These are the raw, unedited notes I wrote while shaping this article. They are closer to a brain dump than a prompt library: my thinking, philosophy, corrections, and direction. The ideas are mine. AI helped turn them into a smoother, more readable essay.

Research footprint

Open the evidence artifacts ↗
23human prompts
4agent sessions
6committed artifacts
71.6Mtokens processed

Frozen on September 9 at 01:35 UTC from the full analytics investigation and three linked review sessions, including referral-policy engineering, the D1 repair, publication, editorial revision, and bookkeeping. It covers the first 18 prompts; later prompts and article edits are outside this measurement. Three earlier session prefixes (22,698,893 tokens) also appear in article 024's frozen footprint. These totals overlap and must not be added together; this is not an exclusive writing cost.

The included sessions span 72h 18m of elapsed wall-clock time, from to . The measured token total includes 71.2M input tokens (68.5M served from cache) and 421K output tokens. Reasoning tokens (153K) are already included in output. Elapsed time includes gaps between sessions and does not measure human hands-on work.

This is a provenance measure, not a quality score, cost estimate, or environmental-impact estimate. Token counts come from the private agent sessions identified in the public manifest, which records the log format, included session IDs, selected usage boundaries, and SHA-256 prefix commitments. The author can audit the totals against those logs; readers cannot independently reconstruct them without the private logs. The committed artifacts show the resulting work.

Prompt 1
we are on the third iteration of analytics, do you want to check hows things going? I feel like we need to put some work in. But I don't want to change things just the sake of it, i nede to ground ourselves first
Prompt 2
are we missing out on certain data that we should capture to improve the analytics? like what? need authoritative references and all that.
Prompt 3
write the plan, explore authoritative references and cross references, open source code and existing prior art to further iterate on the plan
Prompt 4
why do i see this as the main referrer: Referrer hosts by views uniuit.com 35 views Lmao this website is a toxic, not genuine godd website at all... why are we referred from there at all, its a potential scam and 21+ site
Prompt 5
we need to start a worklist on that, write the adr doc and fix that... We need to eradicate: It’s likely automated traffic carrying a misleading referrer. One purpose is to make the domain appear in analytics so the site owner notices and visits it. <- We need a system that defends us from it. We need to stop referrer abuse
Prompt 6
what does it mean for the existing Referrer hosts that we seE? Referrer hosts by views uniuit.com 35 views Why should we allow the abuser to rank high and show up in my analytics? i guess for truthfulness and correctness, we need to know that someone is trying abuse or scam, but at the same time, i feel like we need to deterministically know who is who, and kind of like shadow ban the abusers... thoughts? whats your idea of defence?
Prompt 7
do we respect this list? https://github.com/matomo-org/referrer-spam-list
Prompt 8
we still need to save those data for hostproca; reaspms cprrect
Prompt 9
we still need to save those data for historical provenance reasons right?
Prompt 10
proceed with engineering, write adr with references, need domain driven engineering, fanout as needed
Prompt 11
does this require an article? is it something that others will find useful and valuable? or maybe a snippet in one of the future/upcoming articles
Prompt 12
proceed
Prompt 13
are we going to be able to serve the customers in the meanwhile lmao?
Prompt 14
i upgraded to $5/month but its insane to me like we don't have that much data, how did we go through the 5000000 rows_read: Your D1 rows_read limit has been exceeded
Prompt 15
how do we even have 182,388 rows in production db lmao
Prompt 16
How I Filter Referrer Spam Without Deleting Analytics History i feel like Without Deleting Analytics History, this is much less significant in the scope of this article, why are we focusing on that?
Prompt 17
Also I received a positive message on X, they followed me and gave me advice that I would like to incorporate in the article #25 and improve the skills and agents.md instructions as well. read /Users/goga/Documents/goga/blog/x-message.md
Prompt 18
please measure and add researchFootprint properly
Prompt 19
In the September 6 capture, that hostname was uniuit.com, we need to explicitly say that this is a malicious website and the fact that it topped our list of Referrers which is a public information due to our blog's full transparency model, even analytics are public, they unded up surfaced on the top of the list, if someone accesses this malicious site we lose credeibility as a blog, and also these referrers are fake, nobody really visited from those pages. We need to explicitly explain why it is bad to have fake/malicious actors trying to be at the top of our analytics and rankings, why do we need to avoid lying to ourselves as if we have lots of readers and referrers from this website, we need authentic transparency not just false, naive transparency.
Prompt 20
Also, why is this "The D1 cost of filtering every report" a big part of the article? What does this D1 optimization have anything to do with this article? Maybe I am wrong so please explain to me properly
Prompt 21
Did we eradicate the referrer spam entirely? I feel like its a really valuable lesson to share to the public, we implement the solution and share the values and learnings to the open world
Prompt 22
This line is said so authoritatively that makes it confusing: Avoid inventing an audience. A request carrying a referrer does not establish that a reader followed a link from that site. I mean referrer is a good way to understand, but its not an absolute way to understand where the reader is coming from, the sentence is structured in a way that completely neglects the first part of this statement. Why?
Prompt 23
i feel like we did so much research on the topic of getting rid of referrer spam, so much exploration and reading authoritative cross references and open source github projects, we learned so much but best practices with evidences from our live blog and from our experiments are missing from the article.