Harden the core privacy guarantee:
- Add `strict` mode (default true): after detection, verify no mapped value
survives as a whole token in the output (ignoring placeholders, whose context
may legitimately echo a value like `B+`). Catches a model that redacts one
mention of a value but leaves another in clear — which the placeholder/mapping
bijection check missed. Fail-closed: throws AnonymizationError naming only the
non-secret placeholder key, and runs on the final result so it is not swallowed
into the regex fallback (which can't fix a name leak). Set strict:false to opt out.
- Add `prefilter` option (default true): decouple the PII pre-filter from the
presence of a regex fallback. Set false to always consult the LLM while keeping
the fallback for LLM failures (max recall + graceful degradation).
- Boundary-aware value substitution: applyKnown and the leak check now match
values only as whole tokens (Unicode letter/digit boundaries, regex-escaped),
so "Ann" no longer replaces inside "Anna" and "jean@x.ch" no longer matches
inside "jean@x.church"; accented/non-Latin names preserved.
- deanonymize restores longest placeholder keys first (prefix-overlap defense).
Restructures anonymize/anonymizeChunks/anonymizeTurn to a single exit so the
leak check runs once on the final result. Behavior is unchanged for callers that
were already leak-free.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Enables consuming the package as a git dependency (e.g. from Gitea): npm runs
'prepare' on install, which builds dist/ (git-ignored).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each example now shows the secret `mapping` (placeholder → real value) as a
table alongside the `legend`, and the intro states the results were produced
via Infomaniak's Swiss AI API (Gemma 4).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Each example now shows the shareable `legend` as a table (Abbreviation →
meaning), marking 🆕 the abbreviations the model coined for attributes absent
from the base prompt (ASSUR_ID, PERMIS, PASSPORT, RCC, BIO_ID, SANG, DOSSIER…).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- examples/demo.mjs: live demo (5 single messages up to 5 people with rare
auto-coined attributes + 5 escalating multi-turn conversations with
long-range back-references); self-references the package via exports
- examples/RESULTS.md: its generated output (10 examples)
- examples/.env.example + examples/README.md; `npm run example` script
- README: replace the long examples section with two representative examples
(dense 5-person message + 6-turn conversation) and a badge/link to
examples/RESULTS.md for the full set
- eslint ignores examples/**, prettier ignores generated examples/RESULTS.md,
.gitignore un-ignores .env.example
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Scripted, deterministic conversations (fictional Swiss e-learning chats) via a
mock provider, each asserting exact per-turn round-trips + stable entity ids:
login support (attributes attach to PER_1), two learners (applyKnown reuse of a
re-mentioned name), old/new IBAN kept distinct, parent+minor+doctor (3 people),
and HR (manager+org reused, two employees). 46 tests total.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
anonymizeTurn(text, session) threads a serializable AnonymizerSession
({mapping, legend, history}) so one entity keeps one id across a whole chat
(applyKnown reuse + usedIds + de-collision merge). Adds conversation()
in-memory wrapper and an optional LlmProvider.anonymizeInConversation(text, ctx)
for rich cross-turn context; providers without it fall back to the batch path.
openAICompatibleProvider gains includeMappingInContext (default false — only
send real values to a trusted anonymizer endpoint) + historyMaxTurns (default
10). Verified live vs Gemma 4: Nora stays PER_1 across 4 turns (name/email/AVS/
IBAN), Yanis = PER_2; 41 tests, 98% coverage.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two coherence rules added to DEFAULT_SYSTEM_PROMPT:
- different values of the same type get distinct placeholders (context),
e.g. old/new IBAN -> [PER_1.IBAN:Ancien] / [PER_1.IBAN:Nouveau]
- never absorb adjacent punctuation/separators into a placeholder
Took the live Gemma-4 (Infomaniak) e-learning demo from 4/5 to 5/5 exact
round-trips. 36 tests passing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
- response_format is omitted by default (Infomaniak rejects the legacy
json_object → HTTP 422); opt in via the new `responseFormat` option
({type:'json_object'} or a json_schema object). BREAKING for endpoints
that relied on the forced json_object.
- parse LLM JSON leniently (tolerate markdown fences / surrounding prose)
- fold strict-coherence rules into DEFAULT_SYSTEM_PROMPT (exact
placeholder<->mapping-key identity, values are originals, strict format,
mask value not adjacent label) → reliable output across models
Verified live against Gemma 4 (google/gemma-4-31B-it, Infomaniak v2): all
demo phrases anonymize with clean round-trips, no custom provider needed.
36 tests passing.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
BREAKING CHANGE: regex fallback is now opt-in (patterns no longer defaults
to presets.swiss). With no fallback an LLM failure throws AnonymizationError
(fail-closed) and the pre-filter is bypassed; at least one of llm/patterns is
required. AnonymizationResult gains a required `legend`; anonymizeChunks seed
is now { mapping, legend? } and returns legend.
- prompt: model may coin new UPPERCASE abbreviations and returns a 'legende'
explaining every abbreviation used (French); backfilled by DEFAULT_LEGEND
- PatternDef.meaning surfaces in the legend; swiss/generic presets get meanings
- AnonymizationError (exported) wraps the cause on fail-closed
- README: drop the chatbot provenance line; add 'How it works' + nLPD sections
- 34 tests / 99% coverage; bump to 0.2.0
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
