How it works

sigcue reads public discussions through official APIs: GitHub issues and comments today, and other permitted sources as they are enabled. It does not scrape pages, use session cookies, or read private data.

Each record keeps its original source, community, identifier, and timestamps so any claim on the site can be checked against the source.

Every record is read for the same structured facts: who is affected, the workflow involved, what failed, any workaround, and the consequence. Records that cannot be described that precisely are not used.

Summaries on this site are written from those facts. The original excerpt and a link to the source stay attached to the evidence.

Records are matched by source identifier, source-scoped thread, canonical link, author hash, and near-duplicate text. Several comments inside one discussion are one report, not many.

Independent reports are separate people or threads describing the same failure. Syndicated copies and near-identical reports from the same author do not add independent support.

Reports are grouped when their structured meaning agrees across affected people, job, workflow, failure, root cause, and desired outcome. Deterministic term normalization recognizes common paraphrases, while complete-link checks prevent one broad bridge report from joining incompatible problems. A shared tool name or broad category is never enough on its own.

Related concrete signals can also roll up into a broader problem, which stays linked to each specific signal underneath it.

Insufficient: fewer than two independent reports. The problem may be described well without being shown to recur.

Early: at least two independent reports, with coverage still thin.

Developing: at least three independent reports across two communities or two weeks.

Strong: at least five independent reports across two communities, two source platforms, and thirty days, with high evidence-set quality, consistent failure descriptions, recent evidence, and counter-evidence reviewed.

Confidence is an inspectable 0–100 evidence score. Its displayed components are independent evidence, source diversity, recency, consistency, and evidence quality, minus a duplicate or dependency penalty.

Low-quality, single-platform, stale, or duplicate-dominated evidence is capped so record volume alone cannot create high confidence.

Reports that resolve, contradict, or qualify a problem are recorded and shown on the signal page rather than dropped. A reported workaround is not treated as a fix.

A public complaint does not prove market size, willingness to pay, or that a business would work. Nothing here is a recommendation to build a particular product.

Public discussions also overrepresent some groups and omit purchasing context, offline work, and people who never post. Use this to ask better questions, not to skip talking to people.

Live signals require either human review or an auto-publish gate that still demands ≥2 linked original sources and skips obvious noise titles. Evidence is never attached without that bar.

Some markets are placeholders for future expansion. They are labeled “Coming soon” and are never counted as live evidence until enough high-quality data has been collected and verified.

When a source record is deleted or becomes unavailable, it is marked and excluded from future metrics, and the signal is recalculated. Correction and removal requests are logged for review through the deployment workspace.