A Factory Strike Looks Like a Supply Chain Shock. US Stoppage Data Disagrees.

The mental image is fixed and satisfying. Pickets go up outside a plant gate in one country; four weeks later an assembly line idles in another, staffed by people with no dispute. It reads as a parable about interdependence, and it is recycled whenever a contract expires at a supplier nobody outside the industry can name.

Run the image against the stoppage record and it survives only as a special case. Of the 306,800 workers idled by major work stoppages beginning in the United States in 2025, manufacturing accounted for 6,200 — two percent. The remaining 98.0 percent sat in service-providing industries, education and health services alone accounting for 196,500. That does not make propagation a myth; it relocates the question. Propagation is a property of the node, the buffer, and the legal architecture governing how a dispute may travel, and those vary far more by jurisdiction than the strike does.

What the stoppage record counts, and what it leaves out

The reference series is the Bureau of Labor Statistics work stoppages programme, and its definition needs stating first. A major work stoppage “involves 1,000 or more workers and lasts at least one shift during the work week, Monday through Friday excluding Federal holidays.” It counts strikes and lockouts together, and nothing below that line.

Within that frame, 2025 was active by recent standards. Thirty major stoppages began, against an average of 17.8 per year over 2006 to 2025 — an average made nearly meaningless by a historical band running from 5 in 2009 to 470 in 1952. Cumulative idleness reached 711,600 days in government and 784,400 in private industry. Composition is what matters: state and local government accounted for 232,800 idled workers across 17 stoppages, private industry for 74,000; public administration contributed 82,300 and other services 21,800. An above-average year was, distributionally, a year of classroom, hospital and municipal disputes.

Early 2026 is running colder. Preliminary figures show a single major stoppage beginning in June 2026, involving 1,000 workers, with 10,000 days of idleness in effect. One month is not a trend and preliminary data gets revised, but nothing recent supports treating factory stoppages as a rising source of aggregate disruption.

The censoring problem cuts the other way

The complication runs against everything above. The 1,000-worker threshold censors exactly the case the archetype describes. A plant with 400 employees that is the only qualified source of a machined component never enters the series, yet its stoppage could idle more capacity than a 5,000-worker dispute at a firm with three qualified alternates. The series is strong evidence about where labour conflict concentrates and weak evidence about where supply chain risk sits.

Three gates stand between a stoppage and a supply chain event

Gate one is concentration, not headcount

Transmission depends on whether the struck node has qualified alternates, and qualification is a technical and regulatory status rather than a commercial one. A supplier can have five competitors with idle capacity and still be a single point of failure if the buyer has qualified only one. In regulated categories — aerospace structures, implantable devices, automotive braking components — requalification runs on audit calendars indifferent to the shortage. Struck workforce size describes the bargaining unit, not the substitutability of its output.

Gate two is buffer, measured at the part number

The useful measurement is days of cover on the specific input, not the aggregate inventory-to-sales ratio a downstream firm reports. Aggregate inventory can be ample while a part number is thin, and firms do not disclose cover at that granularity. Only direction is observable.

Gate three is the legal shape of the spread

The third gate is where jurisdiction enters, and most commentary passes over it. In the United States, section 8(b)(4) of the National Labor Relations Act makes it an unfair labour practice for a union to “induce or encourage any individual employed by any person engaged in commerce… to engage in, a strike or a refusal” to handle goods, where an object is to force a neutral employer to stop doing business with the struck employer.

The consequence is structural. A dispute cannot lawfully be propagated sideways by the union itself, so downstream disruption is physical — the part does not arrive — and is bounded by the struck node’s output share rather than compounding through sympathy action. The ceiling is calculable. Where solidarity action is permitted, the ceiling is organisational reach instead.

Jurisdiction by jurisdiction: who transmits, who absorbs

United States, private sector under the NLRA

The statutory fuse is long and public. Section 8(d) requires a party seeking termination or modification of a contract to serve “written notice upon the other party to the contract of the proposed termination or modification sixty days prior to the expiration date,” and to notify the Federal Mediation and Conciliation Service within thirty days. For health care institutions those periods extend to ninety and sixty days, and section 8(g) requires a labour organisation to notify the institution and the FMCS in writing at least ten days before any strike or picketing.

Read against the 2025 composition, the sector dominating idled-worker counts operates under the longest statutory notice regime in the private economy: ten days’ warning of the action, ninety of the bargaining. Manufacturing counterparties get sixty days on the contract and no separate warning of the walkout. The sector with the most strike activity has the most advance notice, which is one reason counts and disruption diverge.

United States, rail and air under the Railway Labor Act

The network layer runs on a separate statute with an explicit freeze. Under 45 U.S.C. 160, an emergency board must “make a report thereon to the President within thirty days from the date of its creation,” and “for thirty days after such board has made its report to the President, no change, except by agreement, shall be made by the parties to the controversy in the conditions out of which the dispute arose.” Two sequential thirty-day windows, both dated from an observable administrative act.

This converts open-ended labour risk into a calendar. Board creation is the event to watch; after it, the earliest lawful action date is arithmetic. RLA disputes therefore surface as pre-deadline freight surges and precautionary embargoes rather than post-strike shortages.

United States, the national emergency route

A third layer exists for disputes judged to imperil national health or safety. Under 29 U.S.C. 179, at the end of a sixty-day period the board of inquiry reports to the President; the National Labor Relations Board “within the succeeding fifteen days, shall take a secret ballot of the employees” on the employer’s last offer and “shall certify the results thereof to the Attorney General within five days thereafter.” Sixty, fifteen, five — but only where the imperilment finding can be made, so it is no general backstop against economic disruption.

Canada, federal jurisdiction: the most permissive of the three

Canada pairs short notice with a deliberately narrow essential-services carve-out. Under the Canada Labour Code a union must give the employer notice “at least seventy-two hours in advance,” an employer owes the union the same for a lockout, and a copy goes to the Minister in both cases.

The carve-out is where the design shows. Section 87.4(1) requires the parties to “continue the supply of services, operation of facilities or production of goods to the extent necessary to prevent an immediate and serious danger to the safety or health of the public.” Commercial harm downstream is not the test and cannot be made the test by argument. A maintenance agreement is due “no later than 15 days after the day on which notice to bargain collectively has been given,” and the Board must decide any application within 82 days of receiving it.

The federal Canadian regime is therefore, on its face, the most permissive of the three toward propagation and among the most predictable in timing — a choice to protect the public from danger rather than commerce from interruption.

Germany: the works council firewall

Germany produces a different shape, and one provision explains it. Section 74(2) of the Works Constitution Act states that “industrial action between the employer and the works council is unlawful; the foregoing does not apply to industrial action between collective bargaining parties.” The plant-level body with the most detailed knowledge of the plant is barred from calling a stoppage there; industrial action belongs to the sectoral union and the employers’ association.

German industrial action consequently arrives sector-wide rather than node-wide: broad and shallow rather than narrow and deep. That is painful for output statistics and comparatively benign for single-point-of-failure analysis, because it never produces the isolated choke point. A buyer facing sectoral action loses a fraction of many suppliers; one facing single-node action loses all of one.

Winners and losers when a node goes dark

The losing side is mostly known in advance. Buyers with a single qualified source in a regulated category are exposed on gates one and two at once, build-to-order sequencing operations carry less cover by design, and firms whose supplier map terminates at tier one cannot see the node that fails.

The winning side is less discussed and more durable. A dual-sourced buyer gains allocation priority at the surviving supplier and often keeps it after settlement, because switching back carries its own qualification cost. The struck node itself can end better off if the shortage lifts realised pricing after restart.

The category that is rarely counted

A fourth group distorts the evidence base. A downstream firm running its own inventory correction acquires, during a supplier dispute, an attribution unfalsifiable at the time: a margin miss readable as demand weakness gets read as a supply event. So treat named attribution differently from unnamed. Naming the struck counterparty and quantifying units foregone is a checkable statement; citing unspecified “supplier disruption” is not.

The 1998 case, and what it does not prove

The strongest documented instance of narrow-and-deep transmission in the United States is well measured, which is why it is cited constantly. The Federal Reserve industrial production release of 14 August 1998 records “strikes at key General Motors parts plants, which began in early June and were settled in late July.” Motor vehicle assemblies “dropped from a seasonally adjusted annual rate of 12.4 million units in May, to 8.3 million units in July,” and production of motor vehicles and parts fell about 15 percent in each of those two months. A dispute originating at parts plants produced a footprint visible in national industrial production.

What it does not establish is a base rate. It is cited because it is the largest such episode in the modern series, not because it is representative — the selection is on the outcome. The vertical structure that made it possible, one assembler drawing critical components from captive parts operations, is not how most current supply relationships are built, and the episode is twenty-eight years old. It sets an upper bound, not a central estimate: the tail when gates one and two fail together at a node feeding an entire assembly network.

What a Skeptic Would Press On

Several conditions would invalidate the framing above.

• Sub-threshold concentration. If stoppages below the 1,000-worker line are rising, the headline series would show nothing while node risk climbed. No published national series exists at that granularity, so this objection cannot be resolved with public data.
• Qualification-bound categories. Where a regulator rather than a purchasing department controls whether an alternate may be used, a small stoppage can produce a long disruption and the strike-size heuristic fails completely.
• Non-labour shutdowns. A plant closed by a regulator, a fire or a customs action produces the same physical result with none of the notice structure, and statutory fuse analysis has no purchase on it.
• Bargaining structure drift. If North American bargaining moves toward pattern or multi-employer forms, the German broad-and-shallow shape displaces the narrow-and-deep one, and single-node modelling becomes the wrong tool rather than an incomplete one.
• Financial rather than physical transmission. A thinly capitalised supplier that breaches covenants during a prolonged stoppage can fail permanently. The disruption outlives the settlement and no gate analysis applies, because the node does not return.

What to Watch Next Week

• The BLS monthly work stoppages detail. Separate stoppages beginning in the period from days idle in effect during it. A low beginning count with high in-effect idleness means disputes are running long.
• Creation of any Presidential Emergency Board. This starts the thirty-day report clock and the subsequent thirty-day status quo period, fixing the earliest lawful action date.
• Section 8(g) ten-day notices in health care. With education and health services accounting for 196,500 of 2025’s idled workers, this is where the volume sits. Notices go to the institution rather than a central register.
• Canadian seventy-two-hour notices. Copies go to the Minister, and any maintenance-of-activities application carries the 82-day determination deadline.
• Attribution language downstream. Whether firms name the struck counterparty and quantify units foregone, or cite unspecified disruption.

Concrete Framework — Where to Look First

A checklist for assessing whether a stoppage will travel, in the order the questions should be asked.

1. Establish qualification status first. Ask how many sources the buyer has qualified for the part, not how many exist in the market. If the answer is one, headcount is irrelevant.
2. Determine the governing regime and its fuse. Sixty days, ninety plus ten, thirty plus thirty, or seventy-two hours. In notice-heavy regimes the response clusters ahead of the earliest lawful action date — positioning after the walkout is positioning after the information.
3. Read any maintenance-of-activities standard literally. A public-health-and-safety standard, as in Canada, will not protect commercial continuity. Assuming otherwise is a common cross-border error.
4. Check whether the dispute can spread organisationally or only physically. Where secondary action is barred, the ceiling is the struck node’s output share; where it is not, the ceiling is organisational reach. Separately, a stated inventory policy is a direction, not a number, and should not become days of cover in a model.
5. Track allocation after settlement, and judge the base rate on composition rather than the count. Volume that moved to an alternate frequently does not move back, because reversing carries its own qualification cost; that is where the durable effect sits. And thirty stoppages with 2.0 percent manufacturing participation is not the signal that thirty with thirty percent would be.

The unresolved question is one public data cannot answer. Everything above rests on a series beginning at 1,000 workers, and the disruptive case may sit below that line. The defensible posture is that the visible strike map has moved away from manufacturing and the invisible one has never been measured.

The parable is not wrong. It describes a tail event and is read as a description of the distribution. The stoppage that matters is rarely the biggest one in the headlines; it is the small one at a node with a single qualified alternate, in a jurisdiction whose essential-services standard was never written to keep an assembly line running.

This is analysis of published statutory frameworks and official statistics, not investment advice, and no recommendation regarding any security or issuer.