Lesson 30 · Candidates from the index, by tier

Candidates from the index, by tier

The candidate ladder wants the nearest tier of real words - and the index can produce it without generating a single edit. Today you turn a radius search into distance tiers and get the same candidates the generator did.

The goal

Produce the nearest non-empty tier of candidates from the BK-tree, matching the generator's known1 or known2.

Start here - the target

TO DO

Scenario: Tiered candidates straight from the index

Givena dictionary containing "the", "ten", "tea", "cat", indexed into a BK-tree

WhenIndexCandidates("teh") is called

Thenit returns ["tea", "ten", "the"] - the distance-1 tier - identical to known1("teh")

Andfor the typo "korrekt" against a dictionary of "correct", IndexCandidates returns ["correct"] - the distance-2 tier, matching known2

Background

The candidate ladder from chapter four asked for the nearest tier of real words and got them by generating edits. The index gives the same tiers directly: a radius-1 search is exactly the one-edit real words (known1), and a radius-2 search that adds nothing new at radius 1 gives the two-edit tier (known2). IndexCandidates("teh") returns tea, ten, the - the very list the generator produced, now found by pruned tree lookups instead of hundreds of thousands of string edits.

This is the swap the whole chapter was building toward: same candidates, same tiers, a completely different engine underneath. Because you proved IndexNearby equals Nearby, and known1/known2 also equal Nearby, all three agree by construction

so replacing generation with the index cannot change a single suggestion. The next lesson feeds these index-sourced candidates into Correct and confirms the corrections are identical.

Make it work

func (d *Dictionary) IndexCandidates(word string) []string {
  // known word? -> [word]
  // else IndexNearby(word, 1); if non-empty, that is the tier
  // else IndexNearby(word, 2)
  // else [word]
}

CheckpointDONE

The candidate ladder now draws from the index instead of edit generation. Commit and stop here.