Chapter 16 templates · The One-Hour-a-Month Tracking Recipe Card
How to use. This card covers change in the craft of "doing research with AI" itself, not your research field (that is the job of the Chapter 4 frontier layer), and not the row-by-row re-derivation of your own map (that is the job of the Chapter 13 review cadence). One fixed hour a month, in three steps. 20 minutes on the sources → 30 minutes running the fifth column on the signals that hit → 10 minutes changing a status or a date and writing one line of change log. Checked and nothing moved, the date still changes.
Part one, the source diet list (fillable template)
The rules get locked in first:
- Total capped at 5. Add one and you must drop one first.
- A retirement review once a quarter. Any source that has not triggered a single "run the fifth column" in the past three months gets downgraded or removed.
- Influencer newsflashes and algorithmic feeds do not use up quota. That kind of information reaches you anyway and needs no active subscription.
| Class | Source name | Why it stays (one sentence) | Date it last triggered an action |
|---|---|---|---|
| First-hand source (original papers / system technical reports, not retellings) | ______ | ______ | ______ |
| First-hand source (an optional second one) | ______ | ______ | ______ |
| Production-retrospective source (an account written by someone who really used it for three months, not a day-one review) | ______ | ______ | ______ |
| Controlled-measurement source (comparison studies with a baseline and a stated basis) | ______ | ______ | ______ |
| Your choice (a high-quality aggregator and the like, use sparingly) | ______ | ______ | ______ |
Part two, the list of signals that trigger a re-derivation (any hit → this month's 30 minutes goes to it)
The general rule is one sentence. A piece of news is worth an hour of yours if and only if it could change the status of some row on some map of yours. In practice, hold it up against that row's fifth column ("what evidence would change its status") and ask, is this the evidence that column describes?
- [ ] A production-grade measurement has appeared. Not a demo, not a launch event. Real accounts like merge rate, rework rate, verification pass rate, claim survival rate.
- [ ] An independent reproduction or an independent refutation has appeared. A second pair of hands with no stake in the original authors got the same result, or the opposite one.
- [ ] A criterion locked in earlier has been tripped. A preregistered falsification condition, or the evidence described in the fifth column of some row of your own map, has surfaced.
- [ ] Repeated rework in your own workflow. The same process step crashes several times running. Your first-hand data is evidence too, and the closest kind there is.
- [ ] A policy or basis change from an authority. Journals, reviewers, or regulators changing the rules for AI's part in research. It reshapes the terrain of the "deliver and take the hits" stop directly.
Part three, the ignore list (let it flow straight past, into no notes and no anxiety)
- Demo videos and vendor launch events. Handled by row 2 of the transfer map. Qualified only to trigger an investigation, not to decide a status, and least of all to trigger anxiety.
- Self-reported speedups with no control. "I got three times faster after using X." See row 3 of the transfer map, self-perception is not trustworthy, calibration comes from measurement.
- Model version-number news. "X.5 is out" is not a signal by itself. A controlled measurement of it on your own tasks is.
- The Nth round of the "AI replaces scientists" debate. An argument fought in the air produces no evidence that can hit the fifth column.
- Secondhand retellings with no link to the original. Every hand a retelling passes through drops one layer of qualifiers. A claim whose original you cannot find is treated as nonexistent.
- Anonymous leaks and rumors. Evidence that cannot be pointed at is not evidence.
Part four, the hour's time budget (copy it as is)
| Slot | Action | Output |
|---|---|---|
| 0–20 minutes | Run through the sources on the diet list, sorting into two piles only, hits the signal list / flows past | 0–3 candidate signals |
| 20–50 minutes | For the signals that hit, run the fifth column of the matching map row. Has that kind of evidence actually appeared | Change the status / change the evidence / confirm nothing moved |
| 50–60 minutes | Update the map, changing a status or only a date, and write one line of change log | Every review date refreshed |
Part five, Self-check (failure modes condensed)
- [ ] Are you over 5 sources? Cut. The number of sources is not proportional to the quality of your judgment, it is proportional to your anxiety.
- [ ] Has the hour swollen into three hours? Then you are reading a feed, not running the recipe. Go back to "ask only against the fifth column."
- [ ] Three months running of nothing but date changes, and yet more anxious? The anxiety comes from not executing the ignore list. Without a rule that permits things to flow past, tracking is slow poisoning.
- [ ] Have you handed the map to AI to maintain fully automatically? AI can run the searches for you, but a status change must pass through a human (the last cordon of Chapter 13).
- [ ] When was the last retirement review? A list that only takes in and never lets go becomes, three quarters later, another forty-seven browser tabs.