# Hiring signals

Read when hiring intent selects the jobs lane. Jobs-only planning exceptions apply only within this mode.

### Hiring Signals mode (`--hiring-signals`)

Use `--hiring-signals` when the user asks what a company's jobs page, careers page, LinkedIn jobs, or competitor hiring suggests about strategic focus. This is strongest for early-stage startups and weaker for large companies, where many unrelated roles are hiring noise.

**Hit the company's OWN job board - that is the entire point.** The engine fetches the company's direct ATS (Greenhouse, Ashby, Lever, Workable, SmartRecruiters) via careers-page-first discovery: it reads the careers page, detects the ATS provider + slug from the embed/link, and calls that API for the full structured board. Aggregators (Glassdoor, Indeed, ZipRecruiter, LinkedIn) are a noisy, lossy last resort, not the source. The engine's output records which `tier` produced the data (`ats` = authoritative, `careers-jsonld` = structured page data, `web` = noisy fallback); weight your confidence accordingly and say so if the run fell to the `web` tier. On Claude Code you can help discovery: read the company's careers page during pre-research, find the ATS board URL (e.g. `jobs.ashbyhq.com/{slug}`), and the engine will resolve the rest.

**Weight by novelty and departure-from-baseline, not raw role count.** A single strategic role can outweigh a department's worth of headcount. The engine surfaces a `Strategic single-role signals` list (founding / first-of-function / specialized / new-geo flags) that is NOT count-gated - read it and judge true novelty yourself, because "is this domain new for this company?" needs world knowledge a keyword map cannot encode. Concretely: 5 engineer roles in a company's core area = "doubling down" (scale signal); 2 roles in an area they have never worked in = a "new bet" (direction signal) and usually the more important story. A `Founding {Role}, {New Capability}` posting (e.g. "Founding Research Scientist, Human Simulation" at a company built on real human interviews) is exactly the high-signal tell that raw counting buries. In synthesis, distinguish "new bets" from "doubling down" in the prose rather than ranking purely by how many roles share a theme.

**Output title for a scoped `--hiring-signals` report.** This is a scoped report, not a general run - it gets a scoped title instead of the `What I learned:` label. Badge on line 1, blank line 2, then `# {Company} - Hiring Signals` on line 3, then the synthesis. Lead with the strongest strategic signal (often a new bet), then the scale signals, then the engine's `## Hiring Signals` evidence block.

**`--hiring-signals` is jobs-scoped - do not build a multi-source plan for it.** When `--hiring-signals` is set the engine searches the jobs source only (it ignores the per-subquery `sources` in your `--plan`). So for a pure hiring-signals run, skip the Step 0.75 multi-source plan work - a 1-subquery plan (or no `--plan` at all) is sufficient, and a rich reddit/x/youtube plan is wasted effort because it gets discarded. If the user wants hiring signals AND community sentiment in one run, pass an explicit `--search=reddit,x,jobs` alongside `--hiring-signals` (the explicit `--search` flag is what keeps the other sources alive).

The output must distinguish evidence from interpretation. Good: "3 current roles mention SSO, SOC 2, and procurement workflows, which signals increased enterprise-readiness focus." Bad: "They will ship enterprise SSO next quarter." In standard `/last30days Company` runs, include Hiring Signals only when the engine surfaces a strong signal; otherwise omit the topic entirely.

---