# Changelog of Y Combinator Jobs Scraper | YC Startup Hiring Data (`tqm/ycombinator-scraper`) Actor - **URL**: https://apify.com/tqm/ycombinator-scraper/changelog.md - **Full Actor documentation**: https://apify.com/tqm/ycombinator-scraper.md ## Changelog ### 2026-09-10 (later) — `roles` failed OPEN ```ts if (roles.length > 0 && role) { ... } // before ``` A card with no detected role **bypassed the filter** instead of being excluded. This repo's own `README.md` recorded it as *"`roles` leaks"*; the Store listing never mentioned it, and it was never fixed. **Now fails closed**, with `roles_unknown` as its own rejection reason. **It does not currently bite** — `role` was populated on 20 of 20 cards measured 2026-09-10, so no row reaches the bypass. That is a property of today's board, not of the code. It is the same failure mode that **inverted `categories` on `remoteok-scraper`**, found the same day and biting there because 14 of 99 rows were unclassifiable. > ⚠️ **A filter that skips rows with missing data is a latent inversion.** It is invisible while the > field happens to be well-populated, and it surfaces on the day a buyer pays for the run. ### 2026-09-10 — YC-EQUITY-01: `equityOnly` and `hasEquity` removed `hasEquity` was derived by searching `` `${title} ${companyDescription} ${role}` `` for the word "equity". That text never contains it, so `hasEquity` was **`false` on every row ever returned** and `equityOnly: true` filtered out the **entire** dataset, always. Confirmed against the live Store Actor 2026-09-10: `equityOnly: true`, `maxJobs: 50` → 0 rows, FAILED. **Both are now gone** — from `src/main.ts`, `.actor/INPUT_SCHEMA.json`, `.actor/dataset_schema.json` and both READMEs. **Why removal rather than a louder warning.** The listing already warned — the input was titled *"(rarely matches)"* and its description opened with *"⚠️ Almost always returns nothing"*. That was not good enough. **An input whose documented behaviour is "does nothing" is worse than an absent one**, and a field that is 100% populated with the same wrong value is worse than an honest null, because nobody audits a field that looks full. **And "rarely" was too generous.** Measured against the live board 2026-09-10, **YC's list page carries no equity element at all** — a card's `.whitespace-nowrap` details are `[jobType, department, role, salary?]` and nothing more. Equity is on the individual job page, i.e. one extra page-load per row: the cost structure that already killed the Upwork detail tier (N1) and `indiehackers` (N3). Do not reintroduce this without pricing that fetch first. **TQM's nightly is unaffected** — `apify_ingest.py` never sent `equityOnly` (its config carries a comment saying why) and nothing in the pipeline reads `hasEquity`. Checked before removing, not assumed. ### Unreleased #### 2026-09-08 — a short result set now says whose limit it was Measured against the live Store listing: asking for the advertised maximum of **50** returned **20** (40%), and the count reproduced exactly across separate runs — so this is the board's real inventory, not a flaky scrape. The status message said only `Returned 20 YCombinator postings.`, which a buyer who asked for 50 cannot tell apart from a broken Actor. `finishRun()` now receives `requested: maxJobs` and appends the reason: ``` Returned 20 YCombinator postings. That is every posting on the board matching your filters — asking for 50 will not return more today. ``` `RUN_STATS` also carries `requested` and `boardExhausted`. Logic and tests live in `apify-actor-kit` (`inventory.ts`); `src/kit/` here is generated. ### \[1.1.0] - 2026-08-27 #### Added — a zero-row run now says why, and fails Adopts the shared `src/kit/`, generated from [`apify-actor-kit`](https://github.com/Atredies/apify-actor-kit). Every filter here was previously a bare `continue`, so a run that returned nothing gave **no clue which filter emptied it**. Rejections are now counted per filter and surfaced in the message and in `RUN_STATS.filteredOut`. A refused listing page is counted too, so `blocked` is distinguishable from "no postings found". `failOnZeroResults` (default `true`). #### Fixed — two inputs advertised more than they deliver The breakdown made both measurable in a single run, on 2026-08-27: | input | measured | now | |---|---|---| | `equityOnly: true` | **rejected 20 of 20** | titled *rarely matches*; description states that YC's list page does not state equity for most postings, so the filter nearly always empties the result | | `techStackFilter` (devops/kubernetes/aws) | **rejected 19 of 20** | description states the stack is inferred from a very short title and blurb, and suggests filtering client-side for recall | Neither is silently broken any more: an emptied run now fails and names the filter that did it. **A filter that silently does not apply is a refund request.** #### Fixed — `maxJobs` advertised a ceiling it cannot reach The schema offered up to **1,000**. YCombinator's jobs page is a single un-paginated list — measured 2026-08-27 it carried **20** postings, and this actor does not paginate. Maximum corrected to **50**. #### Changed - One batched `Dataset.pushData()` per run instead of one call per row. ### Unreleased — 2026-08-13 Documentation only. **No behavior changes**; build `0.0.3` is unmodified. #### Added - Full documentation suite: `docs/ARCHITECTURE.md`, `docs/DEVELOPMENT.md`, `docs/YC-DOM.md`, `docs/IMPROVEMENT-PLAN.md`, and `CLAUDE.md`. - `.actor/dataset_schema.json` — the output surface was previously undeclared. - README rewritten as a buyer-facing listing with an explicit known-limitations table. #### Verified against live YC Two platform runs on 2026-08-13: | Input | Rows | Cost | Duration | |---|---:|---:|---:| | `maxJobs: 10` | 10 | $0.0010 | 5 s | | `maxJobs: 100` | 20 | $0.0052 | 43 s | 20 rows is source exhaustion: one server-rendered page, no pagination. Runs on `CheerioCrawler` over plain HTTP with no browser, which is why it is among the cheapest actors in the portfolio. Data quality is the best of the TypeScript actors — `ycBatch`, `role`, `companyDescription` and a genuine multi-currency salary parser (`USD`/`GBP`/`EUR`/`INR`) all populate reliably. #### Findings recorded (not fixed) - A zero-row run exits **SUCCEEDED** — total scrape failure is indistinguishable from success. - **`equityOnly: true` returns zero rows, always.** `hasEquity` is derived from title + company blurb + role, which never mention equity, so it is `false` on every row. - Every selector is a **Tailwind utility class** (`li.my-2.flex.h-auto.w-full`, …). Any YC restyle breaks extraction silently. Most fragile actor in the portfolio. - Detail chips (`jobType`, department, `role`, salary) are read **by array index** — a missing chip shifts every later field into the wrong chip. - `maxJobs` advertises a maximum of 1,000; ~20 is achievable. - `postedAt` is relative prose (`"2 days ago"`), not a timestamp. - `techStack` is matched against title + company blurb + role, never a job description — empty on \~8 of 10 rows. - An unparseable salary is dropped entirely rather than kept as raw text. - The `K`/`M` multiplier is applied string-wide rather than per number. - The `roles` filter is bypassed for jobs with no detected role. - The `jobId` fallback embeds `Date.now()`. - `department` is extracted into a local variable and never emitted. All are itemized and prioritized in `docs/IMPROVEMENT-PLAN.md`, which flags one investigation to run **before** any of the DOM work: check whether YC ships its listings as embedded JSON, since that would replace the fragile selectors and solve pagination in a single change. *** ### 0.0.3 and earlier No changelog was kept. Build `0.0.3` is the deployed build as of 2026-08-13; the last source commit was 2025-11-16.