The week after the three verdicts were finalized, Mio called Yui in. On the desk sat four envelopes. A fail for Higuchi, a conditional pass for Minami, a clear pass for Wada. "Yui, someday you'll be the one writing these envelopes. Today is practice for that." With that, Mio set a still-unaddressed envelope down in front of Yui. "Do you understand what it really means to draw the line?"
Where Is the Floor — Anchoring First
Yui raised a plain question. "Higuchi fails and Wada clears. But isn't that just your own feel, Mio? If someone else did the reviewing, couldn't it come out the opposite way?"
Mio looked a little pleased. A good question, she said. "If this verdict were just my mood, then it's not a system, only personal likes and dislikes. That's why, before the pass line is ever put to use, there's one thing we always do first."
What Mio described was anchoring first. An anchor is a sample case that fixes where the pass/fail floor sits. "You pick several real materials and mix in dangerous ones, safe ones, and ones where judgment splits. All the reviewers look at them together beforehand and align on, 'This much should obviously be caught,' and, 'Missing this is an automatic fail.' A sample that has already been calibrated — that is the anchor."
Think of a bathroom scale. Even a brand-new scale, you first set the needle to 0 kg. If you measure 50 kg without zeroing it, you can't be sure it's really 50 kg. Anchoring first is the act of zeroing the scale that is human judgment.
"Only once this calibration is done first do different nationalities, different departments, all read the same '50 kg' as the same 50 kg. If everyone measures without dropping an anchor, a fail in Tokyo becomes a pass in Osaka. That isn't a standard."
AI Drafts, the Human Stamps — Human Confirmation
Yui had one more thing on her mind. For half a year, the review room has had a support tool that underlines problem spots in the materials. "If that thing flags 'danger present,' isn't it the same as the machine deciding?"
Mio shook her head. "What the tool produces is an assessment. In doctor's terms, it's the automated comment on a health check. It may print 'further examination recommended,' but it's the physician who confirms the diagnosis and decides the treatment plan, right?"
The AI's Role
A draft assessment. It marks suspicious spots and reduces oversights. Fast. Never tired. But it cannot weigh "why this is dangerous."
The Human's Role
The final judgment. Deciding how this wording moves a healthcare professional's prescribing, including the asymmetry of harm. Carrying the responsibility of signing.
"Remember Higuchi's weak point? Good at explaining, but weak at finding danger. If everyone simply took the tool's assessment at face value, the pairing of Higuchi and the tool would look perfect — 'can explain, and detection shows up too.' But the moment you let the tool carry the responsibility for detection, no one is bearing the weight of harm."
Mio's voice dropped a little. "It's the human who stamps the approval. This isn't about efficiency. It's about who bears the responsibility."
The Four Tiers Are Not Brands — Non-Punitive Development
Yui looked at the fail envelope and unconsciously frowned. "Higuchi must be discouraged…"
"Don't get that wrong." Mio spread the four envelopes out like a fan. "Fail, conditional pass, clear pass, high-rank clear pass. These four tiers are not a ranking of people's worth. They are a signpost for what to grow next. Same as a health check result. 'Re-examination required' is not a denial of your character; it's a map of where you should take care, isn't it?"
| Tier | Read as a brand | Read as a signpost |
|---|---|---|
| Fail | "A useless person" | Toward training that rebuilds detection skill |
| Conditional pass | "Half-fledged" | A stage of gaining real-case experience under supervision |
| Clear pass | "Excellent" | Toward working independently and mentoring those coming up |
"Hand it over as punishment, and from then on people hide their weaknesses. Once hidden, danger becomes invisible. That's the scariest thing. So the pass line is never made a tool of punishment. The more someone has fallen short, the more carefully you hand them the map of their room to grow."
Maps for the Three
One by one, Mio had Yui read the development plans.
Higuchi. Articulate, with a real ability to command a room. But low sensitivity for picking up danger. In retraining, he is pulled off the explaining role for now and, next to Wada, drills only on "finding the one dangerous line in a real piece of material." "His explanatory power becomes his greatest weapon once detection skill has grown. The order was simply reversed."
Minami. A theorist who can name every type of problem. But on the actual material in front of her she can't pick it up — detection on paper, so to speak. As a conditional pass, she stacks up real-case reviews under the supervision of a clear-pass reviewer. "She builds the bridge between knowledge and the field with her own hands. Re-evaluation in half a year."
Wada. Plain and unassuming, but holds the genuine detection skill to pick up danger in real material. A clear pass, recognized as an independent reviewer. "And next, Wada becomes the one who shows Yui the anchor samples."
Keep this in mind. The responsibility of drawing the line is not the failing of people. It's drawing, together, the very foothold from which the one who fell short stands up again. If you can't draw that foothold, you're not yet qualified to draw the line.
The Unaddressed Envelope
Yui noticed the blank envelope still sitting on the desk. "And this one…?"
"The envelope you'll write someday. A day will come when you hand it to someone." Mio looked out the window. "When that day comes, remember. Drop the anchor first. The human is the one who stamps. The tiers are signposts, not brands. A line drawn forgetting these three is nothing but power."
Yui took the blank envelope in both hands. It had almost no weight. And yet the whole road she had walked — from the day in the third installment when she first met Higuchi, to the day in the ninth when the three verdicts were finalized — felt folded inside that light sheet of paper.
"I'll try." That was Yui's answer. Mio gave a small nod and tucked the four envelopes back into the drawer. And so the pass line was handed on to the next person.
The Qualification Bar ── Map of all 10 episodes
- Vol. 2: The Asymmetry of Harm ── A Miss Is Orders of Magnitude Heavier ── Why you must not draw the line with an average, part 1: a miss and a false alarm are not equal harms
- Vol. 3: The Compensation Trap ── Eloquence Hiding a Gap in Detection ── A reviewer who is brilliant at explaining and at getting along with people is weak at just one thing: spotting danger (risk detection). Average the scores and they pass. But someone who cannot spot danger yet talks well will push risky material through on charm alone. Why you must not decide pass or fail on an average — explained gently through real Case A.
- Vol. 4: Floor vs. Aggregate ── Non-Compensatory Gates and the Weighted Score ── Pass/fail is decided by minimum bars (floors); the total score is used only to rank. Fall below even one bar and a perfect score still fails. This is the unbreakable rule of the qualifying line.
- Vol. 5: The Highest Floor for Detection ── Why Risk Detection Exists ── Material review — the job of checking a drug company's promotional materials for doctors before they go out — exists to find the dangerous spots. So among eight abilities, the minimum bar for the power to spot danger (risk detection) is set highest. To pass as someone who can review alone (qualified) you need level L3, the second-highest rung, plus a real-world spotting range of 2 or more. A person who stops one rung lower, at L2, lets the most dangerous materials slip right through.
- Vol. 6: A Floor on Two Axes ── Not Letting Desktop Detection Pass ── The pass line for detection cannot be drawn with a single score. It needs two rulers: how well you can explain the danger, and whether you can catch it in the real material in front of you. A textbook-only spotter may look like L3 on paper but does not clear for solo work.
- Vol. 7: Calibration as a Gate to Independence ── Overconfidence Disqualifies ── A look at the gate (calibration gate G2) that asks: do you estimate your own seeing-power correctly? Working alone means no one checks behind you. A person who thinks their detection skill is higher than it really is (gap Δ of +2 or more) waves through danger without noticing their own blind spot. This gap (Δ) is not skill itself, but it decides whether someone may work alone.
- Vol. 8: The Four Gates G0–G4 ── The Logic of Early Rejection ── A reviewer's pass or fail is decided at four checkpoints in order. Anyone who fails an earlier checkpoint is not re-measured at a later one. A non-negotiable minimum line (a "floor") cannot be patched over by other strengths, and the total score never flips the result.
- Vol. 9: Three Profiles ── How One Line Sorts Them ── The eloquent talker, the textbook thinker, and the real deal — where one pass/fail line sends each
- Vol. 10 (this episode): The Responsibility of Drawing the Line ── Anchors First, Human Confirmation, Non-Punitive Growth ── The closing chapter that turns the pass line into something a workplace can actually use. Only when a shared book of agreed examples exists does the line become a common yardstick. The four verdict tiers are not a brand of failure but a signpost for what to grow next. AI gives a rough first reading; a human makes the final call.
In the end, the pass line becomes the field's standard through three mechanisms. Anchoring first, where everyone aligns their eyes on sample cases before the line is used. Human confirmation, where the machine's assessment stays a draft and the human does the final stamping. And non-punitive development, where the four tiers are handed over as the next foothold rather than as punishment. Only when these three come together does a verdict rise above one person's likes and dislikes, so that whoever reviews, they share the same floor.
Higuchi, Minami, and Wada were not lined up by superiority. Each took the map of where to grow next and set off down a separate road. The responsibility of the one who draws the line was not to fail people, but to draw all the way through to the foothold from which those who fell short stand up again.
- Anchoring first — the substance of the floor is fixed before use, with sample cases calibrated among reviewers. Only then does it become a shared standard across nationality and department. A ruler with no calibration is nothing more than personal feel.
- Human confirmation — what the AI produces goes only as far as an assessment (a draft) of suspicious spots. Weighing the gravity of harm, signing, and bearing responsibility — the final judgment — is the human's to carry. Don't let the machine shoulder the responsibility for detection.
- Non-punitive development — the four tiers of fail / conditional / clear pass / high-rank clear pass are not brands but signposts for what to grow next. Make them punishment and people hide their weaknesses, and danger becomes invisible. The more someone has fallen short, the more carefully you hand them the map of their foothold.
- \n
- Angoff, W. H. Scales, Norms, and Equivalent Scores. American Council on Education, 1971. (The classic on standard setting; the origin of grounding thresholds in expert consensus.) \n
- Messick, S. Validity. In Educational Measurement (3rd ed.). Macmillan, 1989. (Validity and consequential validity of judgments; theoretical backdrop for appeals and transparency.) \n
- Cizek, G. J. & Bunch, M. B. Standard Setting: A Guide to Establishing and Evaluating Performance Standards on Tests. Sage, 2007. (Practice of setting anchors and cut scores and recalibrating them.) \n
- Spencer, L. M. & Spencer, S. M. Competence at Work: Models for Superior Performance. Wiley, 1993. (Foundation for competency judgment linked to development.) \n
- Swets, J. A., Dawes, R. M. & Monahan, J. Psychological Science Can Improve Diagnostic Decisions. Psychological Science in the Public Interest, 2000. (Sensitivity/specificity and the asymmetric treatment of misses versus over-detection.)