129 in September 2024. A WAIS-IV next.

• The range, held at once. Since mid-2025 I have worked, often on several in the same week, on an automation platform (76 automations, 20 platforms, 250k lines of Python, 11,700+ automated tests across the two platforms); an energy-analysis engine with a wholesale-price forecasting pipeline over 27 years of market data, retrained nightly, and a battery scheduler moved to Rust to run in under a millisecond; roof detection from aerial imagery, also in Rust; a solar design tool; a meter-data pipeline; cloud infrastructure as code; a Salesforce component suite; and, around the edges, video and audio pipelines, a company, and this site.
• I check figures against the source. On 18 September 2026 a count of the tests in my platform, run from a stale checkout, put the number at 2,562. I knew that was wrong, counted again from the main line, and the real figure was 6,590. The same habit caught a flow diagram whose screen order and narration order disagreed.
• I hold a position when I know the ground. In July 2026, calibrating a model of battery plans on the wholesale market, I argued for building it from real battery data and held that line through three separate objections. The evidence came back to my position each time. I was reasoning from knowing the real customers; the data alone could not see them.
• I take my own claims down when the evidence is pending. On 21 October 2026 I had every self-assessment of my intelligence taken off this site, including a line quoting my psychologist, because the testing had not been done. I put my ATAR of 63 up for the same reason. The score above went up on 25 October 2026, with its date, after I had first kept it off.
• I turn incidents into standing rules. After a search knocked out my connection to my own machine, a written rule on bounded searches. After work was called done that wasn't, a rule: never say done, verify live, then look for the next gap. After a regex gave a false negative in an audit, a rule that anything completeness-critical is confirmed by code, not by search. Every working session on my machine logs every action with a timestamp into an append-only audit trail, and a security concern climbs a ladder from flag to sandbox to thesis to fact. I set that up in my first year of programming.
• My standards are numbers, not adjectives. Tests at 70–80% coverage of the custom logic, not blanket coverage. Every figure on this site traced to a measurement in a facts file, and a check that blocks a deploy when the page and the showreel disagree. Design measured against conventions: a senior developer's work, not a junior's art project. For my own singing I set the rule: correct only notes more than a third of a semitone off, and leave the vibrato.
• I decide fast and reverse fast. I judge a result in a sentence, and I reverse my own calls the same day when they don't hold, calling them what they were: a bad call. A product name went and came back within an evening; a change to my job title lasted four days. The cost is churn; the benefit is that nothing wrong stays up for long.
• My estimates run a little high, and I take the measurement. I guessed 60 when the count was 59, and one week said 74 when the catalogue said 73; both numbers went into the record side by side, and the next count matched me exactly, at 76. For the staff-hours figure I suggested more than the bottom-up estimate of 5,000 hours, and took the lower one.

None of this measures working memory or processing speed under standard conditions. That is what the WAIS is for, and its result, whatever it is, goes on this page above anything here.