Conflict-of-interest check
A local registry of your matters and their parties, fuzzy name search, and a log of every check.
Tool for legal professionals. Your data stays in this browser, on this device: nothing is sent to our servers, so client information does not leave your device. Make copies (export): clearing the browser’s data deletes it. The tool assists; the professional judgment is yours.
How names are compared
Before any comparison, each name is normalized:
- accents, capitals and ligatures are neutralized (“Hélène Côté” = “helene cote”);
- punctuation becomes a space, “&” counts as “and”, and initials are joined (“A.B.C.” = “ABC”);
- stop words (the, of, and, le, la, les, de, du, des, et) are removed;
- legal forms at the end of a name are removed: Inc., Ltd., Limited, LLC, LLP, LP, Corp., Co., ltée, cie, S.E.N.C., S.E.N.C.R.L., PLC, GmbH, etc.;
- word order does not matter (“Smith, John” = “John Smith”);
- “d/b/a”, “doing business as”, “a.k.a.”, “formerly”, “f.a.s.” split a name into variants, each compared; a party’s other and former names are compared too.
The score, between 0 and 1, is the highest of the following:
- 1.00: identical after normalization; 0.99: identical without spaces (“Mac Donald” / “MacDonald”);
- 0.90: every word of the shorter name (at least two words) appears in the other (“Jean Tremblay” / “Jean-Pierre Tremblay”); 0.85: a one-word name of at least three letters contained in the other;
- otherwise, a word-by-word average: each word is paired with the closest word of the other name. The closeness of two words is the mean of the Jaro-Winkler similarity (scaling factor 0.1, common prefix up to 4 characters, boost applied above 0.7) and the Levenshtein similarity (1 − edit distance ÷ length of the longer word). The average is weighted by word length and computed in both directions. Generic business words (construction, group, services, management, enterprises, holdings, Canada, Quebec…) count for 40%; numbers, such as a corporation number, must be identical; an initial (“J.”) scores 0.9 against a word starting with the same letter, 0 otherwise, and each word weighs at least three characters.
Default threshold: 0.85, adjustable from 0.70 to 1.00. Scores measured by the tool’s tests: “Jean Tremblay” / “Jean Tremblai” 0.94 (flagged); “Jean Tremblay” / “Jeanne Tremblay” 0.92 (flagged); “Jean Tremblay” / “Marie Tremblay” 0.76 (flagged only if you lower the threshold to 0.75); “Construction ABC inc.” / “Construction XYZ inc.” 0.67 (not flagged). A score of 0.97 or more shows in red, others in amber.
The search is tuned not to miss a close name: it produces false positives, which you dismiss. It cannot find a name that is not in the registry, nor a relationship nobody recorded (parent company, spouse, nominee).
What the tool does, and what it does not
- It searches and it records. Every check is written to the local log: date, time, person, names checked, threshold, algorithm version, results, then the decision you enter.
- It does not decide whether there is a conflict. A similar name is not a conflict, and no result rules nothing out: the assessment is your professional judgment under the rules of the jurisdictions where you are admitted.
- It only sees this browser. The registry is shared neither among colleagues nor among devices. To check a whole firm, consolidate the exports (merge on import).
- Nothing is sent. No account, no server: the registry is kept in this browser’s IndexedDB storage.
Backup, transfer and encryption
The JSON export contains everything (registry and log). The CSV export contains the registry, one row per party, with the columns matter_ref, matter_title, opened, closed, matter_notes, party_name, party_kind, party_role, party_aliases, party_related, party_notes. Import accepts JSON, encrypted JSON and CSV, merging or replacing.
The encrypted export uses the browser’s Web Crypto interface: the passphrase becomes a key through PBKDF2 (HMAC-SHA-256, 600,000 iterations, random 16-byte salt), then the content is encrypted with AES-GCM (256-bit key, random 12-byte initialization vector, 128-bit authentication tag). The file holds the salt, the vector and the ciphertext, never the passphrase. A forgotten passphrase cannot be recovered; an altered file is refused on import.
Sources
The American Bar Association’s Model Rules of Professional Conduct are a model: each state adopts its own rules of professional conduct, and its version is the one that governs you. The three rules below are described in general terms, not quoted; read their text on the ABA site and in your state’s rules. The addresses were confirmed in a search index on November 6, 2026; the ABA site refuses automated reading, so the pages themselves were not opened by our verification tooling.
- ABA Model Rule 1.7, Conflict of Interest: Current Clients · concerns representation that is directly adverse to another current client, or that risks being materially limited by duties to others or by the lawyer’s own interests, and the conditions under which a client’s informed consent may allow it.
- ABA Model Rule 1.9, Duties to Former Clients · concerns acting against a former client in the same or a substantially related matter, and the use of information from the former representation.
- ABA Model Rule 1.10, Imputation of Conflicts of Interest: General Rule · concerns when one lawyer’s conflict extends to the other lawyers of the firm, including screening.
For Canada, see the Canadian version of the tool, which quotes the Quebec Code of Professional Conduct of Lawyers: Conflict-of-interest check (Canada).
Other tools
- Time tracking and invoicing · timers, matters, printable invoice
- Case-citation linker · reporter citations linked to CourtListener, table of authorities
- All tools for legal professionals