Press n or j to go to the next uncovered block, b, p or k for the previous block.
| 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 | 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 9x 9x 9x 9x 1x 1x 9x 9x 9x 1x 1x 9x 9x 11x 11x 11x 11x 7x 7x 11x 8x 16x 16x 16x 16x 14x 14x 14x 14x 14x 14x 14x 14x 14x 16x 16x 16x 16x 16x 16x 16x 16x 16x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 6x 6x 6x 6x 6x 6x 1x 1x 6x 1x 1x 6x 1x 1x 6x 6x 6x 6x 6x 6x 6x 6x 6x 1x 1x 1x 1x 1x 1x 1x 1x 6x 6x 6x 6x 6x 6x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 7x 7x 7x 7x 7x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 4x 16x 4x 5x 5x 4x 5x 5x 5x 5x 5x 5x 5x 4x 16x 16x 16x 16x 2x 2x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 14x 42x 42x 13x 13x 13x 42x 14x 14x 13x 13x 14x 1x 1x 1x 1x 1x 6x 6x 1x 1x 1x 1x 6x 1x 1x 1x 1x 1x 1x 1x 14x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 16x 49x 49x 49x 49x 49x 49x 16x 16x 16x 49x 16x 16x 16x 16x 16x 16x 1x 1x 1x 1x 1x 1x 4x 4x 4x 4x 3x 3x 3x 3x 3x 3x 3x 3x 3x 4x 4x 1x 1x 1x 3x 3x 3x 1x 6x 6x 6x 1x 3x 3x 3x 2x 2x 1x 1x 1x 3x 3x 3x 3x 1x 2x 3x 1x 1x 1x 3x 3x 1x 3x 3x 1x 1x 1x 3x 3x 3x 3x 3x 3x 3x 1x 1x 1x 2x 2x 2x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x 1x | import type {
ExtractArgs,
ExtractionProvenance,
JobAnalysis,
PageExtract,
Salary,
SalaryPeriod,
StructuredSource,
} from "../types/job";
/**
* Extraction payload as produced in the page; canonicalUrl is added by the
* caller. `structuredSources` and `provenance` are optional on the wire type
* (older extensions omit them) but always present here — the page either
* produces them or does not run.
*/
export type RawPageExtract = Omit<
PageExtract,
"canonicalUrl" | "structuredSources" | "provenance"
> & {
structuredSources: StructuredSource[];
provenance: ExtractionProvenance;
};
/** Cap applied to mainText before transmission (postings are short; cf. 80k video cap). */
export const MAIN_TEXT_CAP = 40_000;
/** Below this many characters (with no JSON-LD) the page is not worth analyzing. */
export const MIN_TEXT_CHARS = 300;
/**
* Runs inside the target page via chrome.scripting.executeScript({ func, args }).
* The function is serialized, so it MUST be self-contained: no imports, no
* references to module scope, no chrome.* (constants are duplicated inline on
* purpose). Everything it needs from the extension arrives in `args`.
*/
export function extractPage(args: ExtractArgs = {}): Promise<RawPageExtract> {
const learnedSelector = args.learnedSelector;
const CAP = 40_000; // keep in sync with MAIN_TEXT_CAP
const MIN = 300; // keep in sync with MIN_TEXT_CHARS
/**
* Hard ceiling on the readiness wait. The corpus's slowest fixture
* (phenom-intuit) settles at 4883 ms, so anything under ~5 s fails the page
* that motivated the feature; 6000 ms is that floor plus headroom, and still
* far inside the 30 s analysis ceiling (constitution IV).
*/
const BUDGET_MS = 6000;
/**
* Quiet period required after text clears MIN. Returning at the first
* crossing captures ~3% of the Phenom posting: its text crosses 300 at
* ~3440 ms and keeps growing until ~4280 ms.
*/
const STABILITY_MS = 400;
/**
* Shorter quiet period for a document that was already complete and already
* had usable text when extraction started. This is the cost charged to the
* five fixtures that settle in 973-1589 ms, so it is deliberately small.
*/
const FAST_PATH_MS = 150;
const POLL_MS = 50;
const startedAt = Date.now();
const NOISE_SELECTOR = "nav, header, footer, aside, script, style, noscript";
function harvestJsonLd(): Record<string, unknown>[] {
const jsonLd: Record<string, unknown>[] = [];
const scripts = document.querySelectorAll('script[type="application/ld+json"]');
for (const script of Array.from(scripts)) {
let parsed: unknown;
try {
parsed = JSON.parse(script.textContent ?? "");
} catch {
continue;
}
const nodes: unknown[] = Array.isArray(parsed) ? parsed : [parsed];
for (const node of [...nodes]) {
if (node && typeof node === "object" && Array.isArray((node as { "@graph"?: unknown[] })["@graph"])) {
nodes.push(...((node as { "@graph": unknown[] })["@graph"]));
}
}
for (const node of nodes) {
if (!node || typeof node !== "object") continue;
const type = (node as { "@type"?: unknown })["@type"];
const types = Array.isArray(type) ? type : [type];
if (types.includes("JobPosting")) {
jsonLd.push(node as Record<string, unknown>);
}
}
}
return jsonLd;
}
function textOf(element: Element): string {
const clone = element.cloneNode(true) as HTMLElement;
for (const noise of Array.from(clone.querySelectorAll(NOISE_SELECTOR))) {
noise.remove();
}
const raw: string =
(clone as { innerText?: string }).innerText ?? clone.textContent ?? "";
return raw
.replace(/[ \t]+/g, " ")
.replace(/\s*\n\s*/g, "\n")
.trim();
}
/** class/id substrings that mark chrome rather than posting content (FR-012). */
const NOISE_KEYWORDS_RE =
/nav|menu|footer|cookie|consent|banner|sidebar|related|similar|promo|skip/i;
/** Minimum text length for an element to be considered a scoring candidate. */
const CANDIDATE_MIN_CHARS = 200;
function attrsOf(el: Element): string {
const id = typeof el.id === "string" ? el.id : "";
const className = typeof el.className === "string" ? el.className : "";
const role = el.getAttribute("role") ?? "";
// Frameworks that hash class names for CSS-in-JS (Workday's Emotion
// classes, e.g. "css-3zv1wr", are opaque) instead carry the semantic
// label on a data-* attribute (data-automation-id="footerContainer").
// class/id alone would never see this signal.
let dataAttrs = "";
for (const attr of Array.from(el.attributes)) {
if (attr.name.startsWith("data-")) dataAttrs += ` ${attr.value}`;
}
return `${id} ${className} ${role}${dataAttrs}`;
}
/**
* Text-density / link-density container scorer (FR-012). Body-fallback
* loses on 3/6 corpus fixtures today because the three-selector guess has
* no middle ground between "a semantic container exists" and "swallow the
* entire body" — this replaces that cliff with a score every element earns.
*
* score = textLength * (1 - linkDensity) * penalty
*
* `penalty` is 1 for a clean candidate. It drops sharply (0.1) when the
* candidate itself, or an ancestor up to (not including) <body>, carries a
* noise-keyword class/id — this is what keeps a cookie-consent dialog's own
* subtree from ever winning even though its text alone clears the minimum.
* It also drops proportionally to how much of the candidate's OWN text
* lives inside a noise-keyword descendant (a skip-link or footer sitting
* alongside real content inside the same wrapping element) — without this
* second term, a big-enough wrapper containing both the posting and a
* banner would still out-score a tighter container purely on raw length,
* exactly the failure mode a naive length-only scorer has.
*/
function scoreCandidate(el: Element): { text: string; score: number } | null {
// NOISE_SELECTOR strips these as DESCENDANTS of some other candidate, but
// that never disqualifies the noise element itself from being scored AS
// a candidate — a <script type="application/ld+json"> is plain text with
// zero link density and no noise-keyword class, so it would otherwise
// outscore the real content it sits beside.
if (el.matches(NOISE_SELECTOR)) return null;
const text = textOf(el);
if (text.length < CANDIDATE_MIN_CHARS) return null;
const clone = el.cloneNode(true) as HTMLElement;
for (const noise of Array.from(clone.querySelectorAll(NOISE_SELECTOR))) {
noise.remove();
}
let anchorLength = 0;
for (const anchor of Array.from(clone.querySelectorAll("a"))) {
anchorLength += (anchor.textContent ?? "").length;
}
const linkDensity = text.length > 0 ? Math.min(1, anchorLength / text.length) : 0;
let selfOrAncestorNoisy = false;
for (
let node: Element | null = el;
node && node !== document.body && node !== document.documentElement;
node = node.parentElement
) {
if (NOISE_KEYWORDS_RE.test(attrsOf(node))) {
selfOrAncestorNoisy = true;
break;
}
}
let noiseDescendantLength = 0;
if (!selfOrAncestorNoisy) {
const seen = new Set<Element>();
for (const descendant of Array.from(clone.querySelectorAll("*"))) {
if (!NOISE_KEYWORDS_RE.test(attrsOf(descendant))) continue;
// Skip a match nested inside an already-counted match, so a banner's
// own inner divs are not double-charged against the total.
let counted = false;
for (const seenEl of seen) {
if (seenEl.contains(descendant)) {
counted = true;
break;
}
}
if (counted) continue;
seen.add(descendant);
noiseDescendantLength += textOf(descendant).length;
}
}
const noiseFraction = text.length > 0 ? noiseDescendantLength / text.length : 0;
const penalty = selfOrAncestorNoisy ? 0.1 : Math.max(0.15, 1 - noiseFraction);
const score = text.length * (1 - linkDensity) * penalty;
return { text, score };
}
/** JobPosting microdata (`itemscope itemtype=".../JobPosting"`), FR-008. */
function harvestMicrodata(): Record<string, unknown>[] {
const results: Record<string, unknown>[] = [];
const roots = Array.from(
document.querySelectorAll('[itemscope][itemtype*="JobPosting"]')
);
for (const root of roots) {
results.push(readMicrodataItem(root));
}
return results;
}
function readMicrodataItem(scope: Element): Record<string, unknown> {
const data: Record<string, unknown> = {};
for (const el of Array.from(scope.querySelectorAll("[itemprop]"))) {
// Don't cross into a nested itemscope's own props — read it as one
// nested object instead of flattening it into the parent.
let owner: Element | null = el.parentElement;
let nested = false;
while (owner && owner !== scope) {
if (owner.hasAttribute("itemscope")) {
nested = true;
break;
}
owner = owner.parentElement;
}
if (nested) continue;
const prop = el.getAttribute("itemprop");
if (!prop) continue;
const value = el.hasAttribute("itemscope")
? readMicrodataItem(el)
: microdataValueOf(el);
if (value === null || value === "") continue;
data[prop] = value;
}
return data;
}
function microdataValueOf(el: Element): string {
const tag = el.tagName.toLowerCase();
if (tag === "meta") return el.getAttribute("content") ?? "";
if (tag === "a" || tag === "link") return el.getAttribute("href") ?? "";
if (tag === "img") return el.getAttribute("src") ?? "";
if (tag === "time") return el.getAttribute("datetime") ?? el.textContent ?? "";
return (el.textContent ?? "").trim();
}
/**
* One synchronous pass over the current DOM — the whole of what extraction
* used to be. Pulled into a closure so the readiness loop can call it
* repeatedly within a single injection: one isolated world, one
* MutationObserver. Repeated executeScript calls could not share observer
* state (research R2).
*/
function attempt(): {
jsonLd: Record<string, unknown>[];
mainText: string;
strategy: RawPageExtract["provenance"]["strategy"];
structuredSources: StructuredSource[];
/**
* Reflects THIS attempt only, never a running/sticky flag: a selector
* missing at t=0 on a hydrating SPA is not a miss (contract "Interaction
* with readiness") — if a later attempt within the same readiness cycle
* succeeds, this must read false again, not stay stuck true from an
* earlier attempt.
*/
learnedSelectorMissed: boolean;
} {
const jsonLd = harvestJsonLd();
const microdata = harvestMicrodata();
const structuredSourcesRaw: StructuredSource[] = [
...jsonLd.map((data): StructuredSource => ({ kind: "json-ld", data })),
...microdata.map((data): StructuredSource => ({ kind: "microdata", data })),
];
// FR-011: cap total transmitted size, dropping sources whole rather than
// truncating mid-object — a truncated JSON blob is worse than none.
const structuredSources: StructuredSource[] = [];
let structuredSize = 0;
for (const source of structuredSourcesRaw) {
const size = JSON.stringify([source]).length;
if (structuredSize + size > 20_000) continue;
structuredSources.push(source);
structuredSize += size;
}
// Learned selector (FR-028, FR-029, contracts/learned-selectors.md):
// applied AHEAD of every automatic strategy, including the Google overlay
// special case below. A miss here is not necessarily a site redesign — on
// a hydrating SPA the selector may simply not exist yet at this attempt,
// which is exactly why this runs inside the readiness loop rather than
// once before it. The caller (jobFlow.ts) is the one that turns a
// still-missing selector, after the budget expires, into recordMiss().
let learnedSelectorMissed = false;
if (learnedSelector) {
const learnedEl = document.querySelector(learnedSelector);
const learnedText = learnedEl ? textOf(learnedEl) : "";
if (learnedText.length >= MIN) {
return {
jsonLd,
mainText: learnedText,
strategy: "learned-selector",
structuredSources,
learnedSelectorMissed: false,
};
}
learnedSelectorMissed = true;
}
// Google's job-detail overlay: the SERP itself is a list page, but the
// selected posting is rendered inside a visible [role=dialog]. Prefer that
// dialog's text — read via innerText on the LIVE node, whose layout-aware
// semantics skip the hidden preloaded detail views of *neighboring* jobs
// stacked in the same dialog. textOf would leak them: it works on a
// detached clone, where innerText degrades to textContent and hidden
// subtrees come back. Scoped to Google search pages: elsewhere a visible
// dialog is more likely a cookie-consent or chat widget than the posting.
let overlayText = "";
if (
/(^|\.)google\.[a-z]{2,3}(\.[a-z]{2})?$/.test(location.hostname) &&
location.pathname === "/search"
) {
for (const dialog of Array.from(document.querySelectorAll('[role="dialog"]'))) {
const rect = dialog.getBoundingClientRect();
if (rect.width < 100 || rect.height < 100) continue;
const raw: string =
(dialog as { innerText?: string }).innerText ?? dialog.textContent ?? "";
const text = raw
.replace(/[ \t]+/g, " ")
.replace(/\s*\n\s*/g, "\n")
.trim();
if (text.length > overlayText.length) overlayText = text;
}
}
let mainText = "";
let strategy: RawPageExtract["provenance"]["strategy"] = "semantic";
if (overlayText.length >= MIN) {
mainText = overlayText;
strategy = "google-overlay";
} else {
// The semantic container wins OUTRIGHT, verbatim, whenever it clears
// the minimum — exactly today's behavior (FR-013: "MUST NOT change the
// selected text on pages where the existing semantic-container path
// already yields the posting"). This is deliberately a hard rule, not
// an emergent property of the scorer: greenhouse-discord's and
// phenom-intuit's own <main> legitimately contains some
// noise-keyword-matched material today (a "related jobs" rail is
// *inside* phenom's <main>, not a sibling of it) that any density
// penalty strong enough to fix Workday's footer would also strip from
// these two protected fixtures, changing their byte count and failing
// FR-013. Scoring therefore only runs — per FR-012's first clause —
// when the semantic path "yields nothing" (or below-minimum text);
// FR-012's second clause ("or yields a container that scores worse")
// is not exercised by any corpus fixture and is intentionally left
// unimplemented rather than tuned against nothing.
let semanticText = "";
for (const selector of ["main", "[role=main]", "article"]) {
const el = document.querySelector(selector);
if (el) {
const text = textOf(el);
if (text.length > semanticText.length) semanticText = text;
}
}
if (semanticText.length >= MIN) {
mainText = semanticText;
strategy = "semantic";
} else {
let bestScore = -1;
let bestText = "";
if (document.body) {
for (const el of Array.from(document.body.querySelectorAll("*"))) {
const evaluated = scoreCandidate(el);
if (!evaluated) continue;
if (evaluated.score > bestScore) {
bestScore = evaluated.score;
bestText = evaluated.text;
}
}
}
if (bestText.length >= MIN) {
mainText = bestText;
strategy = "scored";
} else if (document.body) {
const bodyText = textOf(document.body);
if (bodyText.length > mainText.length) {
mainText = bodyText;
strategy = "body";
}
}
}
}
return { jsonLd, mainText, strategy, structuredSources, learnedSelectorMissed };
}
return new Promise<RawPageExtract>((resolve) => {
let attempts = 0;
let latest = attempt();
attempts += 1;
/**
* Readiness keys on BODY TEXT ONLY.
*
* The old predicate was `jsonLd.length > 0 || text >= 300`, which passes on
* JSON-LD alone before any body text exists — on 3 of 6 corpus fixtures it
* fires while mainText is still empty (Workday at 8 ms with text at 736 ms,
* Ashby at 10 ms / 369 ms, Phenom at 3314 ms / 3440 ms). A retry loop built
* on it would return instantly with an empty body and be no better than
* shipping nothing. Structured data must not satisfy readiness (FR-001),
* and we must not return below-minimum text while budget remains (FR-004a).
*/
const hasUsableText = (): boolean => latest.mainText.length >= MIN;
// A document that was already complete with usable text on the first pass
// gets the short quiet period. Any observed mutation means the page is
// still building, so it graduates to the full stability window.
let quietNeeded =
document.readyState === "complete" && hasUsableText() ? FAST_PATH_MS : STABILITY_MS;
let lastGrowthAt = Date.now();
let dirty = false;
const observer = new MutationObserver(() => {
dirty = true;
if (quietNeeded === FAST_PATH_MS) quietNeeded = STABILITY_MS;
});
observer.observe(document.documentElement, {
childList: true,
subtree: true,
characterData: true,
});
const finish = (budgetExpired: boolean): void => {
observer.disconnect();
clearInterval(timer);
resolve({
url: location.href,
title: document.title,
jsonLd: latest.jsonLd,
structuredSources: latest.structuredSources,
mainText: latest.mainText.slice(0, CAP),
extractedAt: new Date().toISOString(),
provenance: {
strategy: latest.strategy,
attempts,
elapsedMs: Date.now() - startedAt,
budgetExpired,
learnedSelectorMissed: latest.learnedSelectorMissed,
},
});
};
const timer = setInterval(() => {
// Re-attempt only when something actually changed. attempt() clones and
// walks subtrees, so polling it unconditionally would burn CPU on every
// page for the whole budget.
if (dirty) {
dirty = false;
const previousLength = latest.mainText.length;
latest = attempt();
attempts += 1;
if (latest.mainText.length > previousLength) lastGrowthAt = Date.now();
}
if (hasUsableText() && Date.now() - lastGrowthAt >= quietNeeded) {
finish(false);
return;
}
if (Date.now() - startedAt >= BUDGET_MS) finish(true);
}, POLL_MS);
// A page that is already settled still has to clear its quiet period, but
// nothing else is pending, so the first interval tick resolves it.
});
}
/**
* Client-side fallback fields derived from schema.org JobPosting JSON-LD.
* Rendered when the backend is unreachable so the panel is never empty-handed.
*/
export function deriveJsonLdFields(
jsonLd: Record<string, unknown>[]
): Partial<JobAnalysis> {
const posting = jsonLd[0];
if (!posting) return {};
const fields: Partial<JobAnalysis> = {
title: stringOrNull(posting["title"]),
company: orgName(posting["hiringOrganization"]),
location: locationText(posting["jobLocation"]),
salary: salaryOf(posting["baseSalary"]),
};
const locationType = posting["jobLocationType"];
const locationTypes = Array.isArray(locationType) ? locationType : [locationType];
if (locationTypes.includes("TELECOMMUTE")) {
fields.arrangement = "remote";
fields.arrangementConfidence = "explicit";
}
return fields;
}
function stringOrNull(value: unknown): string | null {
return typeof value === "string" && value.length > 0 ? value : null;
}
function orgName(org: unknown): string | null {
if (typeof org === "string") return org || null;
if (org && typeof org === "object") {
return stringOrNull((org as Record<string, unknown>)["name"]);
}
return null;
}
function locationText(location: unknown): string | null {
const place = Array.isArray(location) ? location[0] : location;
if (typeof place === "string") return place || null;
if (!place || typeof place !== "object") return null;
const address = (place as Record<string, unknown>)["address"];
if (typeof address === "string") return address || null;
if (!address || typeof address !== "object") return null;
const parts = ["addressLocality", "addressRegion", "addressCountry"]
.map((key) => (address as Record<string, unknown>)[key])
.filter((v): v is string => typeof v === "string" && v.length > 0);
return parts.length > 0 ? parts.join(", ") : null;
}
function salaryOf(baseSalary: unknown): Salary | null {
if (!baseSalary || typeof baseSalary !== "object") return null;
const salary = baseSalary as Record<string, unknown>;
const value = salary["value"];
const valueObj =
value && typeof value === "object" ? (value as Record<string, unknown>) : salary;
const min = numberOrNull(valueObj["minValue"] ?? valueObj["value"]);
const max = numberOrNull(valueObj["maxValue"]);
const currency = stringOrNull(salary["currency"]);
const period = periodOf(valueObj["unitText"]);
if (min === null && max === null && currency === null) return null;
return { min, max, currency, period };
}
function numberOrNull(value: unknown): number | null {
return typeof value === "number" && Number.isFinite(value) ? value : null;
}
function periodOf(unitText: unknown): SalaryPeriod | null {
if (typeof unitText !== "string") return null;
const map: Record<string, SalaryPeriod> = {
YEAR: "year",
MONTH: "month",
DAY: "day",
HOUR: "hour",
};
return map[unitText.toUpperCase()] ?? null;
}
|