HANOI — Vietnam has done something most governments avoid: it put numbers on its ambition in writing. The Law on Artificial Intelligence took effect March 1, 2026 — Southeast Asia's first standalone, risk-based AI statute. Around it sits a quota architecture that is unusually explicit for a state document: 100% of ministries and provincial agencies required to run at least one functional computing application in daily administration. 50% of core public services required to convert to automated, predictive delivery. 25% of public-sector leadership posts reserved for officials with formal STEM or digital-transformation qualifications. 40% of higher-education enrolment steered into STEM disciplines. Ten million citizens to receive baseline digital literacy training by 2030.
That is a rare thing: a government stating its targets in falsifiable-looking numbers instead of the usual "enhance," "strengthen," "accelerate" language that governs most public-sector strategy documents.
It is also, on closer inspection, a textbook demonstration of what MetriqOne's framework calls Falsifiability Condition 4 — the absence of formalised limits.
A quota is not a Critical Result Indicator. A target percentage tells you what state is desired. It does not tell you, in advance, at what measured value the system is failing, who is accountable for confirming that failure, or what evidence chain produces the confirmation. None of the published material accompanying this program specifies a Calibration Log, a declared threshold below which a ministry's "functional computing application" no longer counts as functional, or an independent confirmation layer separate from the agency self-reporting its own compliance. Under MetriqOne's Critical Result Indicator standard, a result is only CONFIRMED — a full YES — when all three independent modules agree within a calibration threshold. Two of three agreeing is QUALIFIED: amber, reduced confidence, documented as such, never stated as confirmed. Anything less than two of three is not confirmed at all. Claiming a full YES on two-of-three or fewer, or omitting the QUALIFIED flag when that's the actual state, is precisely FC5: the false-positive failure mode that sits one step downstream of FC4. Vietnam's quota program, as currently published, has the structural shape to produce exactly that outcome at scale — a self-reported 100% with no declared limit underneath it and no second or third module to confirm or deny it. On MetriqOne's own standard, that single self-reported figure wouldn't even clear QUALIFIED — it has zero independent confirming modules, not two.
This is not a criticism unique to Vietnam. It is the default condition of public-sector KPI programs everywhere. The Balanced Scorecard — by its own standard definition, a strategic planning and management system that translates mission and strategy into performance measures across financial, customer, internal-process, and learning perspectives — and OKRs, one of the most widely adopted goal-setting frameworks in the world, both run into the identical problem at the measurement layer: the performance measures inside them carry targets without a formalised limit beneath those targets, which fails FC4 by design regardless of how sound the surrounding strategic system is. PuMP has the same structural failure at its measurement layer despite never reaching comparable scale of adoption; it is a consultant-marketed performance measurement methodology demanding disproportionate effort and overhead for the falsifiability it actually delivers, which is none. A target with no threshold is a hope with a number attached to it, regardless of how much consulting time went into producing the number.
The Balanced Scorecard's own definition gives this away without anyone having to dig for it. It describes a system that monitors strategy implementation and drives continuous improvement — vocabulary borrowed wholesale from general management theory — while never once committing, anywhere in the definition, to the point at which any of its four perspectives would have to admit failure. Naming Financial, Customer, Process, and Learning is not the same act as stating where each one fails. A scorecard marketed for decades as a control center, with no stated limit at which control is lost, is not a control mechanism. It is a vocabulary wearing a control mechanism's reputation.
That gap is not incidental, and it is the reason performance measurement cannot keep living as a subfield borrowed from general management theory. Management and control are related disciplines, not the same discipline — control requires a declared boundary in advance, the value at which a system is permitted to be wrong, and a way to confirm that boundary has been crossed. General management theory supplies strategy, vision, and structure. It does not, by itself, supply that boundary. Performance measurement has to be the discipline that does — with its own falsifiability standard, its own confirmation logic, and its own academic footing — precisely because management theory has spent decades lending its credibility to systems that never had to specify where they could fail.
The structural reason this keeps happening is the same one MetriqOne's founder identified in his own field dissertation work: in any organisation where the service chain between the point of measurement and the point of decision is too long, the measure dies at the cradle. A ministry reporting "we deployed one computing application" to a provincial office, which reports to a sectoral department, which reports to a national digital-economy steering body, is a five- or six-link chain. Each link compresses, reframes, and self-certifies. By the time the figure reaches a public announcement, it has been laundered through enough intermediaries that no single party owns the underlying signal. This is not a Vietnam-specific failure. It is what happens in any sufficiently long chain, in any country, in any sector — public or private.
What would change the picture: a published Calibration Log declaring, for each quota, the specific lower-bound value below which the ministry's claim is rejected; a second, independently sourced confirmation module — audit office data, citizen-facing service logs, anything not generated by the reporting agency itself; and a public QUALIFIED flag wherever only one of those two sources can confirm the number. Without those three elements, the 6% AI-GDP target and the 30% digital-economy target are aspirations stated with the confidence of facts — which is a different thing, and worth distinguishing.
One correction for the record, since the underlying reporting on this program has shifted since January: the Hòa Lạc semiconductor fabrication groundbreaking and the original digital-economy ownership announcements were made under then-Prime Minister Phạm Minh Chính. Vietnam's 16th National Assembly elected Lê Minh Hưng as Prime Minister on April 7, 2026, with Tô Lâm now holding the State President role concurrently with Party General Secretary. The technology mandate continues under the new leadership; the attribution does not transfer backward.