The Fall That Happened on the Way Down, Not the Way Up

Before anything else

If a roof access route on one of your buildings currently has no protected transition at the top of the ladder, the action today is to stop rooftop work at that address and send someone to look at it, not to add a line to the toolbox talk. Everything below is a reconstruction, and reconstructions are only useful to shops that act on them before their own version happens.

This article follows one real-shaped case backwards from the paperwork it left behind. The claim it carries is narrow and it is not the one most incident reviews reach: the exposure on a roof service call is concentrated almost entirely in the two transitions, and a fall protection program measured on whether the tech was tied off at height cannot see the transitions at all, because during them the honest answer to that question is yes.

The event as the paperwork described it

A tech went up to service a rooftop unit on an 18-foot flat-roofed single-storey building, worked most of the afternoon, and fell at the top of the fixed ladder on his way down. He had a bag in his hand.

The file afterwards contained: a work order with arrival and departure timestamps, a job hazard analysis with the fall protection line marked required and yes, a harness inspection entry dated that morning, a photograph of the roof anchor near the unit, and a supervisor's note reading "was tied off." Every document in that list was accurate. None of them was false, and none of them was contested.

That is the problem worth writing down. The record was truthful and it described a compliant afternoon, and a man still went off the roof.

What the record could see and what it could not

Read the timestamps rather than the checkboxes and the shape appears.

The work order gave 210 minutes on site. Backing out the ground time at both ends put roughly 148 minutes on the roof. The anchor photograph placed the connection point beside the unit. The ladder was at the far corner, about 40 feet away across the deck.

So the afternoon had four intervals, not one: the climb and the walk in, the connected work, the walk out, and the transition down. Connected time covered the second interval. The first, third and fourth were disconnected by necessity, because there was nothing between the ladder and the anchor to connect to.

Reconstructing from the walking distance and an ordinary pace, the two transit intervals plus the two transitions total about 6 minutes of the 148. That is 4.1 percent of the roof time. It also carried 100 percent of the unprotected exposure, because the other 142 minutes were spent connected to the anchor in the photograph.

The checkbox asked one question - was fall protection used - and the true answer, yes, was recorded. A single Boolean covering an interval that was 4.1 percent one thing and 95.9 percent another will always return the majority answer. It is not that someone lied on the form. It is that the form had no resolution at the scale where the injury lives.

The obvious correction, run through the arithmetic

The instinct after a case like this is that he should have stayed connected through the transition. That instinct is right, and the useful part is what happens when you try to satisfy it with what was on the roof: the only attachment near the ladder was a deck-level anchor.

Total fall distance is measured from the anchorage, and it is a stack of terms, all of them additive:

• free fall, which is the distance travelled before the system begins to arrest;
• deceleration distance, which is the shock absorber's elongation while arresting;
• harness stretch and D-ring shift, published by the harness manufacturer;
• the height of the worker below the D-ring, because clearance is to the lowest part of the body;
• a safety margin, owned by the manufacturer's instructions.

Before running it, one correction that is a re-basing rather than an addition. The required-clearance figure printed on a lanyard's own label is not a raw free fall number. It already contains the deceleration distance, usually the harness stretch, often a margin, and it was computed for an anchorage at or above the user's D-ring. So you do not add those terms to the label figure a second time. What you add is only what the label's basis excluded, and for a deck-level anchor that is the extra free fall created by the anchorage sitting below the D-ring. Read the basis, then re-base.

The stack at a deck-level anchorage. A 6-foot shock-absorbing lanyard with the anchor at the tech's feet and the D-ring about 5 feet above the deck gives a free fall of 6 plus 5, which is 11 feet. That alone breaks 29 CFR 1926.502(d)(16) for construction and its general industry counterpart at 29 CFR 1910.140(d), both of which require the system to be rigged so an employee can free fall no more than 6 feet. Deceleration distance is bounded at 3.5 feet by the same provision, so take 3.5. Harness stretch and D-ring shift come off the harness label, commonly around 1 foot combined. Worker height below the D-ring, about 5 feet. Margin, take 3 feet, rounding the requirement up rather than down because rounding a required clearance down is the error that lands someone.

11 plus 3.5 plus 1 plus 5 plus 3 is 23.5 feet required below the anchorage. The anchorage is on an 18-foot roof deck. He needs 23.5 and the building gives him 18. The deficit is 5.5 feet, which means the system does not arrest him above grade. Connecting to that anchor would have changed how he arrived at the ground, not whether he did.

The same stack at an anchorage 6 feet above the deck. Put the attachment on a post or a rail at the top of the ladder instead, 6 feet up, which is 1 foot above the D-ring. Free fall becomes 6 minus 1, or 5 feet, and that satisfies the 6-foot limit. Deceleration 3.5, harness and D-ring 1, worker below D-ring 5, margin 3. Total 17.5 feet below the anchorage. The anchorage is 24 feet above grade, so 6.5 feet of clearance remains.

Same harness, same lanyard, same building, same man. The only variable that moved was the height of the anchor point, and it moved the outcome from a fatality to an arrest with room to spare. That is the whole reason anchorage location is an engineering decision and not a convenience decision.

One term the stack does not describe at all. The anchor in the photograph was 40 feet away across the deck. A fall taken while connected 40 feet off to one side is a pendulum: the worker swings through an arc, which adds vertical distance the stack above never counted, and ends in an impact with whatever is in the arc - here, a parapet. Swing fall is not a correction to the numbers above. It is a separate injury mechanism that those numbers say nothing about, and the control for it is anchor position, not a longer clearance.

Why the arithmetic was never the first answer anyway

Everything above prices the last line of defence. Arresting a fall is the last resort, and this case shows why: the arrest option at the place where the fall happened did not close numerically, and the option that did close required infrastructure that was not there.

The order that actually works runs the other way. Eliminate the exposure first, then prevent the fall from starting, then arrest.

• Eliminate. A permanent ladder landing with a platform at deck level and a self-closing gate removes the step-over entirely. The tech is never balanced on a rung with one hand committed.
• Prevent. A tool bag sent up and down on a hand line rather than carried leaves both hands on the rails for the whole climb, which is the single change most available to a shop today. Fixed guardrail or a permanent horizontal lifeline from the ladder to the unit removes the disconnected walk.
• Arrest. Only where the first two are genuinely unavailable, and only with an anchorage high enough and close enough that the stack closes with margin.

The general fall protection duties sit at 29 CFR 1926.501 for construction work and 29 CFR 1910.28 for general industry, and the trigger heights are not the same number. A service call on an operating building is usually general industry; the same tech on an active construction project is usually construction. Name the Part before quoting a trigger, because quoting one Part's number as if it were universal is how a shop ends up protected on paper for the wrong regime.

Fixing the record so the next one is visible

The document set was accurate and blind. Three changes make the same afternoon legible, and none of them add a form.

Record the interval, not the state. Replace the single fall protection Boolean with two entries: connected from, and connected to. When those two timestamps do not span the roof time, the gap is on the page and someone can ask about it.

Record the access route as a condition of the site, not of the job. A rooftop address gets a standing note describing how the transition is protected. Blank means unprotected, and blank is a finding.

Record who has authority to stop. In a scaffolding or excavation context this is a defined role in the standard. On roof access it is often nobody in particular, which is how a known bad transition survives a hundred visits. Name the person, and make it explicit that the authority includes refusing the address until the transition is fixed.

The general failure this catches is broader than roofs. Any protection that has to be disconnected in order to travel produces an interval that the paperwork describes as protected, because the state was protected on either side of it. Look for the same shape wherever a program measures a state instead of an interval.

References

• 29 CFR 1926.502(d), personal fall arrest system criteria for construction, including (d)(16) on the 6-foot free fall limit, the 3.5-foot deceleration distance limit, and rigging so an employee cannot contact a lower level.
• 29 CFR 1910.140, personal fall protection systems for general industry, which is the counterpart Part for service and maintenance work on an operating building.
• 29 CFR 1926.501 and 29 CFR 1910.28, the duty-to-provide provisions for construction and general industry respectively. The trigger heights differ; cite the one that governs the work.
• Harness and connector manufacturer instructions, which own the published harness stretch, the required clearance figure and the basis that figure was computed on. Read the basis before adding to the number.
• See related: universal-fall-protection-systems for the components of a personal fall arrest system.