Building a Repeatable Extraction Workflow
Everything we have covered becomes durable edge only when it turns into a routine you run the same way every quarter. Ad hoc reading produces occasional insights and frequent self-deception; a repeatable process produces a dated record you can trust and improve. In this lesson I want to lay out an extraction workflow concrete enough to follow, built from the techniques in the earlier modules and designed so that your conclusions are auditable, even to yourself six months later.
The workflow has five stages, and the order matters. Stage one is preparation, done before the call. Pull the prior quarter's prepared remarks, your notes on that quarter's Q&A, and the trajectory you had inferred. Write down, in advance, the specific questions you want this quarter to answer and what would confirm or contradict your current thesis. This pre-registration is the most important discipline in the whole process, because deciding what would change your mind before you read protects you from the near-universal habit of finding whatever you already believed.
Stage two is capture. As the call happens, or from the transcript, record the prepared remarks and the notable Q&A exchanges verbatim into a running file for that company. Do not summarize yet; summary discards the exact words, and the exact words are the signal. You want the raw material preserved so that next quarter's diff has something precise to compare against. A consistent file structure, one document per company, appended every quarter, is the backbone of the entire method.
Stage three is extraction, and here you apply the specific lenses. Diff the prepared remarks against last quarter and log every substantive change with its date. Pull the capex guidance apart into maintenance and growth, note the timing shape, and record the firmness of the verbs. Hunt the filings for new or changed commitments, leases, power contracts, take-or-pay, and the purchase-obligation schedule, and size them. Note the density of hedging and any high-information moments from the Q&A. The output of this stage is a structured list of dated observations, not a narrative.
Stage four is triangulation, the analytical core. Take your extracted observations and set them beside the other channels: hiring composition and geography, network and partnership activity, and public power and supply-chain signals. Ask the one question this whole course is built around: do the independent channels agree? Mark each inferred intention with how many channels corroborate it and how hard each of those channels is to fake. An intention supported by a language shift alone is a hypothesis. The same intention supported by a language shift, a hiring cluster, a reserved block of power, and a step-up in purchase obligations is a forecast you can act on.
Stage five is scoring and updating. For each thesis you hold about a company, write down the current weight of evidence, what changed this quarter, and, crucially, what would have to happen next quarter to strengthen or break it. Keep the prior quarters visible so you can see whether your inferences are being confirmed by events or quietly failing. This is where the process compounds: over a year, you accumulate a dated ledger of predictions and outcomes that tells you, honestly, which signals have actually been reliable for this company and which have misled you. Most readers never build this ledger, which is why they never learn whether their instincts work.
Two design principles hold the workflow together. The first is that everything is dated. A signal without a date is nearly useless, because the entire value of leading-indicator reading is knowing how far ahead of the headline you were. Date every observation and every inference so that later you can measure your true lead time. The second principle is separation of observation from interpretation. Keep the verbatim capture and the extracted facts distinct from your inferences about what they mean. When a thesis fails, this separation lets you see whether the facts were wrong or your interpretation was, and only one of those is fixable by trying harder.
The workflow is deliberately unexciting. It is the same five stages, in the same order, every quarter, per company. That repetition is not a limitation; it is the source of the edge. Insight in this domain does not come from a brilliant one-time reading. It comes from the accumulated, dated, auditable record that only a consistent process can produce, and from the honest feedback that record gives you about which of your signals deserve your confidence. Build the process, run it faithfully, and the interpretation gets sharper on its own, because you are finally in a position to learn from being right and from being wrong.