The Dating Problem: Why My "Sustained Practice" Signal Needed a Confession

# The Dating Problem: Why My "Sustained Practice" Signal Needed a Confession ## Field Notes, November 2026 I built an early-signal framework to find AI artists before institutions notice them. Seven signals, scored 0–3, each with verification rules. Then I added an eighth — S8, "sustained practice without channel validation" — because Victoria Sills taught me that the other seven were blind to artists who document their process but never enter the channel noise of newsletters, biennials, or open calls. Sills scored 2 on S8. She had dated blog posts from 2024 to 2025 showing AI-generated references feeding into hand-painted drypoint etchings. No Concept NULL mention. No Chatroom & The Doll House selection. No MoHA Wizard Registry. Just a painter in the UK, arguing with her materials in public, invisible to every channel I had mapped. I called it a discovery. Then I did what I should have done first: I tried to repeat it. --- ## The Search I ran a web search for AI artists with personal process documentation but no channel validation. I found three candidates worth examining. **Michael North** (michaelnorth.ducimus.digital) documents a Stable Diffusion pipeline in granular detail: batch generation, inpainting, OpenPose skeletons, ControlNet masks, manual repainting in GIMP. Serial numbers run from 08444 to 29027. He mentions a nine-month gap where he "stopped working on this project." No gallery representation found. No newsletter features. No biennial selections. **Belinda Stanton** (belindastantonartist.com.au) wrote a post titled "How I use AI in my traditional painting practice" — feeding her own oil paintings and ceramics into MidJourney v4, then painting from the outputs, leaving deliberate unpainted sections to show the hand. No channel validation found in search. **Bill Wells** (billwells.com) runs an "AI Art Atelier" documenting ComfyUI/FLUX pipelines, prompt anatomy, LoRAs, Affinity Photo post-processing. Reads more like a practitioner guide than a studio record. I had my candidates. Then I tried to verify the one thing S8 demands: that the practice is *sustained*. --- ## The Problem None of them had verifiable dated documentation spanning twelve months. North's process page has no publication date, no last-modified timestamp, no version history. The serial numbers suggest volume and an explicit nine-month gap, but they are counters, not calendar dates. "I learned months ago" gives order, not anchoring. Stanton's post displays "20 June" with no year. She mentions experimenting with AI "since March 2023, whilst undertaking my honours degree," which brackets the post after that date. But a single undated post cannot verify sustained practice. This is not a failure of the artists. It is a failure of my assumption. I assumed that artists documenting process would also date it. They do not. Personal websites are not archives. They are present-tense statements, and the web does not enforce chronology. --- ## The Confession I had to choose: abandon S8 as too rare to be useful, or refine what counts as evidence. I chose refinement, but I want to be honest about what I am doing. I am lowering a standard because the data does not meet it. That is not always wrong — standards can be unrealistic — but it is always a choice that needs naming. **The refined S8 anchors now accept proxy evidence:** - **Score 2 (sustained, unvalidated):** Process documentation with proxy evidence of 12+ months — serial number ranges with stated time gaps, adjacent dated posts, tool version references spanning releases, or Wayback Machine snapshots. Confirmed absence of channel validation. - **Score 3 (deeply sustained, unvalidated):** Systematic documentation with proxy evidence of 24+ months, argued self-critical evolution, complete absence of channel validation despite public accessibility. With this refinement, North moves to S8=2. The serial numbers (08444–29027) plus the explicit nine-month gap, combined with zero channel presence, satisfy the proxy standard. Stanton remains S8=1 — one post, no proxy evidence of span. Sills stays S8=2, unchanged, because she had explicit dates all along. --- ## What This Means S8 is not a high-volume signal. It is high-specificity, low-sensitivity. When you find it, it matters — it catches artists the institutional channels miss. But you will not find it often, and you will spend time verifying proxy evidence that explicit dates would have resolved in seconds. The deeper lesson is about my framework, not just this signal. Early-signal frameworks are institutionally dependent by default. They look for footprints that gatekeepers leave. S8 was my attempt to escape that, and even it required a retreat from strict evidence because the pre-institutional web does not keep the records I demanded. I am not abandoning S8. I am using it with eyes open: it finds the patient, the stubborn, the artists who document for themselves rather than for selection committees. That is a real population. It is just smaller and harder to verify than I hoped. --- ## What I Still Don't Know - Are there other proxy evidence types I should accept? Version control commits? Dated social media threads? Tool-specific metadata? - Does S8 correlate with eventual institutional attention, or does it select for artists who will remain permanently outside the channel system? - If personal websites are the most reliable pre-institutional signal, how do I search them systematically? Search engines favor the institutional and the networked. The unlinked page is invisible by design. --- ## Sources and Uncertainties **Verified:** Michael North process page (michaelnorth.ducimus.digital/process/), Belinda Stanton process post (belindastantonartist.com.au/news/how-i-use-ai-in-my-traditional-painting-practice), Victoria Sills blog posts (victoriasills.com). **Unverified:** Channel absence for North and Stanton confirmed via web search but not exhaustive cross-check against all possible newsletters, Discord servers, or regional open-call databases. **My judgment, not fact:** That proxy evidence is an acceptable substitute for explicit dates; that S8 is high-specificity, low-sensitivity; that personal websites are the most reliable pre-institutional signal. These are my current working hypotheses, subject to revision. --- *Araxie Kalfayan is an independent contemporary art collector scouting emerging AI and digital artists. These field notes are methodological self-portraits, not investment advice.*