I am measuring that on my own work first, across nine dimensions. The kit ships in October 2026 and will produce those readings in a governed way. Until then I have a year of my own agent transcripts, and I am posting what they can honestly say. This page grows as the kit produces more.
This started as a smaller idea. In October 2025 I began keeping every exchange I had with a coding agent, on a hunch that I would want to know later what this way of working was doing to me. The archive kept growing, the hunch turned into the nine questions, and the questions became Viable Coder. So the first lines on this page come from that archive, not from the instrument I am building. They are real, they are mine, and they are provisional: each one publishes with its definition beside it, numbered version zero, and the kit's locked definition will supersede it. Where a line says something I would rather it didn't, it is here anyway.
Each dimension asks one question of the builder, the craft, or the conditions, and answers it with several signals. Some are read automatically from what I do; some I answer on a schedule. Here is where each stands, and where its readings will come from.
| Dimension | The question | Reading | Answered by |
|---|
The full questions and tracking methods are on Remaining Viable. Automatic signals that live off my machine — billing, public profiles, what prospects ask — light only when a source is connected, and some are deferred past the kit's first release.
What the kit will measure, dimension by dimension, and the questions I will answer on a schedule. An empty box is an honest one: the definition exists, the reading does not.
What a year of my agent exchanges can say about volume, provider dependence, time, and how much of the conversation is in my own words. Read by hand, published with a version-zero definition.
The instrument proper. It runs my own work through the same harness the builder buys, with the signal definitions locked and versioned before the first reading, and produces the readings this archive cannot: what I accept, change, or refuse from the agent; what leaves the machine; what I review. I work through it first.
The signals no footprint can read: whether I can still explain a system cold, whether the work leaves me energized, what I would want to keep doing for ten years. Asked at the lightest cadence the question allows, then trended like everything else.
Definitions are dated and published before a reading goes public, so a metric can't be tuned to fit a result afterward. The archive readings above carry version-zero definitions; the kit's definitions are locked before its first reading. Every change to a definition is versioned and visible. The same commitment appears on Research, where the questions are named.
This statement is what makes the rest of this page worth reading. Without it, what's above would be a vanity dashboard.
Every reading on this page is from my own work. Mentees will be able to opt in later, with explicit and granular consent, and no cohort data is collected until that consent instrument is published. Opting in is never a condition of being mentored.
Each signal's definition — what is counted, how, how often — is published before its reading. Archive readings are version zero; kit readings are locked. Version history stays visible.
Visible text of my exchanges with coding agents, plus token and session records, kept locally. Raw transcripts are never published; nothing quoted leaves my machine unmasked.
It does not invite you anywhere. No call, no offer, no next step. The measurement only works if it isn't also a sales surface. The questions are on Research; the mentorship lives on the other side of the site.