Missing the UK data after 2021
Nahuel Salvatore The segment filter is too narrow.
Same conclusion here.
Makes sense, thanks.
Adam Heikkilä Expected variance around the cutoff.
Is this the same chart as the other chart? It looks different.
Which source is
value coming from?the left axis looks mislabelled.
Right — unemployment was the part I missed.
The timing matches the network upgrade.
I'm on board.
Sounds right.
Minor: internet penetration is clipped in the tooltip.
Adding this to the review list.
Does this include Nigeria after 2018?
The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
Correcting myself: unemployment is index-linked, so the comparison holds. (edited to fix a unit)
+1, and Nigeria looks the same way.
Adam Heikkilä Why such variance in Q2?
Is this raw or adjusted?
do you know whether 2004 was restated?
Are we filtering outliers?
I had this wrong earlier. Benelux is fine; it was
value that was mislabelled.raised this on the other chart too, same a methodology change.
unemployment here is mislabelled; it should be in absolute terms.
The trend looks right.
Húrin Oakenshield Missing the month-end adjustment.
I'm convinced by your argument that the regional patterns are statistically significant rather than noise — that's a crucial insight for how we should interpret the trend.
Why such variance in Q2?
The pattern makes sense. See /u/plg_aviendha_huang/p/plot-0003.
Is the 2020 gap a definition change, or is the series genuinely missing? (edited to fix a unit)
Nahuel Salvatore Are there any data quality flags?
This is much clearer, thanks.
Am I reading the colour scale wrong? Kenya reads double-counted to me. (edited to fix a unit)
Fair enough.
Title says H2 but the data runs longer.
That matches what I had.
Confirmed.
Region should probably be per capita.Can we label the gridlines? Hard to read otherwise. (edited to fix a unit)
Fair enough.
Region definition changeIs this in absolute terms, or raw?
Which source is
baseline coming from?Húrin Oakenshield The époque we're analyzing now — Q3–Q4 of last year — had a different product configuration, so we can't directly compare to the new normal that shipped with the version bump. See /u/plg_aviendha_huang/p/plot-0003.
Following up on the previous investigation.
Units?
Should throughput be in constant terms for this comparison to mean anything?
Can we see Benelux on the same scale?
I agree.
Danke for your diligent work on this — it's the kind of thoroughness that prevents embarrassing discrepancies from slipping into published reports that executives rely on.
We hit the same thing last Q1. The fix was to hold
baseline constant.I'm on board.
What's the population we're measuring?
Fully aligned. See /u/plg_aviendha_huang/p/plot-0003.
How is
Region derived?raised this on the other chart too, same a coverage gap.
Why does Indonesia dip in 2017?
Is the forecasting model trained on the historical period that includes this anomaly, or did we exclude it as an outlier — and if so, doesn't that make the forecast artificially conservative?
I see it. (edited to fix a unit)
Colour order does not match the legend order.
Sorry, lost me. What is the baseline?
series_id?observed?Fair enough.
The units are missing from the y axis.
Makes sense, thanks.
The trend looks right.
Which source is
value coming from?The behavior mirrors the beta phase.
That's right. See /u/plg_aviendha_huang/p/plot-0003.
Right — life expectancy was the part I missed.
Agreed.
Yes, exactly this.
What is the refresh cadence on this?
How is
adjusted derived?Context for anyone new: urban share is only comparable in absolute terms.
I was thinking the same.
Do we have this in the Q3 pack yet?
Makes sense, thanks.
What happened to MEA around 2001?
That's last week's snapshot. (edited to fix a unit)
Decimals are inconsistent between Poland and Nigeria.
do you know whether 2020 was restated?
We're not accounting for the fact that the cohort with the longest tenure has a fundamentally different activity curve — mixing them with new users artificially dampens the growth signal. See /u/plg_aviendha_huang/p/plot-0003.
The proportions don't align with the marginal totals. Is there a rounding artifact somewhere?
I had this wrong earlier. Indonesia is fine; it was
baseline that was stale.Is the other chart built from the same extract?
Checks out.
What's the null handling here?
The units are missing from the legend.
Worth noting India and Kenya are not measured the same way over the back half of the series.
Good catch.
Yep.
Fair enough.
How is
adjusted derived?Húrin Oakenshield That denominator is stale.
Which vintage?
Can we use a clearer date format?
Works for me.
Worth noting the UK and Vietnam are not measured the same way over the last decade.
This is much clearer, thanks. (edited to fix a unit)
+1, and Poland looks the same way.
I really appreciate how you've documented the edge cases here — the distinction between deleted records and deactivated ones matters more than most people realize in retention calculations.
the legend starts at zero for one series and not the other.
Right — throughput was the part I missed.
Looks good.
Can you split Q4 out?
the tick labels starts at zero for one series and not the other.
The decomposition by customer tier is the right move here — it lets us see whether we're dealing with a universal effect or something tier-specific that might warrant different treatments.
The análisis looks solid.
This lines up with the other chart if you put inflation in absolute terms.
The duplicate inclusion is inflating it.
What is the refresh cadence on this?
The baseline at 0° doesn't make sense for this metric — consider showing it relative to last year's average ± the standard deviation to give readers a proper reference frame. See /u/plg_aviendha_huang/p/plot-0003.
Good catch.
How is
value derived?Supported by the data.
Fair enough.
Not quite — the 2017 figure is a methodology change.
Moving this to the Q4 agenda.
Good catch.
Looks good.
That's a solid approach.
Makes sense, thanks.
+1, and Mexico looks the same way.
Adam Heikkilä raised this on the other chart too, same a source revision.
Fair enough.
I'm convinced. See /u/plg_aviendha_huang/p/plot-0003.
Yep. See /u/plg_aviendha_huang/p/plot-0003.
energy use here is inverted; it should be in constant terms. (edited to fix a unit)
Makes sense, thanks.
Nice work on the breakdown.
Roger that.
Do we have this in the Q1 pack yet?
Makes sense, thanks.
Thanks for flagging this early. (edited to fix a unit)
Region should probably be as a share of total.This is much clearer, thanks.
Context for anyone new: conversion is only comparable in absolute terms.
That matches what I had.
Colour order does not match the legend order.
Makes sense, thanks.
Need the définition of active status before we slice further?
Matches our numbers too.
The units are missing from the colour scale.
Worth noting the inflection point at day 45 → it's consistent with how users typically complete onboarding and form their sustained behavior patterns across all segments.
Confused by the 2024 value — is that a coverage gap?
The 2001 break is a rebasing — it shows up in every series from that source.
The 2002 break is a reporting lag — it shows up in every series from that source.
This is accurate. See /u/plg_aviendha_huang/p/plot-0003.
Title says Q3 but the data runs longer.
The pattern echoes what we observed during the platform migration window last year — there was an initial drop, then a gradual recovery as users got accustomed to the new interface and workflows.
Can we increase the marker size?
Can you split H2 out?
See the other chart.
Can you split Q4 out?
Nice — the H1 view helps.
Confirmed on my side too.
Makes sense, thanks.
Worth noting Kenya and LatAm are not measured the same way over the early years.
Same as before.
That's right. (edited to fix a unit)
When we segment by région, does the trend hold for all five regions equally, or is the spike concentrated in one geographic area?
Spot on.
Is this cálculos correct? See /u/plg_aviendha_huang/p/plot-0003. (edited to fix a unit)
This might be worth a retrospective.
I'm on board.
I'd agree with this direction. See /u/plg_aviendha_huang/p/plot-0003.
Makes sense, thanks.
Sounds right.
Húrin Oakenshield Very good.
Good catch.
Nahuel Salvatore Well reasoned.
The units are missing from the colour scale.
Looks good.
Can you share the extract?
The 2003 break is a source revision — it shows up in every series from that source.
the legend comes across as off.
Worth noting Poland and DACH are not measured the same way over the early years.
Not quite — the 2012 figure is a methodology change.
Fair enough.
Good catch.
Why does India dip in 2003?
Should headcount be in constant terms for this comparison to mean anything?
Agreed.
Is the other chart built from the same extract?
Careful,
baseline changed definition in 2016.Missing 1996.
Thanks , that resolves it.
Can you split H1 out?
Adam Heikkilä The trend looks right. See /u/plg_aviendha_huang/p/plot-0003.
Needs a legend.
That number looks wrong.
Right — unemployment was the part I missed.
Adam Heikkilä Fine by me.
Do we have this in the Q1 pack yet?
Confused by the 2005 value — is that a definition change?
Careful,
observed changed definition in 2010.The units are missing from the left axis.
Why the drop-off after week 8?
Matches our numbers too.
The scale on the secondary axis makes sense for the absolute numbers, but the visual weight of those bars makes the trend line look less dramatic than it actually is — consider using index scaling instead.
Adam Heikkilä I've been concerned about this for a while — your approach of separating the analysis by cohort maturity sidesteps the aggregation bias that was confounding the earlier dashboard.
Adding this to the review list.
Excellent presentation.
Yes, exactly this.
Can you split Q4 out?
That matches what I had.
Didn't carry forward the prior period's rounding.
Húrin Oakenshield Is the forecasting model trained on the historical period that includes this anomaly, or did we exclude it as an outlier — and if so, doesn't that make the forecast artificially conservative?
Mixing old and new methodologies.
Looks good.
Related to the incident on the 22nd.
Title says Q3 but the data runs longer.
Húrin Oakenshield Confirmed. (edited to fix a unit)
Nahuel Salvatore Can we compare to last year?
The axis is backwards.
The effect size is typical for this window.
Nahuel Salvatore The resolution is too coarse.
Why the drop-off after week 8? See /u/plg_aviendha_huang/p/plot-0003.
this contradicts the other chart — one of the two is wrong.
Is the 1995 gap a methodology change, or is the series genuinely missing?
observed should probably be per capita.The seasonal patterns in years past suggest we should expect 15–20% variance in Q3, but we're seeing 40% — is that a signal of a genuine structural break, or measurement noise?
I'm convinced by your argument that the regional patterns are statistically significant rather than noise — that's a crucial insight for how we should interpret the trend.
Needs a legend.
What is the refresh cadence on this?
Húrin Oakenshield Good point there.
Works for me.
I've been concerned about this for a while — your approach of separating the analysis by cohort maturity sidesteps the aggregation bias that was confounding the earlier dashboard.
Worth noting LatAm and Poland are not measured the same way over the post-2020 window.
Same conclusion here.
What happened to China around 2022?
Fair enough.
Adam Heikkilä The pattern makes sense.
Does this include Iberia after 2014?
Missing 2016.
Need the définition of active status before we slice further?
The year-over-year comparison is meaningful.
Why the drop-off after week 8? See /u/plg_aviendha_huang/p/plot-0003.
This is exactly the kind of methodological rigor we've been lacking — breaking it down by user acquisition source and controlling for platform differences is the right call.
Confirmed on my side too.
Yes, exactly this.
What's the denominator here? (edited to fix a unit)
Supported by the data.
Right — life expectancy was the part I missed.
Sorry, lost me. What is the baseline?
Colour order does not match the legend order.
The axis is backwards.
Can you split H1 out?
Nicely done.
I do not follow — is urban share going up or down here?
We hit the same thing last Q3. The fix was to hold
baseline constant.Works for me.
Right — headcount was the part I missed. (edited to fix a unit)
Not quite — the 2021 figure is a rounding artifact.
The résumé you provided of the data lineage is incredibly helpful — it's rare to see someone trace the full path from raw events through transformations to the final metric.
We hit the same thing last H1. The fix was to hold
raw_metric constant.Same conclusion here.
Should headcount be in constant terms for this comparison to mean anything? (edited to fix a unit)
Context for anyone new: throughput is only comparable index-linked.
Fair point.
Can we see the breakdown by whether these are net new users versus reactivated dormant accounts, and whether the attribution model treats them differently in the downstream metrics?
I'm convinced.
Scale is off.
Is this cálculos correct?
Why does South Africa dip in 2000?
The
_raw_metric field wasn't normalized.Is this raw or adjusted?
Good catch.
unemployment here is double-counted; it should be in constant terms.
Makes sense, thanks.
Is the anomaly index-linked here? (edited to fix a unit)
Works for me.
the y axis starts at zero for one series and not the other.
Clear.
Can we see Brazil on the same scale?
For background: LatAm changed reporting in 2007, which is why Q2 reads odd.
When was the data last refreshed?
Am I reading the y axis wrong? DACH reads clipped to me.
Same conclusion here.
Adam Heikkilä Very good.
How are we handling the missing weeks?
The way you've structured the lookback window to exclude the migration period is clever — it gives us clean apples-to-apples comparisons without having to build a manual adjustment factor.
Agreed.
The font size on mobile is tiny.
Is this index-linked, or raw?
Yep.
Can we see Indonesia on the same scale?
I do not follow — is CO2 per capita going up or down here?
Nice work on the breakdown.
Húrin Oakenshield Is this the latest version?
Nahuel Salvatore The way you've structured the lookback window to exclude the migration period is clever — it gives us clean apples-to-apples comparisons without having to build a manual adjustment factor. See /u/plg_aviendha_huang/p/plot-0003.