One round, many headlines

The merge rule

Records are grouped by company inside a ten-day window. Two reports collapse into one financing when both amounts are undisclosed, when the disclosed amounts agree within twenty percent, or when one record names a specific round and the other is a generic reported raise. The surviving record is chosen by review status, named round, investor detail, presence of a source URL and headline completeness — in that order. It carries a count of how many reports were merged.

The window is deliberately short. Quantum companies sometimes raise twice in a quarter, and a broad window would erase genuine second rounds. Where two distinct rounds do fall inside the window with clearly different amounts, both are kept.

Nothing is deleted

The merge happens at publication time only. data/funding-events.json retains every collected report, including the duplicates, so a reader can count them, check the discarded wording and disagree with our decision. This is the difference between de-duplication and editing the evidence. A site that silently drops rows from its source file cannot be audited.

The JSON also records autoCollected and reviewed flags, so a merged surviving row still tells you whether its provenance was human-checked. Rows that have not been reviewed keep a visible ⚡ marker and a clickable source URL.

Why this is also an editorial quality issue

Duplicate headlines produce duplicate sentences. Our first attempt at solving this used a rotation of generated interpretive templates, and several policy notices ended up with nearly identical commentary — exactly the mass-produced impression we were trying to avoid. We removed that generator entirely. Every published event now has its own hand-written note, and an event with no note renders factual labels only, never an invented interpretation.

Publication-layer de-duplication in the current build

The raw ledger has 65 reports. The publication layer publishes 52 distinct financings, so 13 syndicated reports were merged. Every raw report is still present in funding-events.json.

The practical takeaway

If a figure from this site disagrees with a count you saw elsewhere, check the raw ledger first. Most disagreements are a difference in whether syndicated headlines were counted as separate financings.