This is the fourth column in the series, and the one that closes it. Let me retrace the road first. Part 1 was about method: how I taught artificial intelligence to read valuation reports. Part 2 was about coordinates: where marketability discounts and control premiums have actually landed in this market over ten years. Part 3 opened the two ends of the distribution, case by case, and traced the divergence between the Hong Kong and mainland valuation traditions. Three columns, and in truth they did only one thing: they placed a common yardstick over a market no one had ever read in full.

This column answers the one question readers have asked most since the series began. Now that the yardstick is in place, what does it mean for regulation, for the profession, and for every gatekeeper who signs a circular?

Consistency: measurable for the first time

When the profession talks about valuation quality, the phrase that appears most is “professional judgment”. Judgment is invisible, so discussions of quality tend to end in a stalemate of opinions. The first change the yardstick brings is that “consistency” turns from an adjective into something measurable, and measurable at three levels.

At the market level, Part 2 already told the story: annual average discounts hovered between 18% and 22% for ten years, premiums broadly between 20% and 27%, and the centre barely moved. The profession as a group does have a consensus, and that is a mark of maturity.

The firm level is new. Every valuation firm has its own numerical fingerprint: the studies it habitually cites, the levels it habitually lands on, the method combinations it habitually runs. Same firm, similar subjects, two engagements apart in time: were the stances consistent? There was no way to check before. Now the cases can be laid side by side. Consistency does not prove correctness, but inconsistency always has a reason, and the reason belongs in the report.

The case level is the most direct: is a number consistent with the very study it cites? Part 3 showed a report citing a study with a median of 15.8% and adopting 50%. Let me say it again: deviation itself is no sin, and reasoned deviation is exactly where professional judgment lives. There has only ever been one question. Was the reason for the deviation clearly written down?

A gatekeeper’s job is not to demand that everyone be identical. It is to demand that differences come with reasons. For the first time, that demand has data behind it.

Transparency: five asks that need no new law

Say “regulation” and many minds jump to new rules and new regimes. I want to argue for something more modest: the five asks below need no new rule at all. The profession can adopt them today, on its own.

First, state which study was cited and which slice of it was taken: the median, or one end of the range. As Part 3 showed, the choice of study often settles the outcome before any analysis begins. That choice should be laid open.

Second, state why that study fits this particular subject. Reports with a derivation but no connection have recurred throughout the decade.

Third, the number must reconcile on the value bridge. How much was adopted, at which step it was added or subtracted, and how the conclusion follows: a reader should be able to trace it.

Fourth, when no adjustment is adopted, write down why. “No separate marketability discount is applied in this valuation because liquidity is already reflected in the discount rate” is a disclosure, not an explanation. Complete silence is, with respect, hard to accept. As Part 3 noted, the absence of a discount line does not mean the absence of a discount effect, and readers deserve to know where the effect lives.

Fifth, take Part 3’s ask one step further. Part 3 proposed that every report should state which set of standards it follows. In preparing this column, I went back through the 104 circulars whose valuations were performed by mainland appraisal firms, and found that disclosing the standard is in fact already done: wherever a standard is stated, it is the PRC Asset Appraisal Standards, and not one of those reports cites the International Valuation Standards (IVS). Among business valuation reports by Hong Kong and international valuers, most cite IVS or the Business Valuation Standards of the Hong Kong Business Valuation Forum. Each side declares its school, and the schools genuinely differ. But the real divergence is not in the name of the standard. No standard prescribes, clause by clause, how discounts and premiums must be handled, and the 51% against 11% divide in Part 3 is convention, not text. But expecting valuers to explain the difference between two conventions inside a report is not realistic either. The workable form is in fact simple: it takes nothing more than answering two fixed questions. Was liquidity considered, and where does it land: a separate discount line, inside the discount rate, or not applicable under the basis of this valuation? The same for control. The answer need not be long, but it must be sufficient. A reader who knows nothing about the two traditions can still see where these two factors went. The name of the standard was always visible to the reader. Where the numbers went is what the reader cannot see.

None of the five waits on regulatory action. What they need is a habit: writing down the judgments the valuer has in fact already made. And the moment they are written down, comparison becomes possible, and gatekeeping gains a foothold.

One more question for every post

Part 3 counted the gatekeeping ecosystem: all 297 transactions carrying a discount or premium were approved by boards, 482 circulars named a reporting accountant, 314 carried an independent financial adviser’s opinion, 118 engaged a further financial adviser, and every circular passed the Exchange’s review. What was missing was never gatekeepers. It was a common yardstick within every gatekeeper’s reach. Now that the yardstick exists, each post needs to ask only one more question.

A board can ask: where does this percentage sit on the ten-year distribution? If it sits at the far end, did the report explain why? The four questions from Part 2, none of which requires valuation expertise, still apply.

An independent financial adviser can ask: before signing a “fair and reasonable” opinion, can one sentence in the letter place the adopted adjustment against the market’s normal range?

Accountants and auditors can ask: purchase price allocations and goodwill impairment tests after completion meet these same discounts and premiums, so does the same yardstick not serve there too?

As for regulators, I am not advocating any new regime here. But the format and consistency of disclosure have always been where regulation can press gently, and every one of the five asks above belongs to that territory.

Last, the valuers themselves. Let me be very clear on one point: this yardstick is not aimed at valuers. Quite the opposite, it protects the serious ones. When a rigorously derived number and a casually written one cannot be told apart in the market, it is the rigorous one that loses, because rigour has a cost. With coordinates in place, a reasoned deviation stands firm and an unreasoned one has nowhere to hide. For the first time, those who do the work properly have evidence to point to.

The division of labour between AI and human judgment

Finally, back to where the series began: where should artificial intelligence stand in all this?

After ten years of data, my answer is simple. The machine’s job is to read everything. The human’s job is to judge well.

In this research, what the AI did was this: read more than 4,000 circulars end to end, trace every figure back to its exact source text, re-check the arithmetic on the value bridge, and pick up each quarter’s new circulars automatically. What these tasks share is scale and discipline, which is precisely what humans cannot sustain and machines excel at. What the humans did was this: judge whether a study fits the subject, whether a premise is reasonable, whether a transaction is fair to minority shareholders. What these tasks share is context and responsibility, which is precisely what machines cannot do and should not be asked to carry.

The limitation must be stated once more: this dataset was compiled by artificial intelligence, my team and I sample-checked it, and perfection cannot be promised. But every figure is traceable and checkable, and the method and its limits are laid open in full. The real dividing line was never “done by AI” versus “done by humans”. It is whether each judgment is laid open for verification.

So I do not worry about artificial intelligence replacing valuers. What it replaces is the state of “no one has read them”. Judgment remains human. It simply has to be exercised in daylight from now on.

The instrument keeps running

The series ends here. The research does not. The instrument keeps reading each quarter’s new circulars, and the coordinates keep growing with the market. And as large AI models mature, the cost of modifying and improving the programme keeps falling. One day soon, you and I could each take up the gatekeeper’s role with ease. The question Part 3 left open is already on the extension-study agenda: in valuations that apply no discount, and discounted cash flow valuations in particular, are the discount rates systematically higher? That question should be answered by data, not by standpoint. And the full research paper is being prepared for publication. My team and I are glad to maintain this yardstick as an independent third party, and we welcome peers in the profession to use it, question it and improve it. The value of a yardstick lies not in who owns it, but in how many people measure with it.

Gatekeeping, in the end, is one sentence: every number capable of moving a valuation should be read by someone, comparable to something, and explainable for a reason. For ten years, that sentence was out of reach, because no one had read them.

Now someone has.

(Gatekeeping Valuation Quality in Hong Kong, Part 4. The end of the series.)