BRENNEN

Industries / Construction and property / Damage restoration

Mold remediation

Mold remediation is removing mold growth from a building under controls that keep spores from spreading, then fixing the moisture that let it grow.

7

Milestones achieved, from 1993 to 2013.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1993

    Cleanup sized by area

    On 7 May 1993 the New York City Department of Health, the city's Human Resources Administration and the Mount Sinai Occupational Health Clinic convened an expert panel on the mold Stachybotrys atra (now Stachybotrys chartarum), after mold problems in several city buildings. Its guidelines, later widened to all molds, set the precautions by the size of the moldy area: Level I up to 10 square feet, which trained building staff can clean, up to Level IV, more than 100 contiguous square feet, which calls for containment, respirators and a health and safety professional's oversight. They say bulk or surface sampling is not required before remediation.

    toronto.ca

    Credit

    New York City Department of Health; Mount Sinai Occupational Health Clinic

  2. 2000

    Infant link not proven

    In December 1994 and January 1997 the CDC's weekly report linked a cluster of 10 infants in Cleveland, Ohio, with bleeding in the lungs, one of whom died, to water-damaged homes and the mold Stachybotrys chartarum. The finding was cited in guidelines, testimony and lawsuits, and helped close public buildings. On 10 March 2000 the CDC reported that internal and outside reviews had found "shortcomings in the implementation and reporting of the investigation", and concluded that the link "was not proven".

    LessonOne unconfirmed study drove years of building closures, cleanups and lawsuits before it was checked.

    cdc.gov

    Credit

    US Centers for Disease Control and Prevention (CDC)

  3. 2001

    Dry within 48 hours

    The US Environmental Protection Agency (EPA) published "Mold Remediation in Schools and Commercial Buildings" in March 2001 and reprinted it in September 2008. It tells building managers to dry wet materials within 24 to 48 hours to prevent mold growth, and sorts jobs by the area affected: small (under 10 square feet), medium (10 to 100) and large (over 100).

    epa.gov

    Credit

    EPA Indoor Environments Division

  4. 2001

    A verdict reshapes cover

    On 1 June 2001 a jury in Travis County, Texas, awarded Melinda Ballard and Ron Allison $32 million against Fire Insurance Exchange, part of Farmers, for mishandling a water-leak claim after which mold spread through their house. The award comprised $6.2 million in actual damages, $12 million in punitive damages, $5 million for mental anguish and $8.9 million in lawyers' fees. In December 2002 the Texas appeals court removed the punitive damages. By then, by the National Association of Independent Insurers' count, 35 states had approved an Insurance Services Office (ISO) homeowners endorsement that excludes mold unless it results from a covered water loss, with an optional mold limit of $10,000 a year.

    LessonA slow water claim became a far larger mold loss, and insurers answered by capping mold cover.

    insurancejournal.cominsurancejournal.comleagle.com

    Credit

    Travis County jury; ISO

  5. 2003

    A standard for mold work

    The IICRC published the first edition of S520, its standard for professional mold remediation, in 2003, followed by editions in 2008, 2015 and 2024. It covers mold-damaged structures, their heating and cooling systems and their contents.

    iicrc.org

    Credit

    IICRC S520 Consensus Body

  6. 2003

    Testers cannot be cleaners

    A Texas law in force from 1 September 2003 requires a license for mold assessment and for mold remediation, and bars any license holder from doing both on the same project, or owning both firms on it. New York's Labor Law Article 32 sets the same split, requires an independent assessment to set the scope of the work, and requires a post-remediation assessment.

    LessonSeparating the person who finds the problem from the person paid to fix it removes the incentive to inflate the job.

    texas.public.lawdol.ny.gov

    Credit

    Texas Legislature; New York State Department of Labor

  7. 2013

    An index not validated

    EPA researchers built the Environmental Relative Moldiness Index (ERMI), which scores a home's mold against a national sample of indoor mold values measured with MSQPCR, an EPA-patented DNA method that identifies and counts mold species. EPA licensed MSQPCR to companies (10 active licenses in January 2013), and firms sold ERMI tests to the public. On 22 August 2013 EPA's Office of Inspector General found that EPA had never validated or peer-reviewed either tool for public use, and that some licensees' advertising could make people think they were EPA-approved methods.

    LessonA research tool sold as a home test can drive cleanup decisions it was never checked for.

    epa.gov

    Credit

    EPA Office of Inspector General

Watch for

  1. Mold scopes from photos

    QuoteIQ, which sells business software to contractors, says its AI Estimator produces per-square-foot mold remediation estimates, covering containment, HEPA air filtration, antimicrobial treatment and demolition, from job descriptions or site photos, and that technicians "who used to spend 45 minutes on estimates are closing scopes in under 10". It gives no accuracy figures.

    Achieved whenan outside test compares the tool's areas and containment levels with assessors' measured scopes for the same jobs.

    myquoteiq.com

    Claimed by

    QuoteIQ, June 2026

Open

  1. 01

    Spore results, many readers

    A test set of past post-remediation air-sample reports, with names and addresses removed, each giving the lab's indoor and outdoor spore counts by type. At least ten assessors, working separately, and three AI models each call every report pass or fail and give their reasons. Publish how often they agree on the same numbers.

    The brief
    Credit · unclaimed
  2. 02

    Mold area from photos

    A test of whether AI vision models can place visible mold in the size bands that decide the cleanup method, under 10 square feet, 10 to 100 or over 100 by EPA's guide, from the photos in past assessment reports, scored against the area the assessor measured on site.

    The brief
    Credit · unclaimed

Open 01 · the brief

Spore results, many readers

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A test set of past post-remediation air-sample reports, with names and addresses removed, each giving the lab's indoor and outdoor spore counts by type. At least ten assessors, working separately, and three AI models each call every report pass or fail and give their reasons. Publish how often they agree on the same numbers.

Why it is still open

There is no official number to pass. New York's Department of Labor says "there are currently no EPA or Federal standards for mold, so any interpretation of sampling results would not be based on any established health or environmental standard". Robertson and Brandys (Mycologia, 2011) compared seven labs' counts of samples taken in the same building and found indoor counts ranging from 1,510 to 15,278 spores per cubic meter under one condition, and said spore traps should not be the only method. AIHA's 2019 guidance stresses how variable sampling conditions make results hard to interpret. Studies found measure the labs' counts; none measured how often assessors reach the same verdict from the same report. Techcyte's Sporecyte software has counted spores on slides since 2020, but the pass or fail call is still a person's.

dol.ny.govebi.ac.ukmed.navy.miltechcyte.com

Built with what exists

Reports donated by assessment firms with their clients' permission; a simple online form or Label Studio for the calls; AI models such as Claude, GPT and Gemini given the same report and prompt; Fleiss' kappa, a measure of agreement among many raters, computed in Python with statsmodels.

  1. First step

    Collect 30 redacted post-remediation reports from two firms and have three assessors call them independently.

  2. Done looks like

    300 reports, each called by ten assessors from at least three states and by three models, with Fleiss' kappa for pass or fail overall and by ratio of indoor to outdoor counts, and all calls and reasons released under an open licence.

Open 02 · the brief

Mold area from photos

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A test of whether AI vision models can place visible mold in the size bands that decide the cleanup method, under 10 square feet, 10 to 100 or over 100 by EPA's guide, from the photos in past assessment reports, scored against the area the assessor measured on site.

Why it is still open

A University of Salford team's YOLOv9 model finds mould regions on indoor walls and ceilings (a 76% average classification rate on its own dataset, published January 2025) but does not measure them. QuoteIQ prices mold jobs from photos without published accuracy. No public set pairing mold photos with measured areas was found.

api.crossref.orgmyquoteiq.comepa.gov

Built with what exists

Report photos with a tape or scale card in frame, given with permission; Label Studio for outlining; Meta's Segment Anything (SAM 2) for masks; Ultralytics YOLO; a small open vision model fine-tuned on one graphics card (see the Note "One GPU to fine-tune"); a commercial model for comparison.

  1. First step

    Ask one assessment firm for 100 past report photos with their measured areas, with faces and addresses removed.

  2. Done looks like

    At least 1,000 photos from at least 200 jobs, each with its measured area, released under an open licence, and for each model the share placed in the right band and the share of over-100-square-foot cases called smaller, the mistake that leaves containment out.