Knowledge quality
Contradiction detection, health scoring and compiled syntheses.
Most memory products accumulate whatever they are fed. Chordian also maintains what it holds: it notices when sources disagree, measures how good its own knowledge is, and compiles what it knows into readable profiles.
Knowledge integrity
LiveThe industry default is last-write-wins: your CRM says a deal is In Negotiation, an email two weeks later says it closed, the later sync overwrites, and nobody is told the data ever disagreed. The discarded value vanishes and there is no decision trail.
Chordian records the disagreement instead. Three conflict types are detected inline during ingestion:
| Type | Trigger |
|---|---|
| Value | Two text values are semantically distant — "closed" against "in negotiation" |
| Numeric | Values differ by more than 15%, with unit awareness so 240k and 240,000 do not false-flag |
| Temporal | Two dates more than 30 days apart |
Detection is an alert, never a gate. Ingestion is never blocked or slowed — a sync does not fail because two sources disagree.
Resolution writes your choice back to the graph and records who decided:
| Resolution | Use when |
|---|---|
accept_incoming | The new value is right |
keep_existing | The current value is right |
both_valid | Both were true at different times — records a transition |
ignore | Not a real conflict |
both_valid is the one people forget, and the one that preserves history.
Attribute-level comparison is live. Model-driven fact-level comparison across free text — checking whether two documents assert contradictory claims in prose — is on the roadmap, not shipped.
Health
LiveA report card for the quality of the knowledge, not its security. Four dimensions, weighted into one A–F grade, scored for the whole memory and for each department:
| Dimension | Weight | A low score means |
|---|---|---|
| Freshness | 0.30 | A source stopped syncing |
| Coverage | 0.25 | Many entities, thin supporting evidence |
| Connectivity | 0.20 | Isolated entities with no relationships |
| Coherence | 0.25 | A backlog of duplicates or unresolved contradictions |
A worked example — Grade F (48%): Connectivity 100%, Freshness 50%, Coverage 17%, Coherence 35%. In plain English: the graph is superbly connected but thin on documents (1,749 entities against 415 documents), carries a clean-up backlog, and has a connector that has not synced in about ten days. Three different fixes; the grade alone would not have told you which.
Scored nightly, with history retained so you can watch it improve.
Compiled syntheses
LiveAsking about an account normally means searching hundreds of documents and re-deriving the answer every time — slow, an inference call per question, and subtly different on each asking.
A synthesis is a pre-built profile per entity — overview, current relationship, key contacts, recent activity, risks, likely outcome — kept current in the background.
| Without | With |
|---|---|
| Search 300 documents per question | Read one compiled page |
| An inference call every time | Cost paid once, reused |
| Answer varies between askings | Consistent every time |
A synthesis is a summary, never ground truth: every response carries its
source_count and source_ids.
Compilation, storage, refresh and the console UI are live. Ranking syntheses
first inside search results is not yet wired — today you fetch a synthesis
explicitly rather than receiving it automatically from /search.