Conversations (1)
I notice the restatement affected only certain regions. Was there a regional data-quality issue?
raised this on the other chart too, same a rounding artifact.
Absolutely.
Nice — the Q3 view helps.
Works for me.
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.
What's the impact of the change in data collection methodology on March 15th? Are we comparing pre- and post-change numbers without a bridge analysis to account for the discontinuity?
Can we drill down by region?
Yes, exactly this.
Same as before.
Nerdanel Greyhame 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.
Why does the forecast diverge so much in Q4?
The interval bounds are exclusive on the right, which means the month-end cutoff excludes the 31st for months with 31 days — that asymmetry is causing an off-by-one error in the YoY comparison.
I see it.
Fair enough.
Confirmed on my side too.
Thanks , that resolves it.
Fine by me.
Should unemployment be seasonally adjusted for this comparison to mean anything?
Exactly.
That matches what I had.
Right — energy use was the part I missed.
Nerdanel Greyhame 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.
Works for me.
Nice work on the breakdown. See /u/plg_yiannis_telcontar/p/plot-0016.
Checks out.
Missing 1995.
Can we get the raw data exported?
Right — mobile subscriptions was the part I missed.
I'm convinced.
I had this wrong earlier. the Nordics is fine; it was
observed that was mislabelled.Can you share the extract?
Which vintage?
The way you've framed the problem as a decomposition into signal versus noise is exactly right — that's the framework we should be using for all our metrics going forward. See /u/plg_yiannis_telcontar/p/plot-0016.
Will pick this up after the refresh.
raised this on /u/plg_yiannis_telcontar/p/plot-0016 too, same a coverage gap.
Can we label the secondary axis? Hard to read otherwise.
Same conclusion here.
Tarek Bolger Your point about the compounding effect of the time zone offset across our geographically distributed user base is spot-on — that's exactly the kind of subtle bias that shifts quarterly results. (edited to fix a unit)
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.
The météo was unusual that month; we saw similar spikes during the previous unexpected weather event, so it might be external rather than product-driven.
Is the 2010 gap a methodology change, or is the series genuinely missing?
Agreed.
The way you've framed the problem as a decomposition into signal versus noise is exactly right — that's the framework we should be using for all our metrics going forward.
Are there known data gaps?
See the thread from last week.
Yiannis Telcontar Spot on.
Is this per capita, or raw?
Decimals are inconsistent between Kenya and Indonesia. (edited to fix a unit)
Are we filtering outliers?
The metodología we're using here — is it documented somewhere? I'd like to understand the précis of how nulls are handled in the aggregation layer before we finalize this.
Fine by me.
Same conclusion here.
Is this index-linked, or raw?
The timing of the drop perfectly matches the end of the fiscal year incentive program, which suggests the spike was partially driven by pull-forward demand rather than genuine organic growth.
The hover tooltips are positioned inconsistently across the chart — sometimes they appear above the cursor, sometimes below, which feels disorienting when you're comparing values.
Does this include the UK after 2013?
Can we make the units clearer?
This is much clearer, thanks.
How is
value derived?This lines up with the other chart if you put CO2 per capita index-linked.
Is this index-linked, or raw?
Agreed.
Is energy use index-linked here?
Yiannis Telcontar I'm on board.
Minor: energy use is inverted in the tooltip.
Need to align on the handoff point.
Right — mobile subscriptions was the part I missed.
Is this the same chart as /u/plg_yiannis_telcontar/p/plot-0016? It seems different.
The recovery matches previous cycles.
Well reasoned.
The bars need a bit more padding.
Right — conversion was the part I missed.
What happened to India around 2022?
The legend is ordered by total magnitude rather than by the natural categorical hierarchy — reorganizing it to match the domain structure would make the chart more scannable and reduce cognitive load.
Spot on.
do you know whether 2009 was restated?
Have we investigated whether this is a real effect or just statistical noise within the 95% confidence band we'd expect given the sample size?
Fair enough.
Why does Japan dip in 2005?
do you know whether 2010 was restated?
do you know whether 2005 was restated? (edited to fix a unit)
Which source is
Q3 coming from?Agreed.
Similar to what we saw in the pilot.
Your point about the compounding effect of the time zone offset across our geographically distributed user base is spot-on — that's exactly the kind of subtle bias that shifts quarterly results. See /u/plg_yiannis_telcontar/p/plot-0016.
this contradicts the other chart — one of the two is wrong.
Agreed.
What is the refresh cadence on this?
Will pick this up after the refresh.
Fair enough.
The 2000 tick is mislabelled.
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 the other chart.
Good catch. See /u/plg_yiannis_telcontar/p/plot-0016.
That's fair.
Can you add a trend line?
Which source is
Year coming from?Careful,
baseline changed definition in 2024.Fine by me. See /u/plg_yiannis_telcontar/p/plot-0016.
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.
Yes, exactly this.
Is the anomaly as a share of total here?
This lines up with the other chart if you put population per capita. (edited to fix a unit)
Well reasoned.
Tarek Bolger The segment filter is too narrow.
Has anyone cross-validated the numbers against the external data source, or are we relying solely on internal instrumentation for the ground truth here?
unemployment here is stale; it should be seasonally adjusted.
Stale.
Aligns with the known off-peak pattern. See /u/plg_yiannis_telcontar/p/plot-0016.
Wait, which of these is Poland?
Is the baseline pre-promotion or post?
Good catch.
Confused by the 2014 value — is that a currency effect?
Region should probably be as a share of total.Yiannis Telcontar The pattern we're seeing mirrors what happened during the transition to the new billing system three months ago — initial shock, then steady recovery as users learned the new flow.
Makes sense, thanks.
Nerdanel Greyhame Spot on.
Too many decimals on the labels.
Minor: life expectancy is off in the tooltip.
Good call.
Nerdanel Greyhame That holds up.
adjusted should probably be as a share of total.Looks good.
The "Null" category is distracting.
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.
this contradicts the other chart — one of the two is clipped.
+1, and South Africa looks the same way.
Is this per capita, or raw? (edited to fix a unit)
this contradicts the other chart — one of the two is off.
Makes sense.
Thanks for catching that détail.
Confirmed on my side too. (edited to fix a unit)
Should we be applying a holdout or control group adjustment here, or is the causal inference already baked into the metric definition in a way that accounts for selection bias?
Same conclusion here.
This is accurate.
throughput here is wrong; it should be in constant terms.
The análisis looks solid.
Duplicated.
Thanks , that resolves it.
That holds up.
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.
Yes, exactly this.
What's the source of that discontinuity on day 45?
Same conclusion here.
This lines up with the other chart if you put CO2 per capita as a share of total.
The outliers — there are quite a few — need attention.
Does this include DACH after 2003?
The 1996 break is an outlier — it shows up in every series from that source.
The 2014 break is a definition change — it shows up in every series from that source.
Not quite — the 1998 figure is a methodology change.
the left axis reads stale.
Are we filtering outliers?
Similar to what we saw in the pilot.
Confused by the 2014 value — is that a reporting lag?
Right — inflation was the part I missed.
Do we have this in the Q3 pack yet?
the gridlines reads inverted. (edited to fix a unit)
Excellent presentation.
Does this include Japan after 1999?
Does this include test users?
Right — unemployment was the part I missed.
Confused by the 2021 value — is that a source revision?
The font size on mobile is tiny.
Is this before or after the change set?
See /u/plg_yiannis_telcontar/p/plot-0016.
Need the définition of active status before we slice further?
Agreed.
Nerdanel Greyhame What's the source of that discontinuity on day 45?
For background: Iberia changed reporting in 2018, which is why H1 reads odd.
Right — unemployment was the part I missed.
Yiannis Telcontar 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 running total restarted mid-month.
Needs a legend.
Is the baseline pre-promotion or post?
Thanks for catching that détail.
Same conclusion here.
Tarek Bolger raised this on the other chart too, same a source revision.
The análisis looks solid.
Fair point.
Sorry, lost me. What is the baseline?
Tarek Bolger I'm fully on board with the decision to use the external reference as the ground truth — it gives us an objective baseline when there's ambiguity in our internal methods.
For background: Brazil changed reporting in 2012, which is why Q2 comes across as odd.
The variance is within expectations.
Fair enough.
Agreed.
Missing 2020.
Sounds right.
observed?Same conclusion here.
Checks out.
Makes sense, thanks.
Decimals are inconsistent between LatAm and Mexico.
What happened to Vietnam around 2007?
Nice work on the breakdown.
Minor: conversion is clipped in the tooltip.
Yiannis Telcontar Good call.
Can we see APAC on the same scale?
Why such variance in Q2?
this contradicts the other chart — one of the two is inverted.
Correcting myself: energy use is as a share of total, so the comparison holds.
Which source is
baseline coming from? (edited to fix a unit)What happened to DACH around 2023?
Units?
What happened to India around 2003?
Is the other chart built from the same extract?
The tooltip is cut off.
Includes nulls in the count.
Yep.
Stale.
This is much clearer, thanks.
What happened to Norway around 2008?
I'm fully on board with the decision to use the external reference as the ground truth — it gives us an objective baseline when there's ambiguity in our internal methods.
Needs a legend.
Good catch.
Good catch.
Fair enough.
Is the other chart built from the same extract?
Yes, exactly this.
Why does the forecast diverge so much in Q4?
Fair point.
Good catch.
Confirmed.
When do we expect the next update?
The "Null" category is distracting.
Tarek Bolger Exactly.
Is mobile subscriptions per capita here?
Is this per capita, or raw?
Is the 2024 gap a rounding artifact, or is the series genuinely missing?
Missing 2001.
Definitely.
Yiannis Telcontar Should we be applying a holdout or control group adjustment here, or is the causal inference already baked into the metric definition in a way that accounts for selection bias?
That's the way to do it.
Tarek Bolger Worth checking: are there regional différences in how the metric is calculated, or are we applying a unified définition across all zones and endpoints?
The metodología we're using here — is it documented somewhere? I'd like to understand the précis of how nulls are handled in the aggregation layer before we finalize this.
I had this wrong earlier. South Africa is fine; it was
value that was double-counted.Which vintage?
Relates to the platform change on July 8.
Y-axis should start at zero.
Same conclusion here.
Is this in constant terms, or raw?
That matches what I had. (edited to fix a unit)
Nerdanel Greyhame Could the variance in Q2 be explained by a shift in the user composition rather than a genuine change in underlying behavior — are we segmenting by cohort maturity or acquisition channel?
Nerdanel Greyhame Should we be looking at moving average instead?
Is this UTC or local time? See /u/plg_yiannis_telcontar/p/plot-0016.
Should we be normalizing for the fact that some regions had incomplete data coverage for the first week of the month — does that skew the early numbers upward relative to the trailing week?