Interface: ViterbiBiases

Defined in: types/privacyFilter.ts:22

Six Viterbi transition biases that match the operating-point schema from the openai/privacy-filter viterbi_calibration.json. Each value is added to the decoder score whenever the corresponding BIOES transition is taken.

Positive values encourage the transition; negative values discourage it. Defaults are zero (neutral validity-only Viterbi).

Properties

backgroundStay?

optional backgroundStay: number

Defined in: types/privacyFilter.ts:24

O -> O (background persistence). Higher = stay in background more, fewer false positives.

---

backgroundToStart?

optional backgroundToStart: number

Defined in: types/privacyFilter.ts:26

O -> B-* / S-* (span entry). Lower (negative) = enter spans more eagerly, higher recall.

---

endToBackground?

optional endToBackground: number

Defined in: types/privacyFilter.ts:28

E-/S-* -> O (span closure to background).

---

endToStart?

optional endToStart: number

Defined in: types/privacyFilter.ts:30

E-/S-* -> B-* / S-* (back-to-back spans).

---

insideToContinue?

optional insideToContinue: number

Defined in: types/privacyFilter.ts:32

B-/I-X -> I-X (span continuation). Higher = longer spans.

---

insideToEnd?

optional insideToEnd: number

Defined in: types/privacyFilter.ts:34

B-/I-X -> E-X (span closure). Higher = shorter spans.