About Rewindex
Rewindex is an archive of how SaaS vendors change over time. Vendor pages overwrite themselves: a status page shows today’s version of an incident, a pricing page shows today’s prices, a subprocessor list shows today’s list. Once it changes, the old version is gone. Rewindex keeps every version.
What is archived
- Status-page incidents from six platforms: Atlassian Statuspage, incident.io, Better Stack and Instatus (public JSON APIs) and status.io (components API, RSS feed and incident pages), polled every 10 minutes, plus the Salesforce Trust API, polled every 30 minutes. Each page’s published history is archived as far back as the platform exposes it.
- Deprecations, shutdowns, version sunsets and breaking changes via Dead Canary (deadcanary.io), and pricing, licensing and support-lifecycle changes via Viz Canary (vizcanary.com), matched to the vendors tracked here (see /deprecations).
- Silent edits: every distinct version of every incident is hashed and kept, so a rewritten title, an edited or deleted update, a downgraded impact or a deleted incident is visible.
- Pricing, trust/security and legal pages (terms, privacy, subprocessors), fetched once a day. A new version is stored only when the visible text changes.
Method
- Change detection uses a SHA-256 hash of a canonical form of each document, so cosmetic churn (timestamps, counters) is not a change.
- Failures are data: every fetch attempt is logged, so gaps in coverage stay visible and uptime figures stay honest.
- Incident impact is the vendor’s own label. Durations are resolved minus started, as published.
- Pages are fetched politely by RewindexBot (see /bot), at most a few requests per host per day for pages.
Hacker News signals
Two Hacker News signals appear on incidents. Both are automated, computed from public data, and shown with their numbers so you can judge them yourself.
- Discussed on Hacker News: an HN story is linked to a vendor’s status incident when the vendor is named in the story or its comments, the thread is about an outage (a high-confidence match such as the vendor’s status-page host, or an outage word in the title), and the first mention falls within 3 hours of the vendor’s first published update. At most three threads are shown per incident, highest points plus comments first. Low-confidence matches (for example a thread about another vendor’s outage that mentions this one in comments) are shown only with at least 100 points and 50 comments. “Posted N min before/after” compares the HN post time with the vendor’s first published update, or with the listed start time when the platform publishes no update times.
- Downplayed?: HN outage threads where the vendor is the subject get a perceived-outage score from 0 to 1: 35% traction (points and comments), 30% velocity (comments in the first hour), 20% distinct commenters in the first hour and 15% people reporting first-hand symptoms or confirmations in the first 6 hours, each as a percentile against the other outage threads of the last 90 days. A thread is flagged at 0.6 or higher (and at least 50 points or 30 comments). When a flagged thread lines up with an incident the vendor rated minor or none (posted from 3 hours before the incident to 1 hour after it was resolved), the incident gets a “Downplayed?” badge. No language model is involved.
- Limits: HN users skew toward developers, so heavy discussion shows broad developer impact, not impact on every customer. Popular vendors (AI tools above all) draw more discussion for the same outage. One thread can cover several vendors at once (e.g. simultaneous OpenAI, Claude and Grok problems). Vendor impact labels are set early and are not always revised. The badge is a prompt to look closer, not a claim that a vendor misrepresented an incident. Scores, points and comments refresh every 30 minutes with the HN poll.
Archive size
433companies
135,583incidents since 2014-10-31
1,048page versions
Access for people, agents and code
- Every page has a Markdown twin: add
.mdto the URL, or sendAccept: text/markdown. - JSON API at
/api/v1(OpenAPI 3.1). - MCP server at
/mcp(streamable HTTP). - llms.txt and llms-full.txt.