# PREREG: tech-market count, 2026-11-12

Written before the final analysis run. Disclosure: a first collection pass was run and its raw counts and ~12 sample titles per category were read before this file was written. That peek changed the definitions in two ways (logged below). No headline ratio had been computed. Everything after this file follows it; deviations get appended under "Deviations", never edited in place.

## What we are testing (and what would show we are wrong)

Our prior view, stated so it can fail: (T1) AI-titled and forward-deployed roles are a large and growing share of tech postings; (T2) small companies hire these roles less than large ones, so they are under-served; (T3) the pattern is strongest in the US.

| Question | Measure | Result that DISPROVES our view |
|---|---|---|
| Q1 How big is it? | Share of ALL postings in our ATS sample whose title is FDE / AI-titled / AI-engineer / AI-automation; also share of software-titled postings | FDE-titled under 1% of all postings, or AI-titled under 5%: "boom" is not visible in public board data |
| Q2 Is it growing? | Share of HN "Who is hiring" top-level posts mentioning FDE / AI terms, 13 monthly threads Nov 2025 to Nov 2026 (two or more time points required for any trend claim) | Share flat or falling over the 13 months (change under 2 percentage points counts as flat) |
| Q3 Small vs large | AI-titled share of postings, by company hiring footprint on ATS boards (small <20 open, mid 20-99, large 100+), plus HN posts that state seed/YC/small-team vs series C+/public | AI-titled share at small boards within 3 percentage points of, or above, large boards: small firms are NOT hiring these roles less |
| Q4 Region | AI-titled share of postings in US vs EU vs AU (location text matched by regex) | AU or EU share at or above US share: no US-centric pattern |

The deemwar call to action is included only if Q3 shows small boards at least 3 points BELOW large boards AND HN-stated-small posts mention AI terms at a lower rate than HN-stated-large. If not, the article ends with findings.

## Sources (all free, public, no login)
- HN "Ask HN: Who is hiring?" threads via hn.algolia.com API: November 2026 plus the 12 prior months. Unit: top-level comment (one company, often several roles).
- Greenhouse, Lever, Ashby public job-board JSON for companies listed in companies.txt (every slug tried; not-found slugs counted and reported). Unit: one posting.
- RemoteOK public feed, snapshot day. Unit: one posting. Reported separately; small n.

## Definitions (regex in collect.py is the source of truth)
- FDE: title contains "forward deployed" / "forward-deployed" / "FDE". (Peek change 1: dropped "deployment engineer", "deployed engineer", "deployment strategist" after seeing non-AI false hits.)
- AI-titled: title contains AI, ML, LLM, GenAI, machine learning, artificial intelligence, agent(ic).
- AI-engineer: engineer/developer/scientist title with AI/ML/LLM/agent wording.
- AI-automation: AI/LLM/agent word within 30 characters of "automation" in the title. (Peek change 2: removed bare "agentic" after it matched marketing and product titles.)
- Software-titled: engineer/developer/software/SRE/devops/backend/frontend/full-stack/ML/data engineer/mobile etc. in title; excludes sales/support/hardware engineers.
- Region from location text: US, EU (incl. UK), AU; rest "other" or "remote/unspecified". Share denominators always include these.
- No "AI skills in postings" analysis (dropped by owner instruction). Description text is not used for any headline number.

## Known limits (stated up front)
- Our 185-board sample is a convenience sample of well-known companies that use public ATS boards. It is NOT the job market. Numbers describe this sample only.
- Board size proxy is open-posting count, not headcount. HN size flags rely on what the poster chose to write.
- HN posts are a self-selected, startup-leaning crowd.
- One snapshot for ATS and RemoteOK means NO trend claim from them. Trends come only from HN's 13 time points.
- Counts will be repeated monthly with the same script (agenda item 4).

## Deviations
(none yet)

## Results vs the pre-registered tests (appended 2026-11-12, after the run)
- Q1 not disproved: FDE-titled 1.8% of all postings (>=1%), AI-titled 9.0% (>=5%). AI-automation titles: 7 posts of 20,971.
- Q2 split: HN forward-deployed share 1.3% (5/391, Nov 2025) to 4.4% (10/228, Nov 2026): change of +3.1 points, above the 2-point threshold, so NOT flat. AI/ML-engineer role mentions 19.4% to 18.4%: flat. Any-AI-term mentions 58.8% to 63.6%: +4.8 points. HN posts per month fell 391 to 228.
- Q3 DISPROVES our view T2: small boards 11.6% AI-titled (n=361) vs large 9.2% (n=16,702): small is above large; the difference is inside the 95% intervals. HN stated-small 69.1% (n=55) vs stated-large 57.1% (n=14, too few to lean on). The CTA condition (small >=3 points BELOW large) failed, so no CTA.
- Q4 not disproved: US 10.3%, EU 9.1%, AU 6.5% (n=353).
- Exploratory, NOT pre-registered: AI-native companies (hand-picked list in results.json) vs the rest. Labelled as such wherever shown.