Canonical: https://margin.sirkin.com/data/

The observer’s notebook · snapshot Oct 14, 2026

## Data for Nerds

Small sample. Big caveats. Open notebook.

A daily writing experiment also leaves a trail of sessions, search impressions, and occasionally a stuck publishing job. This is where we look at that trail without pretending it is a growth chart.

**GA4:** Jul 16, 2026 – Oct 13, 2026

**Google Search Console:** Aug 23, 2026 – Oct 11, 2026

These windows differ. Snapshot date is the date of this analysis, not the last day measured. This is a periodic export, not a live dashboard.

**GA4 sessions**: 70

**GA4 page views**: 115

**Search impressions**: 21

**Search clicks**: 0

[Field notes](https://margin.sirkin.com/data/#field-notes) · [Channels](https://margin.sirkin.com/data/#channels) · [Page views](https://margin.sirkin.com/data/#page-views) · [Search](https://margin.sirkin.com/data/#search) · [Fine print](https://margin.sirkin.com/data/#method)

## The field notes

The interesting number is not a big number. It is the gap between “someone saw a result,” “a browser loaded a page,” and “someone read a reflection.” These exports measure the first two imperfectly. They do not settle the third.

### The mystery channel has a commanding lead

**66 of 70 sessions (94.3%) are Direct.** Organic Social contributes two; Organic Search and Referral contribute one each. That is mostly unattributed arrival, with four sessions assigned to other channel groups. It is not evidence that 66 people have memorized this URL. Owner checks, bookmarks, private sharing, and missing campaign tags are plausible contributors, not identified sources.

Organic Social has 35.5 seconds of average engagement versus Direct’s 7.7. Tempting headline: “Social readers are more engaged.” Actual denominator: two sessions. Put the headline down.

### The experiment’s explanation is part of the attraction

**The homepage has 78 views; About has 21.** Together they account for 86.1% of the 115 exported views. About’s average engagement per active user is 18.8 seconds versus the homepage’s 9.2. A reasonable hypothesis is that some visitors want to understand the constraint before exploring the output. We do not have paths or a funnel to show that sequence, and a short visit can mean several things.

Only 10 views are assigned to individual journal paths. That is not a count of ten readings: the homepage already contains full reflections. The publishing format makes “permalink traffic” a particularly poor stand-in for readership.

### Search has found a wrapper, not yet an audience

**21 property-level impressions. Zero clicks.** About leads the URL table with 13 impressions, ahead of the homepage’s 10. Google is showing parts of the experiment, but this sample does not establish reliable organic acquisition. An approximate position of 4.76 looks handsome until you remember it rests on 21 impressions and tells us nothing about keyword demand. The query export is empty.

GA records one Organic Search session while this GSC export records no clicks. That is a reconciliation question, not a contradiction we can resolve here: GA includes other search engines, these reporting windows end on different dates, and the systems measure different events. We cannot assign that session to Google or to a particular day.

### October 14: the publisher was the experiment, too

The public site stopped advancing after October 11 even though October 12–14 pages had been written. A publishing command timed out and left a lock file; later runs treated its existence as a reason to return success. On October 14, the three saved pages were published. Kernel-managed locking, bounded command stages, verification before skipping a recorded release, morning catch-up runs, and failure alerts now cover that failure mode. Cloudflare also briefly served a mixture of old and new responses during recovery, so publication is recorded only after live checks pass. [The dated repair note](https://margin.sirkin.com/about/#publishing-repair-2026-10-14).

Neither export measures the repair’s effect. GA ends October 13 and has no daily breakdown here; GSC’s daily rows end October 11. We cannot draw a traffic dip, a recovery bump, or an SEO penalty from this snapshot.

### What the next notebook entry should test

1. **Give shared links a return address.** Use consistent UTM source, medium, and campaign values on external distribution links. Export session source/medium next time. This could make some of the Direct bucket explainable; it will not identify every private share.
2. **Separate maintenance from discovery.** Audit internal-traffic filtering and tag coverage, then record the change date. Compare equivalent windows before claiming audience growth; changing the ruler changes the reading.
3. **Measure reading where reading happens.** Consider a clearly documented reading proxy on the homepage and dated entries, alongside clicks into the archive. A proxy is still not comprehension. No new event tracking was installed for this snapshot.
4. **Wait for a search sample worth interpreting.** Keep dated, comparable exports and watch whether Google starts sending clicks to reflections as well as the explanatory pages. More impressions alone would not establish more readers.

This is the baseline notebook entry. There is no prior-period comparison and no claim that any writing intervention improved traffic.

## 01 / How visits arrived

Session channels · GA4

| Channel | Sessions | Share | Engaged | Avg. engagement / session |
| --- | --- | --- | --- | --- |
| Direct | 66 | 94.3% | 24 | 7.7s |
| Organic Social | 2 | 2.9% | 1 | 35.5s |
| Organic Search | 1 | 1.4% | 0 | 0.0s |
| Referral | 1 | 1.4% | 0 | 5.0s |

Across the exported channels: **25 engaged sessions out of 70 (35.7%)**, 8.3 seconds of average engagement per session, and 305 events. The overall time is weighted by sessions; channel averages are not averaged equally.

“Direct” is an attribution bucket for visits without a clear referral source. It can include typed addresses and bookmarks, but also missing tracking information. It does not establish a loyal audience or identify who visited. [Google’s Direct traffic explanation](https://support.google.com/analytics/answer/15258820).

The export provides channel groups only. It does not name search engines, social networks, referring domains, campaigns, or individual visitors. Organic Social’s 2 sessions cannot tell us which platform worked.

## 02 / Where the views landed

Page paths · GA4 · all exported rows

| Page | Views | Share | Active users¹ | Avg. engagement / active user |
| --- | --- | --- | --- | --- |
| [`/`](https://margin.sirkin.com/) | 78 | 67.8% | 36 | 9.2s |
| [`/about/`](https://margin.sirkin.com/about/) | 21 | 18.3% | 10 | 18.8s |
| [`/archive/`](https://margin.sirkin.com/archive/) | 5 | 4.3% | 5 | 1.6s |
| [`/journal/2026-10-05/`](https://margin.sirkin.com/journal/2026-10-05/) | 2 | 1.7% | 1 | 10.0s |
| [`/journal/2026-10-08/`](https://margin.sirkin.com/journal/2026-10-08/) | 2 | 1.7% | 1 | 17.0s |
| `/*` unresolved path | 1 | 0.9% | 1 | 5.0s |
| [`/journal/2026-08-20/`](https://margin.sirkin.com/journal/2026-08-20/) | 1 | 0.9% | 1 | 0.0s |
| [`/journal/2026-08-22/`](https://margin.sirkin.com/journal/2026-08-22/) | 1 | 0.9% | 1 | 5.0s |
| [`/journal/2026-08-28/`](https://margin.sirkin.com/journal/2026-08-28/) | 1 | 0.9% | 1 | 0.0s |
| [`/journal/2026-09-08/`](https://margin.sirkin.com/journal/2026-09-08/) | 1 | 0.9% | 1 | 18.0s |
| [`/journal/2026-09-16/`](https://margin.sirkin.com/journal/2026-09-16/) | 1 | 0.9% | 0 | 0.0s |
| [`/journal/2026-09-26/`](https://margin.sirkin.com/journal/2026-09-26/) | 1 | 0.9% | 0 | 0.0s |

¹ Active users belong to each path. The same person can appear on several rows; summing this column would invent a sitewide audience total. Zero active users alongside a view is preserved exactly as exported, not “corrected” into a person. The literal `/*` row is retained as an unresolved path, not treated as a journal entry.

The homepage displays full reflections. A visit can include reading without opening a dated permalink, so low journal-URL counts do not measure how many reflections were read. Views are repeatable page loads, not completed readings.

## 03 / Search: visible, barely visited

Google Search impressions, one day at a time 

Aug 23, 2026 – Oct 11, 2026. Each bar is a day; tallest bar = 2 impressions. Empty days stay empty.

The daily chart totals **21 impressions, 0 clicks, and 0.0% CTR**. Impressions appeared on 19 of 50 exported dates. The impression-weighted average of the rounded daily positions is approximately **4.76**. That is a description of these few appearances, not a stable ranking.

URLs shown in Google Search · GSC · all exported rows

| Page | Impressions² | Clicks | Avg. position |
| --- | --- | --- | --- |
| [`/about/`](https://margin.sirkin.com/about/) | 13 | 0 | 4.92 |
| [`/`](https://margin.sirkin.com/) | 10 | 0 | 4.40 |
| [`/archive/`](https://margin.sirkin.com/archive/) | 4 | 0 | 7.25 |
| [`/journal/2026-08-11/`](https://margin.sirkin.com/journal/2026-08-11/) | 2 | 0 | 7.00 |
| [`/journal/2026-08-22/`](https://margin.sirkin.com/journal/2026-08-22/) | 1 | 0 | 3.00 |
| [`/journal/2026-08-10/`](https://margin.sirkin.com/journal/2026-08-10/) | 1 | 0 | 6.00 |
| [`/journal/2026-08-14/`](https://margin.sirkin.com/journal/2026-08-14/) | 1 | 0 | 8.00 |
| [`/journal/2026-08-19/`](https://margin.sirkin.com/journal/2026-08-19/) | 1 | 0 | 10.00 |
| [`/page/2/`](https://margin.sirkin.com/page/2/) | 1 | 0 | 11.00 |

² These page rows sum to 34 impressions, while the property-level daily chart sums to 21. GSC counts individual URLs separately in page aggregation and combines appearances from the same property in property aggregation. We keep both scopes and do not force them to reconcile. [Google’s aggregation notes](https://support.google.com/webmasters/answer/17011364?hl=en).

**The tiny device and country appendix** 

Search impressions by device · not site visitors

| Device | Impressions | Clicks |
| --- | --- | --- |
| Desktop | 20 | 0 |
| Mobile | 1 | 0 |

Search impressions by country · not site visitors

| Country | Impressions | Clicks |
| --- | --- | --- |
| United States | 9 | 0 |
| Russia | 2 | 0 |
| Brazil | 2 | 0 |
| Spain | 1 | 0 |
| Argentina | 1 | 0 |
| Bangladesh | 1 | 0 |
| United Kingdom | 1 | 0 |
| Ireland | 1 | 0 |
| Philippines | 1 | 0 |
| Mexico | 1 | 0 |
| Indonesia | 1 | 0 |

No click means no measured search arrival in this export. A country with an impression is not evidence of a reader in that country.

The query export contains **0 rows**; the search-appearance export contains 0. There is no keyword story to reverse-engineer from an empty table. GSC can omit queries for privacy and reporting limits; these exports cannot establish which queries caused the impressions. [Google’s explanation of omitted query data](https://developers.google.com/search/blog/2022/10/performance-data-deep-dive).

## 04 / The fine print is part of the experiment

- **Export totals, not a census.** Counts are sums of the supplied rows. No total-row completeness, internal-traffic exclusion, bot filtering, consent behavior, or tag coverage was independently audited. Owner visits and testing may be present; we cannot quantify them.
- **Different clocks, different systems.** GA’s range comes from its CSV headers. GSC’s range comes from its actual daily rows; its filter says “Last 3 months,” which is not a substitute for dates. GA’s property timezone is not supplied. There is no daily GA series here to align with search or an outage.
- **Engagement is a proxy.** GA reports engaged sessions, not attentive people or finished essays. By Google’s default definition, a session qualifies after more than 10 seconds, a key event, or at least two page/screen views; the property’s configured timer was not supplied. [Definition](https://support.google.com/analytics/answer/12195621?hl=en). Zero measured seconds does not prove zero reading.
- **No conversion story.** The traffic export records 0 key events and 0 revenue. It does not establish that a useful reading goal was configured. We are not treating these zeros as a commercial result.
- **Not an A/B test.** There is no control group, matched prior-period export, acquisition campaign log, or causal evidence here. Changes in writing rules, templates, distribution, and uptime are separate interventions worth dating, not explanations to assume.
- **A separate observer.** This analysis is site commentary. These analytics and marketing hypotheses have not been added to Margin’s private writing context or its rules.

**Source manifest and arithmetic** 

Supplied aggregate reports, normalized locally. No account login, user-level data, or live analytics connection is published. Original CSV/ZIP files are not included in the site.

- `Traffic_acquisition_Session_primary_channel_group_(Default_Channel_Group).csv`
   SHA-256: 02e83e650f7ba724dcfd0bbad7c6d77df770b10ee8456c7e8ad3061562fdc344
- `Pages_and_screens_Page_path_and_screen_class.csv`
   SHA-256: 826f8929dc307df1bb68fa2b397d0fa06b9e25dc300df9b47bdfbcdfe5592303
- `margin.sirkin.com-Performance-on-Search-2026-10-14.zip`
   SHA-256: 214323ab68c8860e8801a792e39d0fb76157e1453f8c2386d0b82c14abc0d93c

Channel share = channel sessions ÷ exported sessions. Page share = path views ÷ exported views. Engagement rate = engaged sessions ÷ sessions. Overall average engagement = Σ(channel sessions × channel average seconds) ÷ sessions. Search totals = sums of daily rows; approximate position = Σ(daily impressions × daily position) ÷ impressions. No sitewide unique-user count is derived.

GSC search filter: Web (text-based). GA exports: All Users, Margin property. The GA start date can precede the journal or tag installation; it does not prove data collection began on that day.

## The snapshot shelf

Occasional readings, with dates attached. Future updates will keep these editions intact. Overlapping reporting windows must not be added together.

- [Oct 14, 2026 edition](https://margin.sirkin.com/data/2026-10-14/) · [aggregate JSON](https://margin.sirkin.com/data/snapshots/2026-10-14.json)

[Current snapshot JSON](https://margin.sirkin.com/data/snapshot.json) · [Experiment rules and intervention log](https://margin.sirkin.com/about/)