🇰🇷 한국어 버전 (Korean version)

Seventy years ago, a quality engineer insisted on a single word — and that word is now reappearing, unchanged, in the design of long-term memory for AI agents. Not Check. Study. This article traces the history of that word, records a hands-on session with pdsa — a CLI tool that turns Deming's philosophy into graph engineering — and finally introduces the record of 22 cycles in which the tool was used to build itself.

1. The History — Quality Is Learned, Not Inspected

🎨 [Image #1 — Crossing out Check, circling Study] Japan, 1950. CHECK is erased from the blackboard and STUDY gets the circle. This entire article is the story of that one stroke of chalk.

From Shewhart to Deming

In the 1920s–30s at Bell Labs, Walter A. Shewhart established statistical quality control along with an iterative improvement loop later called the "Shewhart Cycle." His pivotal shift was simple: instead of filtering out defects through inspection at the end of the line, make the production process itself produce uniform results.

W. Edwards Deming, who worked with Shewhart in the 1930s, pushed this further — from shop-floor statistics into a philosophy of management. PDSA is the direct descendant of the Shewhart Cycle, refined by Deming.

The Price of Putting Quality Last — the Atari Landfill

The clearest cautionary tale of what volume-without-quality costs is the 1983 North American video game crash. Low-quality games from unlicensed developers flooded the market, and home video game revenue collapsed from roughly $3.2 billion (1983) to about $100 million (1985) — a nearly 97% drop. The emblem of that era is Atari's E.T. (1982): rushed out in five to six weeks, five million cartridges produced, only 1.5 million sold. Around 800,000 leftover cartridges were buried in a landfill in Alamogordo, New Mexico. Long dismissed as an urban legend, they were actually excavated in 2014.

What revived the dead market was also quality. Nintendo launched the NES (1985) with the Seal of Quality and a licensing lockout chip (10NES), enforcing a release gate. The strategy was "fewer, better games." Where uncontrolled volume had destroyed trust, enforced quality rebuilt it.

Deming Goes to Japan — and the West Relearns, Late

Invited by the Japanese Union of Scientists and Engineers (JUSE), Deming arrived in Japan in July 1950 and taught statistical process control together with management philosophy. Toyota adopted company-wide quality control in 1961 and won the Deming Prize in 1965; those principles became part of the philosophical foundation of the Toyota Production System. (TPS is, of course, also the achievement of Taiichi Ohno and the Toyoda family, and Japan's industrial rise had many engines. Quality was a major factor, not the only one.)

Quality theory was born in America, but Japan practiced it first and most completely. The moment the U.S. began to relearn it is usually traced to NBC's June 1980 documentary "If Japan Can... Why Can't We?" (contrary to popular belief, it was NBC, not the BBC). The Malcolm Baldrige National Quality Award (1987) and Europe's EFQM (1991) were later modeled explicitly on Japan's Deming Prize.

The Final Correction — Study, Not Check

Here is the most important part. The cycle is commonly taught as PDCA (Plan-Do-Check-Act), but in 1986 Deming warned his audiences that this version is not accurate. His reason is striking — in English, "check" means "to hold back." In letters he was even more blunt: "Be sure to call it PDSA, not the corruption PDCA." What he insisted on was PDSA (Plan-Do-Study-Act), with Study as the third step.

It can look like quibbling over a single word. But the more you unpack it, the more remarkable the observation becomes. A check is literally a stopping motion — the moment a verdict lands, the activity ends. Pass, and the next batch moves on; fail, and it gets sent back. Either way, the activity needed to improve the next iteration of the process never happens there. This is not to say checking is meaningless. It is that checking alone stops the loop in its tracks. To roll the loop into the next cycle, learning has to sit on top of the verdict.

PDCA asks "did it work?" PDSA asks "what did we learn, and how should that change our theory?" It is the second question that lets knowledge compound across cycles.

Harness × Loop — a Blueprint Drawn Before IT Was Born