1. Load the gap ledger. Cell 1 holds all thirteen rows of Table 9 with position, binding constraint and status. Positions are the authors' assessment out of ninety degrees and are not measurements, so treat them as ordering and not as data.
2. Count the constraints. Engineering implementation 6, regulatory 3, AI algorithms 2, computation efficiency 1, workforce 1. Nine of thirteen bind on engineering or regulation. Two bind on AI algorithms.
3. Read the two open rows. A reference drift implementation at 10 of 90, and system identification priced in energy at 5 of 90. Both ask for an instrument rather than a result, which is why they are the cheapest of the thirteen to move.
4. Load the absence ledger. Cell 2 holds Table 10. Two of its six rows are held by every region that searched, and both of them are energy rows.
5. Walk the metric-class ladder. Battery envelope, production cycle energy and point-to-point task energy all have located instances. Joules per embodied task does not, in five regions of five.
6. Print the closing pair. One national standard requires the figure as a reportable. Zero measured instances of it were located in any region. That pair is the result the course ends on.
7. Report the boundary of the review as a boundary. Of 74 not-located records, 28 failed on access rather than absence, 11 of those on documents sold rather than posted. So 46 are absence as far as the review reached, and the edge of the finding is partly a paywall.
8. Keep the refutation rate. Twenty load-bearing figures went to independent sceptics instructed to refute. Four came back refuted and were discarded, a fifth of the load-bearing set. Apply that rate to your own numbers before you publish them.
9. Say what was not swept. Five regions is not the world. Russian, Hindi, Portuguese, Spanish, Italian and the Nordic languages were not read. Nothing here is a statement about those records.