A system of record for the facts your organization has to defend: one canonical record per real thing, a cited, verbatim, dated source behind every value, unknowns declared rather than left blank, and any past state reproducible on demand.
We reply within two business days.
A record, as the person acting on it sees it
Halden Therapeutics is fictional. The behaviour is not: select a value, or replay the record at an earlier date.
Halden Therapeutics AG
Lead program
HLD-204
“Following the January reprioritisation, HLD-204 is our lead candidate and our principal development priority.”
chosen over
haldentx.com/pipeline, page not revised since August 2025
A dated presentation outranks an unrevised web page.
In production in two sectors
Deployed inside customer infrastructure, against live private and public sources, with their own policies.
6,000+ identity decisions by rule, none reversed
More than six thousand decisions about whether two records describe one real thing, made by rule. None has since been reversed. The count includes evaluation runs and has not been independently audited (counting rule).
Every value cites a retained source, verbatim
There is no path by which a value reaches a reader without the sentence, the locator and the date it was read.
The sequence
Four properties, in order. Each one is a thing you can check on any value the system shows you, not a description of how the system is built.
The document a value was read from is kept as it was read, with the moment it was fetched.
A value whose source has since moved, changed or vanished can still be defended.
The value points at the sentence it came from, quoted verbatim, with its page or address and the date it was observed.
A paraphrase is not a citation and does not count as one.
Sources that disagree are both kept. The record shows the value it carries, what that value was chosen over, and on what grounds.
Disagreement is reported, never averaged away or silently dropped.
Every value carries the date it was observed and the date it entered the record, and nothing is overwritten.
Any past state of the record can be reproduced and read again.
The four guarantees
These are the whole contract. Everything else the system does exists to make one of them true.
A company, a person, a facility, a program, a contract: each has exactly one record, whatever it has been called. Former names are kept. Records that relate are linked rather than folded together, because merging two real things falsifies both.
exampleA supplier that changed its name twice keeps one identifier; the two former names still resolve to it.
Every value points at the sentence it was read from, quoted word for word, with the page or address, the date it was observed, and a retained copy of the document. A paraphrase is not a citation and is not accepted as one.
exampleA phase reads Phase 2 because a registry record version 9, read on 11 March 2026, says so in those words.
Where there is no evidence, the record says so in words: not found, blocked, or not applicable, each with a reason and the record of what was searched and when. An empty cell is never emitted, because a blank reads as a negative.
exampleA readout date with no published source reads not found, and lists the four sources searched and the dates.
Nothing is overwritten. Every value carries the date it was observed and the date it entered the record, so the record as it stood on any past day can be reproduced and read again, together with the sources that supported it that day.
exampleThe brief you sent in March can be reproduced in September, with the source each line rested on then.
Coverage
The same kernel, with the sources, policies and vocabulary of the organization that runs it. One worked case each.
A pipeline page, a registry record and a February deck name three different lead programs. The record carries one, dated, and shows the two it was chosen over.
pharmaTwo companies share a director, a former trading name and a registered agent. The record keeps them separate, links them, and says on what basis.
financial-servicesA supplier's registered address changed twice in eighteen months. The award file reproduces the address as it stood on the day of award, with the extract.
governmentA deal-room document and a public filing give different completion dates. The memorandum cites both, names the one it relies on, and states why.
legalTwo schedules give different insured values for one site. The reserve carries one figure with the schedule it came from; the other stays on the record.
insuranceA smelter appears under three spellings across four tiers. The record keeps one entity, shows the three names it is known by, and dates each.
supply-chainDeployment
The kernel runs on your infrastructure, with one database per organization. Nothing about your sources, your readers or your retention is decided outside the boundary you draw.
Every value exports with its chain of custody: the source, the excerpt, the locator and the dates, in a form an auditor can read without access to the system.
Research
The problems this product is built on are open research problems. We would rather be corrected in public than be quietly wrong in a customer’s file.
open problems
reading the literature
work with us
Further reading
Entity Resolution Evaluation Beyond Pairwise F1
Pairwise against cluster metrics, sampling design and uncertainty for judging entity resolution, worked through an eight-record example.
Minimum Data Provenance Requirements for Defensible Values
Fifteen minimum provenance fields and five audit questions, mapped to PROV-O, with temporal rules, declared unknowns and validation tests.
Open World Assumption: Why Unknown Is Not False
Open-world negation, explicit negative knowledge and five epistemic states, with the evaluation protocols that keep an unknown from being scored as false.
Valid Time vs Transaction Time: Which Clock Proves What?
Valid, transaction, observation and publication time distinguished through a delayed correction and four as-of tests.
Bring one entity you already argue about internally. We will show you what it looks like when every value on it carries its source.