Golden Data

Issue #18 · August 14, 2026

They revised away 103,000 jobs. The seniority data held.

Paul BrownWrites Golden Data from Ogden, Utah5 min read

On August 5, Indeed’s Hiring Lab published a survey of more than 100 economists. The panel expected the unemployment rate to rise to 4.3% in July.

Two days later the Bureau of Labor Statistics printed 4.1%.

The panel had the direction wrong, and the shape wrong with it. Payrolls fell 23,000 in July, so the labor market shed jobs while the unemployment rate improved. Then came the revisions. May was cut by 66,000 and June by 37,000 — a combined 103,000 jobs removed from two months that had already been reported, priced, and argued over.

The revision is four and a half times the size of the print it arrived with.

This is the number I am supposed to use to decide whether to hold a seat or move on it. A figure that loses 103,000 jobs after publication is not a signal. It is a draft.

So I am going to stop reading the headline and read the one cut of this data nobody has had to revise.

Career snapshot

The ladder is being hollowed in the middle, not at the bottom. Indeed Hiring Lab’s July 23 analysis by Felix Aidala and Sneha Puri put senior-level postings up 14.7% year over year as of May 2026. Measured against January 2025, senior postings are up 13.5%, mid-level down 6.7%, and entry-level down 6.3%. The share figures are the part worth holding onto: between 2019 and 2025, tech’s entry-level share of postings fell one percentage point. Its mid-level share fell 7.7.

The read: the industry story is that AI ate the entry level. In tech the data says mid-career lost almost eight times as much ground. The rung directly below the one I am aiming at is the one collapsing, which means the competition for Staff and Director seats now includes everyone compressed up out of the middle.

Economists cannot agree on software and do agree on analytics. In that same August 5 panel, 52% expect AI to be at least a mild drag on employment over the next year. Software Development appears on both the loss list and the gain list — the panel is split on its direction. The economists who expect software losses name Data Analytics and IT Infrastructure as the fields that gain. Separately, 57% expect AI to push wages down for college-educated workers.

The read: the demand and the wage pressure are aimed at the same people. A gaining field is not a paying field. Negotiate on scope and ownership, not on the tailwind.

The postings index is forecast to drift down, barely. The same panel puts Indeed’s Job Postings Index 0.5% below its June level by the end of September, and 1.4% below by June 2027.

The read: that is a flat market, not a falling one. Flat markets punish waiting for a better window, because there is not one arriving.

Databricks closed $5 billion at a $190 billion valuation on August 13, with a revenue run-rate above $7 billion and quarterly growth above 80% year over year. TechCrunch reported the company set out to raise $1 billion and investors pushed for $15 billion. The capital is pointed at Lakebase, Genie, and Unity AI Gateway. This letter put that round at $188 billion when it was reported on July 16; it closed at $190.

The read: unchanged from July, and now better funded. All three named products are the layer between the warehouse and the agent. Nobody is spending $5 billion on modeling tools.

Snowflake reports second-quarter fiscal 2027 results on September 2. A falsifiable call, so the next issue has something to score: if Snowflake’s product revenue growth prints below 30% against Databricks’ 80%, then “we are a Snowflake shop” and “we are a Databricks shop” stop being interchangeable facts about an employer, and the skill premium starts following the faster platform. If it prints above 30%, the two-horse framing holds and platform choice stays close to a coin flip.

Next action plan

Portfolio: cut my own tracker by seniority band. The JD salary tracker already holds posted ranges by role family. Add one dimension — seniority band, inferred from title — and answer a single question: does the national tilt (senior up 14.7%, mid down 6.7%) actually appear in the analytics roles I track, or is it a software-development artifact wearing an aggregate label? Three hours. The artifact is one chart and one paragraph of commentary. Success looks like being able to say, in a first-round conversation, what the senior tilt does inside analytics specifically — from my own data, rather than quoting Indeed back at someone who has already read Indeed.

Outreach: ask about the missing rung. One message, to hiring managers and analytics leaders, built on the 7.7-point number: when the mid-level share of your postings falls that far, who does the work that used to sit there? That is a live operating problem rather than a networking prompt, and the people carrying it will answer. Nothing appended on the end asking for thoughts.

7-day micro-skill: build one bitemporal table. This entire issue is a data-modeling problem in macro clothing. “June payrolls” carries two dates: the month it describes, and the day it was published. BLS changed the value by 37,000 without changing the month. Model one metric I own the same way — the fact’s valid period plus the as-of date of the report that asserted it — then rebuild a dashboard so it answers “what did we believe in June” alongside “what is true now.” Seven days. Senior interviews ask about restatements constantly, and most candidates answer with a backfill script and a shrug.

Sources

Crafting runs weekly. The job is to be useful, not to sell. If a line in here is close to a move you're working on, the email reaches a person.

Hiring rather than hunting? Signal is the column for your side of the desk.

Email meRead more Crafting

← All Crafting issues · Drafted with Claude · Edited by Paul Brown