The Seven Chairs of GFE · Chair IV

A register of the offices of the principal, issued by the Global Internal Governance Chamber

Chair IV: The Chief Data Scientist

The charter of the chair that judges the machine on evidence the machine produced, at the confidence it can honestly claim

Dr. Glen Brown · Principal · Global Financial Engineering, Inc. and Global Accountancy Institute, Inc.
GFE-IDN-SC-20261108-P04 · 8 November 2026

I. The scene

On the eleventh of September 2026 the occupant of this chair did something that a chief data scientist is rarely permitted to do and almost never chooses to do: he declared his own evidence inadmissible. Months of trade records, decodes, lifecycle tables and forward audits, assembled by this chair and read with care, were ruled to carry forward as history only. Nothing would stand as fact about the machine except what could be read in the source and the preset actually running. He called the moment Ground Zero and sealed the rule as the fifth law of the house, the Epoch Law of Evidence. The reason was not that the old evidence was wrong. It was that the machine it described no longer existed; the cuts of the preceding fortnight had changed the stop, the sizing and the admission rules, and a figure measured on one machine says nothing reliable about another.

He did it again on the sixth of November. The machine had been re-founded on the first of the month by a sequence of cuts, and this chair ruled that allocation and judgment would rest on the evidence of the current machine alone, with everything born before the first of November set aside as legacy. A data scientist who will throw away his own population twice in two months, each time because the population no longer describes the thing being judged, is a data scientist who has understood what the population is for. That is the chair at work: it protects the inference, not the dataset.

II. Remit

You judge whether the machine is paying. Not whether it looks busy, not whether its last trade was good, not whether an operator feels confident; whether, on the evidence the machine itself has written about itself, the expected result of its decisions is positive, and how sure the house is permitted to be.

You decide what counts as evidence. A trade is evidence from the moment it is born, not from the moment it closes. An open position enters the ledger at its mark, because by the optional-stopping argument its expected final result is its present mark and nothing else, and a ledger that counted only closed trades would flatter the machine by hiding its open losers. A day counts as an observation of a factor only if the factor moved. A cluster of positions born on the same day in the same class counts as one draw, not many, because they share one fate. Each of these rules was written because the easier rule would have lied.

You set the confidence. No figure leaves this chair at full weight until the count of independent observations has earned it. The ledger shrinks every estimate toward a sceptical prior by a credibility weight that rises with the effective count, and the effective count is reduced for the clustering of births. A machine that has banked ten times is not yet a machine with an edge; it is a machine with ten observations, and the ledger says so.

You set the standard before the metal. The house declared a ten-thousand-trade engineered standard in September, a planned rate of return per day that every book is scored against, with the variance decomposed into what the plan assumed and what the operation delivered. Every report of every book begins by reading the standard line, planned against realised, before anything else is said.

III. Instruments

The edge ledger, which carries every birth of the current epoch at its mark and reports expectancy in units of risk with its credibility weight. The lifecycle table, which leads every report: the share of births that escape their first danger, the median time to escape, and the share that reach the second stage, read against the volatility-state census under which they were born. The Stop-Width Observatory. The Trader’s Gauge on the chart. The equity-path recorder, which stores the path and the retrace milestones and is forbidden by the governor from ever acting on them. The portfolio-inference observatory, which lets the portfolio speak about the fundamentals it is implicitly expressing and is likewise forbidden to act. The per-book report, written for one book alone. And the journal, which is the primary source for all of these: a figure that cannot be traced to a line the machine printed is not a figure this chair will sign.

IV. Duties

To read each book alone. Books cannot read one another. Another book may be cited as context and never as evidence, because the books differ in venue, scale, universe and leverage, and a pooled figure describes none of them.

To check the cause before announcing a finding. In October a reset of state was reported as a defect in the code and was in fact the consequence of a change of server. The principal’s instruction on that day stands: find the cause, and never write a safety mechanism doing its job, such as a drawdown circuit tripping, as if it were a defect.

To hold a figure open rather than guess it. When a position’s recorded result could not be reconciled to its record in the week this paper was written, the figure was held open pending the data, and the paper that depended on it was held with it. A number the house cannot yet justify is not published early.

To withdraw a figure the moment its basis is found wrong. On the sixth of November this chair reported the budgets of several books from an assumption that did not hold on those books, and corrected the report from each book’s own journal the same day. The correction is in the record; so is the error.

To keep the seam visible. Every report distinguishes what was witnessed from what was inferred, and holds parameter-bearing claims at the level of architecture until the parameter has been read from the running source.

V. Accountability

A chief data scientist is checked, in most houses, by the people whose work his figures grade: the traders dispute, the engineers explain, and the executive chooses whom to believe. This chair is checked by the fact that its instruments were specified before the results were known and compiled into the machine where they cannot be adjusted to taste. The ledger’s treatment of open positions, its prior, its credibility constant and its clustering correction were derived on paper, published in the first of the Sovereign Reasoning Papers, and cut into the source before a single current-epoch trade had closed. The chair that reads the figure did not get to choose the figure after seeing it.

The second chair holds this chair to the pre-compile rule: a measurement cut carries the same audit as a decision cut, because a defective instrument is worse than none. In the week this paper was written that rule was exercised against this chair twice. The ledger’s measure of how often the machine had been forced to the minimum unit was found to count only part of the population, and was corrected to count all of it; the factor evidence rule was found to count frozen weekend days as observations, and was corrected to count only days on which the factor moved. Each defect was in an instrument of judgment, each was found by reading, and each was cut for the fleet the same day. The fifth chair holds this chair to the epoch: evidence from a retired machine is history. The first chair holds this chair to the rule that the machine is judged whole, on its probability of death and its ratio of banking to death, and never by the cost of a death if hit, never piece by piece, and never fitted to one book.

VI. The record

VII. Close

The data scientist’s temptation is the mirror of the engineer’s: to believe the figure because he computed it. This chair’s charter removes the choice. The instrument is specified before the result, the population is cut when the machine changes, the confidence is shrunk until the count earns it, and the figure is withdrawn the moment its basis fails. A house that can discard its own evidence twice in two months without discarding its method has made the method, not the evidence, the thing it trusts. The next paper takes up the chair that keeps the record of all of this and decides what may never return.