 305 / 58 / 1415.2
Elyria, Ohio, United StatesEnglish27230 / 3013 / 1319 / 2416.0
Mercer Island, Washington, United StatesEnglish1,72931 / 358 / 1112 / 1928.2
Winnipeg, Manitoba, CanadaEnglish35932 / 3220 / 2026 / 3120.3
Valladolid, SpainSpanish6825 / 258 / 814 / 1710.3

Everything it did not get fully right is listed here, so nothing hides behind the totals:

  • Tampa: It did not flag one contradiction in the key: the escalator discharge chute had a conflict between the original spec and the drawing, which was resolved by Addendum 1.
  • Tampa: It did not flag one contradiction in the key: the plug valve actuators were electric in W-32.09 and pneumatic on the drawings, which was resolved by Addendum 1.
  • Mercer Island: On Check 4, the run counted the intrinsically safe and level-control panels correctly, but it never said which sites get them or that Flush Station 12 has none.
  • Mercer Island: On Check 11, the run read only the last page of the I/O list, so it gave the 1,852 row count, but it did not give the split into digital inputs, digital outputs and analog inputs, or the new, existing, future and spare counts.
  • Mercer Island: On Check 21, the run noted that Addendum 2 talks about reusing existing conduit and wire, but it never stated the rule that new conduit and wire go only where the circuit and raceway schedules show them.
  • Mercer Island: On Check 32, the run gave the right bid deadline of 2:00 PM, November 3, 2022, but it did not say that bids are emailed as a PDF to the City's public works address, or that a further extension was refused.
  • Mercer Island: It did not flag one contradiction in the key about new conduit and cable versus the reuse of existing wiring.
  • Mercer Island: It did not flag one contradiction in the key about whether the Pump Station 20 PLC enclosure is new or existing.
  • Mercer Island: It did not flag one contradiction in the key about the digital output card channel count on the PLC rack layouts.

The contradictions it flagged, and how many held up

The reading step flagged 228 contradictions of its own. When we checked the pages, 175 held up, 32 did not and 21 were unclear. The ones that did not hold were mostly judgement calls, such as treating two rules written for different parts of a job as if they clashed. The unclear ones rest on its own math or on a reading of a drawing that the documents alone cannot settle.

We publish this number because a list of possible problems is only useful if most of them are real. In a draft quote, every flagged contradiction becomes a question for the engineer, and an applications engineer can drop the ones that do not matter in seconds.

Where the reading found more than our keys

Of the contradictions that held up, 96 were not in our keys at all. Here are three of them:

  • Valdosta: The spec says the sludge is anaerobically digested, but drawing M-3 labels the feed lines as coming from an aerobic digester, and the cake and polymer guarantees depend on which it is.
  • Mercer Island: The flow switch spec asks for a 3/4-inch threaded connection, while the instrument index and the installation detail use 1/2-inch, at 15 stations.
  • Winnipeg: The specification says Pump 2 has a 300 mm suction and 250 to 300 mm discharge, but the Addendum 2 layout drawing shows 350 mm pipe at both Pump 2 connections, which changes the elbow casting and maybe the pump frame.

What makes a water equipment package hard to read

The hardest packages were not always the longest ones. These were the things that made reading hard:

  • Numbers broken by the text layer. In Ann Arbor, superscript footnotes merge into the numbers, so a design dose of 30 reads as "301" unless you look at the page.
  • Drawings and lists with no text. The Valdosta drawings and the Mercer Island I/O list have no text a computer can copy, so every value had to be read from the image.
  • Forms that disagree with the tender. In Winnipeg, the bid forms carry another project's title, a different warranty and section references that are all off by one.
  • Addenda that undo each other. In Ann Arbor, the second addendum reissued the bid form and brought back wording the first addendum had removed.
  • Scope that belongs to someone else. In Mercer Island, the engineer writes all the PLC programs, so an integrator who prices programming would overbid.
  • Language and size. The Valladolid package is in Spanish, and the Mercer Island package runs to 1,729 pages.

What this test does not measure

This test measures reading, not design or pricing. The reading step does not size equipment, select pumps or design a process. It records what the documents ask for, with the page each requirement came from. Pricing depends on each maker's own rates and past jobs, so we test it separately, on a maker's own past bids, before any live bid. Our pilot protocol explains how that test works, and the water and wastewater equipment page shows how the reading turns into a priced draft quote.

How to test any tool on your own spec books

You can run the same test on any AI tool with bid packages you have already answered.

  1. Pick five to ten past spec books where you know what the engineer asked for.
  2. Before the tool sees them, write down the price drivers and the conflicts you found, each with its page.
  3. Give the tool each package once, with no hints, and keep the result as it comes out.
  4. Score it strictly, so a half-right answer counts as wrong.
  5. Open the page behind every conflict it flags, and count how many are real.

The specification compliance matrix is a ready template for writing down the requirements and your answers.

Package by package

A page from the Henniker, New Hampshire, United States water and wastewater bid package

Henniker, New Hampshire, United States

English · 1 file · 18 pages · one screw press dewatering system bought directly by the town

Source: This comes from the Town of Henniker, New Hampshire, with engineer Underwood Engineers, for the Request for Bids, Screw Press Dewatering System for the Henniker WWTF Upgrade (equipment prepurchase), in 2023.

What made it hard: It is only 18 pages, but the specification contradicts itself on panel voltage, the polymer pump drive, the stainless grade and the nameplate colours, and it asks for equality to a basis of design it never names.

Result: The run got 22 of 22 checks right, found 6 of 6 key contradictions, and 13 of the 21 contradictions it flagged held up. It took 9.0 minutes.

One find beyond the key: The interlock and alarm lists need a polymer activation tank level signal, but the polymer system specified is an in-line mixing chamber with no tank or level switch.

A page from the Ann Arbor, Michigan, United States water and wastewater bid package

Ann Arbor, Michigan, United States

English · 3 files · 232 pages · open-channel UV disinfection with gates and controls, chosen by supplier preselection

Source: This comes from the City of Ann Arbor, Michigan (Wastewater Treatment Services Unit), with engineer Black & Veatch, for RFP No. 23-13 Wastewater Treatment Plant UV Disinfection Equipment (UV system supplier preselection), in 2023.

What made it hard: Superscript footnotes merge into numbers in the text layer, so the design dose reads as 301 instead of 30. Two addenda then undo each other on the bid form, and the channel sizes exist only in scanned 1999 drawings.

Result: The run got 32 of 32 checks right, found 8 of 8 key contradictions, and 22 of the 26 contradictions it flagged held up. It took 17.7 minutes.

One find beyond the key: The specification lists an integrated lifting device as a required part of the system, but Addendum 1 accepts systems that rely on a crane the Contractor provides, which changes both scope and how bids compare.

A page from the Valdosta, Georgia, United States water and wastewater bid package

Valdosta, Georgia, United States

English · 4 files · 516 pages · one 2.2-metre belt filter press and a polymer skid inside a construction contract

Source: This comes from the City of Valdosta, Georgia, with engineer CDM Smith, for the Withlacoochee Water Pollution Control Plant Belt Filter Press Replacement (construction invitation to bid), in 2025.

What made it hard: The drawings have no text a computer can copy, the bid form demands two alternates that exist nowhere, and the press on the drawings is wider than the door it must pass through.

Result: The run got 32 of 32 checks right, found 11 of 11 key contradictions, and 26 of the 30 contradictions it flagged held up. It took 21.4 minutes.

One find beyond the key: The spec says the sludge is anaerobically digested, but drawing M-3 labels the feed lines as coming from an aerobic digester, and the cake and polymer guarantees depend on which it is.

A page from the Camas, Washington, United States water and wastewater bid package

Camas, Washington, United States

English · 3 files · 182 pages · two high speed turbo blowers and a master control panel, bought directly

Source: This comes from the City of Camas, Washington (Public Works Department), with engineer HDR, for the Camas Blower Improvements Project, Equipment Purchase, City Project No. S1040 (Camas Wastewater Treatment Plant), in 2024.

What made it hard: The award runs on price plus 20 years of power cost, so the guaranteed kilowatts matter as much as the price, while the start-up section plans for four blowers and the bid form buys two.

Result: The run got 30 of 30 checks right, found 10 of 10 key contradictions, and 17 of the 22 contradictions it flagged held up. It took 15.8 minutes.

One find beyond the key: The bid form asks for the kW of one blower unit at each design point, but Addendum 2 says the top flow of 2,665 SCFM is the total with all machines running, so it is unclear whether bidders enter per-unit or combined kW in the 20-year evaluation.

A page from the Tampa, Florida, United States water and wastewater bid package

Tampa, Florida, United States

English · 4 files · 241 pages · two grit washer systems inside a construction contract

Source: This comes from the City of Tampa, Florida (Wastewater Department, Contract Administration Department), with engineer Hazen and Sawyer, for Contract 17-C-00031, Howard F. Curren AWTP Screen and Grit Building 2 Grit Washer Replacements, in 2017.

What made it hard: The performance guarantee names three different particle sizes, the schedule gives two different unit counts, and an addendum changes the actuators, the signals and the spare parts boxes.

Result: The run got 30 of 30 checks right, found 10 of 12 key contradictions, and 18 of the 24 contradictions it flagged held up. It took 18.8 minutes.

One find beyond the key: The spec gives each grit washer three valves (backwash, blowdown, escalator water), but the drawings show five solenoids per washer and no blowdown valve, which changes the panel I/O, field switches and control logic.

A page from the Essex Junction, Vermont, United States water and wastewater bid package

Essex Junction, Vermont, United States

English · 1 file · 75 pages · one fine screen and washer compactor, equipment only

Source: This comes from the Village of Essex Junction, Vermont, with engineer Aldrich + Elliott, PC, for the Water Resource Recovery Facility New Headworks Screen, Equipment Only (A + E project 17003), in 2017.

What made it hard: This is a short equipment-only package, yet it still rates the same control panel two ways, points to a design section that does not exist and gives two different warranty clocks.

Result: The run got 30 of 30 checks right, found 5 of 5 key contradictions, and 8 of the 14 contradictions it flagged held up. It took 15.2 minutes.

One find beyond the key: The drawing shows two process-water connections to the screen (1-1/2 in and 1 in), but the specification asks for no screen wash and only one 1/2 in supply on the washer compactor.

A page from the Elyria, Ohio, United States water and wastewater bid package

Elyria, Ohio, United States

English · 1 file · 272 pages · one factory-built packaged sewage lift station bought by the city

Source: This comes from the City of Elyria, Ohio, Department of Public Service, Engineering Division, with engineer AECOM, for the Overbrook Pump Station II Procurement: supply of a factory-built packaged wastewater lift station with back-up power, in 2024.

What made it hard: The two specification sections disagree on pump head, pump model, seal faces and test length, the drawing shows a different generator from the specification, and the spare parts clause points to a section that does not exist.

Result: The run got 30 of 30 checks right, found 13 of 13 key contradictions, and 19 of the 24 contradictions it flagged held up. It took 16.0 minutes.

One find beyond the key: The check valve paragraph names a GA 340D lever-and-weight swing check and then asks for a spring-loaded type with an external lever arm, which are two different valves.

A page from the Mercer Island, Washington, United States water and wastewater bid package

Mercer Island, Washington, United States

English · 6 files · 1,729 pages · control panels and instruments at 17 sewer pump stations, quoted by a control system integrator

Source: This comes from the City of Mercer Island, Washington (Public Works), with engineer Brown and Caldwell, for the Sewer SCADA Systems Replacement, Bid No. 22-31 (BC project 153585), in 2022.

What made it hard: At 1,729 pages it is the largest package. The I/O list has a broken text layer and has to be read as images, and the engineer writes all the programs, so an integrator who prices programming would overbid.

Result: The run got 31 of 35 checks right, found 8 of 11 key contradictions, and 12 of the 19 contradictions it flagged held up. It took 28.2 minutes.

One find beyond the key: The flow switch spec asks for a 3/4-inch threaded connection, while the instrument index and the installation detail use 1/2-inch, at 15 stations.

A page from the Winnipeg, Manitoba, Canada water and wastewater bid package

Winnipeg, Manitoba, Canada

English · 18 files · 359 pages · two large dry-pit wastewater pumps bought directly from the maker

Source: This comes from the City of Winnipeg, Water and Waste Department, with Contract Administrator MPE, a division of Englobe, for Tender 879-2023: supply, delivery and on-site inspections of Perimeter Road Pumping Station (PRPS) pumping equipment (project S-1251 pump replacement and electrical upgrade), in 2024.

What made it hard: Four addenda change the motor speed, the damages cap and the cabling scope, while the bid forms carry another project title, a different warranty and section references that are all off by one.

Result: The run got 32 of 32 checks right, found 20 of 20 key contradictions, and 26 of the 31 contradictions it flagged held up. It took 20.3 minutes.

One find beyond the key: The specification says Pump 2 has a 300 mm suction and 250 to 300 mm discharge, but the Addendum 2 layout drawing shows 350 mm pipe at both Pump 2 connections, which changes the elbow casting and maybe the pump frame.

A page from the Valladolid, Spain water and wastewater bid package

Valladolid, Spain

Spanish · 3 files · 68 pages · seven submersible raw-sewage pumps with sensors and control units

Source: This comes from the Ayuntamiento de Valladolid, Área de Medio Ambiente y Sostenibilidad (Ciclo Urbano del Agua), for Expediente CA-11/2016: manufacture and supply of submersible raw-sewage pumps for the main pumping station of the EDAR (wastewater treatment plant), in 2016.

What made it hard: The package is in Spanish, the price formula is an image, and the administrative terms and the technical specification disagree on delivery, installation, penalties and scoring.

Result: The run got 25 of 25 checks right, found 8 of 8 key contradictions, and 14 of the 17 contradictions it flagged held up. It took 10.3 minutes.

One find beyond the key: The ESPD (DEUC) form bound into the administrative terms names a different contract, a waste-container supply V.33/2016, so a bidder copying it would file wrong contract data in envelope 1.

Downloads

The scores and all ten answer keys are free to download. Each key lists every check with the quote and page it came from, and the contradictions we confirmed before the runs.

To cite this page:

Mavlon, Inc. (2026). Water and Wastewater Equipment Bid Benchmark, version 1.0. https://mavlon.co/data/water-wastewater-equipment-bid-benchmark. CC BY 4.0.

Changelog

Version 1.0, 18 November 2026. All ten packages run once and scored: Henniker, Ann Arbor, Valdosta, Camas, Tampa, Essex Junction, Elyria, Mercer Island, Winnipeg and Valladolid.

The step after reading

Reading the package is only the first step of a quote. The AI Quoting Engine then prices every line from your own rates, carries both readings where the documents disagree, and writes the questions for your customer, so your estimator checks a draft instead of starting from a blank sheet.

See how it works for water and wastewater equipment makers

Book a demo
Questions
Can AI read a water and wastewater equipment bid package?
Yes. In this test, the reading step of Mavlon's AI Quoting Engine matched 294 of 298 checks on ten public water and wastewater packages, including design criteria, performance guarantees, named makers and or-equal rules, materials, controls, field service, warranty and bid form items, and it found 99 of the 104 contradictions in our answer keys. Every key and score is published on this page.
Can AI read the drawings and scanned pages in a spec book?
Yes, in this test. The Valdosta drawings and the Mercer Island I/O list have no text a computer can copy, the Ann Arbor package includes scanned 1999 record drawings, and superscripts break the Ann Arbor design table in the text layer. The reading step read those values from the page images.
Does it size equipment or design the process?
No. The test measures reading only. The reading step does not size equipment, select pumps or design a treatment process. It records what the engineer asked for, with the page each requirement came from, so the maker's own engineers can decide how to meet it.
How long does it take to read a water equipment package?
Between 9.0 and 28.2 minutes per package in this test. The time depends on the number of files and how much of the package has to be read from images, more than on the page count.
How often are the contradictions it flags real?
The reading step flagged 228 contradictions of its own. When we checked the pages, 175 held up, 32 did not and 21 were unclear. In a draft quote each flagged contradiction becomes a question for the engineer, who can drop the ones that do not matter.
Does it work for control system integrators?
Yes. One package in the test, Mercer Island, is a control system upgrade at 17 sewer pump stations, read from the integrator's side: panels, I/O, instruments, testing, and which work belongs to the engineer.
Does this benchmark measure pricing?
No. It measures the reading that comes before pricing. Pricing depends on each maker's own rates and past jobs, so Mavlon tests it on the maker's own past bids before any live bid.
Who prepared the answer keys?
Mavlon prepared the keys with AI help before any run, quoting the exact words and page for every check, and a script compared every quote with the page it cites.
Can I use this data?
Yes. The answer keys and the scores are published under CC BY 4.0. You can use them for any purpose if you cite this page.