# Changelog of Instagram Creator & Influencer Marketing Intel (`seibs.co/instagram-creator-intel`) Actor

- **URL**: https://apify.com/seibs.co/instagram-creator-intel/changelog.md
- **Full Actor documentation**: https://apify.com/seibs.co/instagram-creator-intel.md

## Changelog

### 0.2 - 2026-09-27 - Reliability: one shared browser per run

- **Browser tier rebuilt around one shared Chromium per run** (`src/browser_session.py`): each fetch or flow gets an isolated context on the same browser, open pages are capped by run memory (1 below 2 GB, 2 at 2 GB, 3 at 4 GB+), each browser fetch is capped at 60 s and each flow at 120 s, and a circuit breaker switches the tier off after 3 consecutive failures or challenge pages so a blocked run finishes promptly. Previously every escalated request launched its own headful Chromium, several at once, which can exhaust the container's memory and kill the run.
- A private Xvfb starts when the image ships one and no display is running (headful passes edges that block headless); `--disable-dev-shm-usage` keeps renderers alive on the 64 MB container /dev/shm; a missing patchright Chromium falls back to the Playwright build (the Dockerfile now installs both).
- The challenge wait checks before sleeping (no fixed 4 s idle per page) and a missing results selector releases the page after 20 s.
- `browser_telemetry()` on the client reports mode, launches, pages served and breaker state.

### 0.1 - Initial release

- Three modes: creator\_profile, audience\_quality, roster\_compare.
- Primary surface: public Instagram profiles via the logged-out web\_profile\_info endpoint (profile + 12 most-recent posts with like/comment counts). Optional TikTok + YouTube cross-reference for a multi-platform reach view.
- Influencer-analytics layer (the Modash / HypeAuditor product): engagement rate, avg/median likes & comments, like:comment ratio, posting cadence, format mix, top content, follower-tier benchmark comparison, and category inference.
- Audience-quality / fake-follower heuristic estimate from public signals only (engagement rate vs follower-tier band, like:comment ratio, following ratio, engagement uniformity) -> credibility score + band + flags. Clearly labeled an estimate, never a definitive verdict.
- Cross-platform reach rollups (group by canonical handle) and a cross-creator roster-comparison ranking table.
- Anti-bot escalation ladder (httpx -> curl\_cffi Chrome TLS impersonation -> Playwright browser that loads the public profile page and captures its own web\_profile\_info XHR) for Instagram's token-locked endpoint; fail-soft documented notes otherwise.
- Monitor mode: scheduled runs emit a follower / engagement-rate movement digest with optional Slack webhook.
- Portfolio cost-control: pre-flight input caps, demo-mode soft-fail, \_RunBudget guard, in-loop budget checks, hardcoded PPE prices.
- Legal-safety: logged-out public profiles only, no login / no account creation, PII minimized to the public creator/business identity, polite rate limits. Meta v. Bright Data (2024) supports logged-off scraping of public Meta data.