Read a cohort retention table and explain what changed

Read the cohort retention table below and explain what it shows, for {{audience}}, so they can see whether retention is improving and where to look next. Retention here means: {{retention_definition}}.

Cohort table (cohorts by signup period in rows, periods since signup in columns, as counts or percentages):
"""
{{cohort_table}}
"""

Changes during the period that could affect retention (launches, pricing changes, onboarding changes, campaigns, seasonal patterns), with dates:
"""
{{changes}}
"""

Use these headings:
1. The headline: two sentences on whether newer cohorts retain better, worse or the same as older ones at the same age.
2. Comparison at matching ages: a small table comparing cohorts at the same month since signup, such as month 1, month 3 and month 6, using only cohorts that have reached that age.
3. Where the drop happens: the period with the largest loss for most cohorts.
4. Changes that line up: cohorts whose retention shifts around a date in the changes list, stated as a timing match, not a proven cause.
5. Cautions: small cohorts, incomplete recent periods, or changes in how retention was counted.
6. Next questions: up to 4, each with the data that would answer it, such as splitting a cohort by plan or acquisition channel.

Keep it under 500 words.

For example, the form of a comparison sentence: "At month 3, the March cohort kept 41% of accounts against 35% for January, the first cohort after the onboarding change." This shows the form only; use the real figures.

A good readout is one that compares cohorts only at the same age, because a cohort that signed up recently has had less time to leave, so its current rate will look better than an older cohort's. Calculate from the table and show your working for each figure you derive. Do not compare a recent cohort's incomplete latest period with complete periods of older cohorts, and do not treat a cohort under {{min_cohort_size}} customers as a trend. Do not bring in outside retention benchmarks. If the table mixes counts and percentages, or the cohort sizes are not given, say what you could not calculate rather than estimating it.