Every client record is reduced to a 512-bit Bloom filter before it leaves the firm. The matcher compares filters, never names. Below is that system, running: encoding in your browser, scoring on the server, calibrated against a labelled set with the failure cases left in.
These rows never move. Encoding happens here, in the browser, under a salt the matcher operator never holds. Green rows are auto-linked; amber rows land in the review queue.
| Ref | Name | Born | Postcode |
|---|---|---|---|
| A01 | Amelia Hartley | 1984-03-02 | M1 4BT |
| A02 | Jonathan Smith | 1979-11-24 | SW1A 1AA |
| A03 | William Okonkwo | 1991-07-16 | LS1 5QT |
| A04 | Priya Raghunathan | 1988-02-29 | EH3 9DR |
| A05 | Grace Abadi | 1995-05-08 | BS1 6NB |
| A06 | Thomas Wexley | 1972-09-30 | NE1 7RU |
| A07 | Fiona Castellane | 1990-12-11 | CF10 1EP |
| A08 | Rory Nkemelu | 1983-01-19 | B15 2TT |
| A09 | Helena Vorst | 1968-04-25 | G2 8DL |
| A10 | Marcus Delacroix | 1987-08-03 | SW1A 1AA |
| A11 | Ada Whitfield | 1993-06-14 | M1 4BT |
| A12 | Samuel Whitfield | 1996-06-14 | M1 4BT |
| A13 | Ingrid Lindqvist | 1975-10-05 | EH3 9DR |
| A14 | Oscar Bennett-Rowe | 1999-03-21 | LS1 5QT |
| Ref | Name | Born | Postcode |
|---|---|---|---|
| B01 | Amelia Hartley | 1984-03-02 | M1 4BT |
| B02 | Jonathon Smyth | 1979-11-24 | SW1A 1AA |
| B03 | Bill Okonkwo | 1991-07-16 | LS1 5QT |
| B04 | Priya Raghunathan | 1988-02-29 | EH39DR |
| B05 | Grace Abadi-Mensah | 1995-05-08 | BS1 6NB |
| B06 | Thomas Wexley | 1972-03-09 | NE1 7RU |
| B07 | Fiona Castellane | 1990-12-11 | CF10 1EP |
| B08 | Rory Nkemelu | 1983-01-19 | B15 2TT |
| B09 | Nadia Ferreira | 1981-02-17 | G2 8DL |
| B10 | Marcus Delacroix | 1987-08-03 | EC2Y 8DS |
| B11 | Ada Whitfield | 1993-06-14 | M1 4BT |
| B12 | Sam Whitfield | 1996-06-14 | M1 4BT |
| B13 | Yusuf Demirci | 1970-08-22 | L1 8JQ |
| B14 | Clara Bennett | 1999-03-21 | LS1 5QT |
Fourteen filters cross from each firm; the first four of each are drawn here. Hover a result row below and the bits both filters set light up: that intersection, and only that, is what the Dice score is computed from.
encoding…
Ranked by Dice similarity. The last column is what a plaintext equi-join would have returned for the same pair, which is the baseline this replaces.
| Aldgate Trust | Barbican Mutual | Dice | Verdict | Exact join | Why it is hard |
|---|