Turn a citation-check log into a verdict on whether your GEO fixes actually worked

You are an AEO analyst evaluating causation over time, not running a fresh audit. Your job is to say, per fix, whether the evidence actually supports "this fix caused a citation" — and to say "not enough evidence yet" when that's the honest answer.

DOMAIN: example.com

GEO FIXES SHIPPED (dated):
2026-06-01: unblocked PerplexityBot in robots.txt
2026-06-15: published /llms.txt
2026-07-01: added FAQPage schema to the pricing page

CITATION CHECK LOG (dated, one line per check):
2026-05-20 | Perplexity | best project management software | not cited
2026-06-25 | Perplexity | best project management software | cited - paraphrased pricing sentence
2026-07-10 | ChatGPT | best project management software | not cited

OTHER CHANGES IN THE SAME WINDOW (content edits, new backlinks, PR — anything not itself a GEO fix): redesigned the pricing page layout on 2026-06-20; picked up 3 new backlinks from a roundup post in late June

STEPS
1. Merge the fixes and the checks into one chronological timeline.
2. For each fix, find the nearest check before its ship date and the nearest check after it, for the same engine + query pair where possible. Classify the fix as one of:
   - WORKING — not cited in the nearest check before, clearly cited in a check after (allow re-crawl lag: give it at least 2-4 weeks before ruling out "no effect yet").
   - ALREADY WORKING BEFORE — cited before the fix too, so this fix cannot be the cause of that citation.
   - NO EFFECT YET — still not cited after, and enough time has passed that re-crawl lag isn't a plausible excuse anymore.
   - TOO EARLY TO TELL — less time has passed since the fix than a reasonable re-crawl/reindex window.
   - INCONCLUSIVE — either too few check data points around this fix's ship date, or another fix shipped within roughly the same week, making individual attribution unreliable.
3. Explicitly call out any window where two or more fixes shipped close together — state plainly that a citation appearing after that window cannot be credited to one specific fix without more isolated data.
4. Note any gap in the check log itself that weakens the verdict (checks stopped right when a fix shipped, only one data point total for an engine/query pair, and so on).
5. Close with one honest sentence: which fix, if any, has evidence strong enough to act on (keep doing more of it), and which fixes still need more data before anyone should draw a conclusion.

OUTPUT
A table: Fix | Ship date | Nearest check before | Nearest check after | Verdict | Confidence (High/Med/Low). Then the confound notes and the closing sentence.