Rina Dechter's 1986 AAAI paper genuinely uses 'deep learning' and 'shallow learning' as terms of art — for depth of constraint-satisfaction search, not neural networks

Dechter, "Learning While Searching in Constraint-Satisfaction-Problems" (AAAI-86, pp. 178-183), studies backtracking search enhanced by recording constraints discovered at "dead-ends." She defines "deep" versus "shallow" learning as a parameter of how much conflict-set information to record and report:

"Discovering all minimal conflict-sets amounts to acquiring all the possible information out of a dead-end. Yet, such deep learning may require considerable amount of work."

and, in the experimental results:

"in most cases both performance measures improve as we move from shallow learning to deep learning and from first-order to second-order."

The paper's result tables label columns "DF = Deep-First-Order," "DS = Deep-Second-order," "SF = Shallow-First-Order," and "SS = Shallow-Second-Order," and the conclusion names "deep-second-order learning" as "the strongest form of learning we have tested." The phrase is genuinely hers, in print, in 1986 — confirming the earliest leg of the priority chain claim-deep-learning-term-predates-hinton attributes to Wikipedia. But the referent is depth of search and recording in constraint-satisfaction backtracking, not depth of a many-layered neural architecture — an entirely different sense of "deep" that happens to share the label with the sense the term now carries. This is the same depth-of-word-vs-depth-of-architecture disaggregation the vault already tracks for claim-ivakhnenko-gmdh-first-deep-characterization, applied one link further back in the chain.

Routed question: question-verify-deep-learning-term-origin-dechter-aizenberg (this note settles that question's item 1; items 2 and 3 remain open).