rodado/linkme
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: stop title filters from truncating a LinkedIn search early

Browse Source 
fetchJobs ended pagination on `newOnPage === 0`, but newOnPage counts
jobs that survived the title filters. A page whose 25 cards all happened
to be German/Java/junior/non-software read as end-of-results and killed
the whole search, discarding every later page. The end condition is now
new raw data-entity-urn card ids, pre-filter, so what gets filtered can
no longer decide where the crawl stops.

Also add a name-only "Basel (broad)" search alongside the three geoId
ones. LinkedIn's guest search serves a different slice per query shape:
the Basel-city geoId query exhausts at 76 unique cards and never returns
some genuinely Basel-Stadt postings a name-resolved query does. PAS01's
"Full-Stack Software Engineer (60-80%)" in Basel (4461764782) was
invisible to all six Basel searches and reached the digest only because
the Bern query — also name-only — happened to return it. Union across
the Basel searches: 126 unique.
This commit is contained in:
rodado
2026-09-05 18:04:36 +02:00
parent a596281dae
commit e3368ce161
2 changed files with 32 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fetch.ts
+16 -2
View File
@@ -431,6 +431,7 @@ export async function fetchJobs(config: SearchConfig, { raw = false } = {}): Pro
const baseUrl = buildUrl(config);
const allJobs: Job[] = [];
const seen = new Set<string>();
const seenCards = new Set<string>(); // every card id served, pre-filter — drives the end condition
for (let page = 0; page < MAX_PAGES; page++) {
if (page > 0) await Bun.sleep(400);
@@ -452,6 +453,19 @@ export async function fetchJobs(config: SearchConfig, { raw = false } = {}): Pro
}
const html = await res.text();
// End-of-results is decided on the raw card ids LinkedIn served, never on
// how many survived the title filters. A page whose cards all happen to be
// German/Java/junior/non-software read as "no new jobs" and truncated the
// whole search, discarding every later page — pages that do carry keepers.
let newCards = 0;
for (const m of html.matchAll(/data-entity-urn="urn:li:jobPosting:(\d+)"/g)) {
if (!seenCards.has(m[1])) {
seenCards.add(m[1]);
newCards++;
}
}
const jobs = parseJobs(html, config.label)
.filter((j) => !NON_ENGLISH_RE.test(j.title))
.filter((j) => !JUNIOR_TITLE_RE.test(j.title))
@@ -469,8 +483,8 @@ export async function fetchJobs(config: SearchConfig, { raw = false } = {}): Pro
}
}
console.log(`${jobs.length} jobs, ${newOnPage} new`);
if (newOnPage === 0) break;
console.log(`${jobs.length} jobs, ${newOnPage} new (${newCards} new cards)`);
if (newCards === 0) break;
}
return allJobs;