Freelance Lead Radar

One feed. No noise. Sorted by how much money you make per hour.

Most freelance-job scrapers give you what the platforms give them: hundreds of listings, a third of which are people looking for work, a third of which have no budget, and a long tail of ₹200 data-entry gigs. You still spend 40 minutes a day triaging.

This Actor does the triage for you:

1. Aggregates five independent sources — GitHub bounty issues, Hacker News contract posts, Freelancer.com projects, YesWeHack public bug-bounty programs, and Remotive remote jobs.
2. Removes the noise — [FOR HIRE] posts, "seeking work" comments, listings with no stated budget, sub-$20 gigs, equity-only "co-founder" ads, and bounty-farm repos where forks outnumber stars 3:1 (the classic pattern of a repo that manufactures fake $8k bounties for engagement).
3. Scores every survivor 0–100 on money, effort, fit, freshness, and competition, then reports USD/hour so you can pick by return rather than by headline number.

You pay per lead that survives the filter. Noise we removed is not billed.

---

What you get

One dataset row per lead:

Field	Meaning
title	The lead headline
url	Direct link to the issue / comment / project
source	github · hn · freelancer · yeswehack · remote
kind	bounty (fixed-price task) · gig (contract) · job (salaried)
org	Repo (owner/name), company, or posting author
payoutUsd	Budget converted to USD
effortHours	Estimated hours to deliver
roiUsdPerHour	payoutUsd / effortHours
valueScore	0–100 composite
postedAt	ISO timestamp
bidCount	Competing bids already placed (Freelancer.com only)
labels	Repo labels or required skills
summary	First ~300 chars of the description

A SUMMARY key-value record reports how many items were scraped, how many survived de-noising, and which sources failed — so a partial run is visible rather than silent.

---

Why the noise filter matters

Run against live data at the time of writing, one default run pulled 101 raw items and delivered 85 usable leads — 16 were rejected as noise, before de-duplication removed several more. Across a week of runs the rejected share is typically 20–30%, and it is dominated by three things:

• Bounty farms. A repo with 14 stars and 200 forks posting [Bounty $8k] titles is not paying $8k. Real bounty repos have more stars than forks.
• Job-seeker posts. The monthly HN freelancer thread is roughly half people advertising themselves, not hiring.
• Unpriced listings. A gig with no number is usually a "how much would you charge?" fishing expedition.

Filtering these is the entire product. Everything else is plumbing.

---

Input

Field	Default	Notes
sources	["github","hn","freelancer","yeswehack"]	Add remote for salaried roles
maxLeads	25	Each returned lead is one billed event
minScore	0	Raise to ~50 for a high-precision feed
minPayoutUsd	0	Hard floor on money
perSource	max(maxLeads*2, 30)	Raw items pulled per source before filtering

Example: high-precision, high-payout only

{
  "sources": ["github", "freelancer"],
  "maxLeads": 10,
  "minScore": 55,
  "minPayoutUsd": 300
}

Example: broad daily sweep

{
  "sources": ["github", "hn", "freelancer", "yeswehack", "remote"],
  "maxLeads": 50,
  "minScore": 0
}

---

Pricing

Pay per event — you pay for results, not for noise.

Event	Price	Charged when
lead-returned	$0.015	once for each lead written to the dataset
apify-actor-start	$0.005	once per run (one event per GB of memory, minimum one)

With the default maxLeads: 25, a run costs roughly $0.38. Ask for 5 leads and it's about $0.08.

Why this model:

• A run that finds nothing costs only the start event.
• Filtered-out noise is never billed. You pay for leads that survived, not for what we threw away.
• maxLeads caps your maximum spend per run, so cost is predictable.

Apify bills platform usage separately (compute, proxy, storage); this Actor uses no proxies, no browser and no SERP calls, so that part stays in the sub-cent range.

---

Scheduling

Set a daily schedule (08:00 local) and point a notification at the run. The feed is meant to be read once a day over coffee, not monitored.

---

Limits and honest caveats

• Freelancer.com is rate-limited. Requests are spaced out; a run takes roughly 60–120 seconds. This is why the Actor is not instant.
• GitHub's label:bounty signal is polluted. That is precisely why the repo-health check exists. Without it, roughly 95% of that label's newest issues come from a single farm repo.
• effortHours is a keyword heuristic, not an LLM estimate. It is directionally right (a typo fix is 0.5h, a mobile app is 60h) and deliberately conservative on unknowns (6h default).
• Unpriced listings are ranked low on purpose. A post that never states a number scores 55% lower than an identical priced one, because "how much would you charge?" leads are the single biggest time sink. If you want them anyway, they are still in the dataset — just further down.
• Amounts in prose are read conservatively. A GitHub issue whose body says "€230,000 across 214 programs" is a summary, not a €230,000 bounty, so it is not scored as one. Amounts must sit next to a word like bounty, reward, or budget and must not be part of a total/average statement.
• Bug-bounty programs are ranges, not prices. YesWeHack publishes min/max rewards per program, and taking the ceiling would fake a top result (one program's €100–230,000 range would score as $248,400). We take the geometric mean of the range, or max/10 when only a ceiling exists, and keep the raw range in summary. Effort is pinned at 24h because scope is open-ended — you may hunt for three days, or land it in the first hour.
• No bidding, no applying. This Actor only finds and ranks. Submitting proposals is against the terms of every platform it reads from, and automating it gets accounts banned. You still click submit.
• Payout parsing prefers the first amount in the text. A post that mentions "$500 budget, previously paid $50k" yields $500. Prefer the summary field when a number looks wrong.

---

Sources and their terms

All five are public read-only endpoints that require no login:

Source	Endpoint	Auth
GitHub	api.github.com/search/issues	none (optional token raises rate limit)
Hacker News	hn.algolia.com/api/v1	none
Freelancer.com	freelancer.com/api/projects/0.1	none
YesWeHack	api.yeswehack.com/programs	none (public program list only — we never submit reports or scan targets)
Remotive	remotive.com/api/remote-jobs	none

None of these are scraped with a browser, so there is no fingerprinting, no CAPTCHA solving, and no proxy dependency. If a source changes its API the Actor logs the failure and returns the rest.

---

Local development

pip install -r requirements.txt
python -m src            # runs against real APIs, writes to local storage
python tests/test_radar_core.py   # 53 offline unit tests, no network

The scoring and filtering logic lives in src/radar_core.py and has no Apify dependency, so it can be tested and reused standalone.