18.3 Failed Samples, Reclean/Retest, Exterior Visual, and Multi-Family Random Sampling

Fail means reclean what the wipe stood for — not a speech, and not a lucky average

Quick Answer: 745.227(e)(8)(vii) compares each single-surface result to the action levels (on or after January 12, 2026: 5 / 40 / 100 µg/ft²). If the residual lead equals or exceeds the applicable action level, the components represented by the failed sample shall be recleaned and retested. A composite is compared to the action level divided by half the number of subsamples. A four-subsample floor composite is judged against 5 ÷ 2 = 2.5 µg/ft²; 3.0 fails even though 3.0 is below 5. Composites need ≥2 subsamples, must include every component being tested, and must not mix component types (745.227(g)). Exterior paint abatement is a visual of the closest outdoor living-area horizontals and of paint chips on the dripline or foundation, which must be removed and disposed of ((e)(8)(v)(C)). Multi-family random sampling ((e)(9)) requires similarly constructed and maintained units, blind selection from the people who abate or clean, and a sample large enough for 95% confidence that no more than 5% or 50 dwellings (whichever is smaller) exceed the action levels.

Section 18.1 was the scoreboard. Section 18.2 was who wipes and how many rooms. This section is what you do with a fail, how composite math turns a "pretty good" average into a fail, how exteriors actually finish, and when a random multi-family plan is legal.

Single-surface fail: reclean and retest the represented components

Equals or exceeds is the verb. A floor at 5 µg/ft² fails. A sill at 40 fails. A trough at 100 fails. "We rounded down" is not in (e)(8)(vii).

The remedy is not automatically "wash the whole building," and it is not "wipe a 12-inch patch under the fail and call HUD." Components represented by the failed sample means the floor system, interior sills, or troughs that wipe stood for — typically those component types in that room or in the set the sampling design used that wipe to represent. HUD's documented methodology says the same thing in contractor English: reclean the surfaces that failed and all other surfaces represented by the surfaces that failed.

After recleaning:

1. Certified supervisor onsite for that reclean (745.227(e)(2)).
2. Inspector or risk assessor visual must pass again.
3. New ≥1 hour wait after that final cleanup.
4. Retest those components — inspector or risk assessor, NLLAP lab.

Do not rewipe a wet floor to save the second trip. Do not occupy the work area on a failed action-level wipe.

Worked example — one failed sill. Bedroom floor 3 µg/ft², bedroom sill 55 µg/ft², trough 40 µg/ft². The sill exceeds 40. Reclean the interior window sills represented, retest those sills. Do not automatically HEPA the attic, and do not skip the sill because the floor passed.

Worked example — failed outside-containment floor. Interior floors are 2–3 µg/ft². The floor outside the containment area is 8 µg/ft². That floor exceeds 5. Reclean and retest the floors represented by that outside sample — the pathway and room beyond the barrier, not a lecture that interior rooms "already cleared." Containment leaked or the buffer was never cleaned. That is why (e)(8)(v)(A) exists.

When does a fail imply the whole unit? When the sampling plan's failed sample represents the whole unit, or when enough component types fail that you never actually cleaned. If every floor wipe is ≥5, you do not argue room-by-room. If one trough fails and floors are below 5, you reclean that trough family, not the roof as a ritual.

Composite rules and the half-subsample math

745.227(g) allows composite dust sampling only in the situations (c) through (e) already permit (hazard screen, risk assessment, post-abatement). If you composite:

1. At least two subsamples.
2. Every component that is being tested shall be included in the sampling.
3. Do not mix component types. Floors with floors. Sills with sills. Troughs with troughs. A composite that dunks a floor wipe and a sill wipe into one tube is illegal under (g)(3) and uninterpretable against 5 versus 40.

(e)(8)(vii) then changes the pass/fail line. Compare the composite's residual lead to:

action level ÷ (½ × number of subsamples)

Fail if the composite equals or exceeds that reduced number.

The half-subsample divisor exists because compositing dilutes a dirty room with cleaner rooms. Without it, one filthy floor hiding in a four-room average could still print below 5. EPA makes the bar stricter as you add subsamples.

Worked example — four-room floor composite at 3.0 µg/ft². Four floor subsamples, one tube, lab reports 3.0 µg/ft². Single-surface thinking says 3.0 < 5, pass. Composite rule: 5 ÷ 2 = 2.5. 3.0 ≥ 2.5. Fail. Reclean the floors represented by that composite — all four rooms' floors, not the one room you guess was dirty — then retest.

Worked example — four-room floor composite at 2.0 µg/ft². 2.0 < 2.5. The composite beats the action-level comparison. If 2.0 is a reportable number, those floors can still be a dust-lead hazard (Section 18.1) while the job completes. Do not reclean solely because 2.0 is reportable. Do not tell the owner there is no hazard.

Worked example — mixed composite. Supervisor asks the assessor to put two floors and two sills in one tube "to save lab fees." 745.227(g)(3) forbids mixing component types. Reject the tube. Resample.

Single-surface sampling avoids this arithmetic. Many inspectors prefer it on small jobs for that reason. Composite is legal when (g) is followed; it is not a discount that loosens 5 / 40 / 100.

Exterior paint abatement is a visual, not a floor-wipe clone

(e)(8)(v)(C) is short and easy to overwrite with interior habits:

• Following an exterior paint abatement, a visible inspection shall be conducted.
• All horizontal surfaces in the outdoor living area closest to the abated surface shall be found to be cleaned of visible dust and debris.
• A visual inspection shall determine the presence of paint chips on the dripline or next to the foundation below any exterior surface abated.
• If paint chips are present, they must be removed from the site and properly disposed of, according to applicable Federal, State, and local requirements.

There is no 5 / 40 / 100 wipe table in (v)(C). Do not invent federal exterior floor wipes because the interior chapter uses wipes. Do not pass an exterior because the siding looks newly painted while the patio table and the dripline are still peppered with chips. Dripline (745.63) is the area within 3 feet surrounding the building perimeter — the band where chips actually land.

Worked example — siding job, chips in the mulch. Visual of the deck nearest the abated wall is clean. The inspector finds chips along the foundation in the dripline. (v)(C) is not done. Remove the chips, dispose of them as Chapter 17 requires, visual again. Do not bury chips under fresh mulch and call it disposal.

If the project mixed interior and exterior, the interior still gets (e)(8) dust sampling. Exterior visual does not waive living-room floors.

Multi-family random sampling — blindness is the point

745.227(e)(9) lets you randomly sample for post-abatement testing in a multi-family dwelling with similarly constructed and maintained residential dwellings, provided:

1. The certified individuals who abate or clean the dwellings do not know which dwelling will be selected for the random sample.
2. A sufficient number of dwellings are selected to provide a 95 percent level of confidence that no more than 5 percent or 50 of the dwellings (whichever is smaller) in the randomly sampled population exceed the appropriate action levels.
3. The randomly selected dwellings are sampled and evaluated under (e)(8) — visual, hour, room counts, action levels — not a lighter "random" wipe of one hallway.

Similarly constructed and maintained is the eligibility gate. Do not random-sample a 1920 walk-up with original sashes together with a 1975 building that already had window replacement, and pretend one statistical population covers both. Units that were not abated, or that used a different method, are not in that population.

Independence from the crew is the integrity gate. If the supervisor posts "clearance units: 2A, 4C, 6F" on the job board Monday and the crew spends Thursday detailing those three, random sampling is a sham. Selection must be unknown to the people who abate or clean. The inspector or risk assessor (or the project designer running the sampling design) holds the draw. Document it.

95% / 5% or 50 is the size gate. HUD documented methodologies supply the sample-size tables that implement that sentence. You do not sample 3 of 80 units because three is a nice number. For large properties, 50 dwellings can be the smaller cap versus 5% (5% of 2,000 is 100; the rule takes 50). For small properties, 5% of 20 is 1 dwelling — that does not magically mean one wipe on one unit satisfies the confidence statement; the documented table, not a napkin 5%, is how you pick n. If the building is so small that random sampling cannot meet (e)(9), sample the units you abated under (e)(8) and do not hide behind "we randomized."

Worked example — crew knows the list. Forty identical garden units, same window replacement, same crew. The supervisor emails the assessor the unit list and CCs the foreman "so they can be ready." The foreman extra-cleans the named units. (e)(9)(i) fails. Those extra-clean units do not represent the population. Either sample every abated unit under (e)(8), or redraw without telling the people who clean, and reclean any unit that was tipped off as if it were not random.

Worked example — failed random unit. Twenty units drawn under a valid (e)(9) plan. Unit 12's floor is 9 µg/ft². Reclean and retest the components represented in Unit 12. A random-plan fail is not automatically a fail of the 19 unsampled units, but it is evidence about the population the 95%/5% statement was supposed to protect. Repeated fails mean the cleaning system failed; you do not keep drawing new random units until you find a clean one and ignore the dirty ones.

Official sources: 40 CFR 745.227(e)(8)(v)(C), (e)(8)(vii), (e)(9), (g) (exterior visual; reclean/retest; composite comparison to action level divided by half the subsample count; composite construction rules; multi-family random sampling); 40 CFR 745.63 (dripline).