/confidence

What supports this model?

Inspect the recorded sources, judgments, decision frontier, history, and limits without turning a ranking into certainty.

Open source model
  1. How much confidence is actually supported?

    Probability distribution unavailable · calibration unavailable

    This model projection supplies no typed provider probability distribution or evaluated calibration result. Its ranking score is not a probability. No overall confidence number can be established from these records.

    No outcome forecast is supplied by the available decision records.

    Why is this question here?

    Source order · no numerical ranking receipt supplied

    This view preserves the order supplied by its source adapter. Its position is not a probability or a measure of truth.

    How is this answer supported?

    There is no Jev integration or invented confidence threshold on this page. A future provider response would require its question, criteria, model version, source revision and typed result to be retained first.

    Inspect Yawn's model of Jev

    This part has not been completed.

    Cost not estimated.
  2. Which model and sources are we inspecting?

    public · revision sha256:c93f01a19ec46d71199bdad0f54f14b3cd4a297251e19ae90df28a0dd1d8c494

    Jev · typed judgments

    Inspect the source contract, select saved text and explicitly approve one typed judgment. Configuration, authenticated metadata, retained inference and independent calibration remain distinct.

    Coordinate
    yawn.bot/typesafe-ai/jev
    Revision
    sha256:c93f01a19ec46d71199bdad0f54f14b3cd4a297251e19ae90df28a0dd1d8c494
    • YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn · sha256:c93f01a19ec46d71199bdad0f54f14b3cd4a297251e19ae90df28a0dd1d8c494
    • https://docs.typesafe.ai/primitives · source revision unavailable
    • https://docs.typesafe.ai/confidence · source revision unavailable
    • https://docs.typesafe.ai/concepts/system-one · source revision unavailable
    • https://docs.typesafe.ai/introduction · source revision unavailable
    • https://docs.typesafe.ai/api · source revision unavailable
    • https://docs.typesafe.ai/models · source revision unavailable
    • https://docs.typesafe.ai/introduction/quickstart · source revision unavailable
    • https://github.com/typesafe-ai/typesafe-sdk-js/blob/main/src/resources/models.ts · source revision unavailable
    • https://docs.typesafe.ai/llms.txt · source revision unavailable
    • https://docs.typesafe.ai/images · source revision unavailable
    • https://docs.typesafe.ai/primitives/box · source revision unavailable
    • YAWNS/Product/Leaderboard/index.yawn · sha256:f8c87eb8b44fc6efce7f7e707468cfb2ab4514c3e0b17fbe18a2d2615dbad7fc
    Why is this question here?

    Source order · no numerical ranking receipt supplied

    This view preserves the order supplied by its source adapter. Its position is not a probability or a measure of truth.

    Open source model
  3. What questions and judgments are recorded?

    Recorded source content; no new interpretation

    What does TypeSafe say Jev does?

    Provider documentation and consumer boundary · reviewed 2026-10-16

    TypeSafe describes focused typed judgments. Its current Images guide supports images with Jev 1.13 and 1.14, correcting this reference's September text-only finding. The System One overview and Models table still describe text; that documentation inconsistency remains visible. Direct audio and video are not supported. Outputs remain structured judgments, not generated explanations.

    The Confidence panel deliberately accepts selected saved text only and excludes attachments. Its explicit choices are jev-1.14.0 and jev-1.13.0. At the October 16 review, jev-latest still targeted 1.13.0; retained calls use pinned versions.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn https://docs.typesafe.ai/introduction https://docs.typesafe.ai/models https://docs.typesafe.ai/concepts/system-one https://docs.typesafe.ai/images

    Answer or correct at the source

    Which answer shapes does it expose?

    Provider description · reviewed 2026-10-16

    Choice returns an option, its full option distribution and confidence. Score returns a score, level legend, full distribution and confidence. Noul gives the probability of yes; it has no separate confidence field. Experimental Box returns normalized bounding edges and confidence, requires Jev 1.14, and has no not-found answer. Pair Box with a presence Noul. These are provider contracts, not observed Yawn results.

    The current Company and Confidence consumers send text Choice, Score and Noul only. The Forman Brain manifest contract can represent experimental images and Box, while its supplied text transport declines those requests.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn https://docs.typesafe.ai/primitives https://docs.typesafe.ai/primitives/box

    Answer or correct at the source

    Can it inspect images now?

    Provider experimental capability · no Yawn execution

    Yes, experimentally. Images and Box require an explicit per-request TypeSafe-Danger-Zone-Images header. Images can occur in state, instructions or criteria. Still PNG/JPEG only: each side 280 to 1568 pixels, at most 16 MiB per file, with additional request and expanded-image limits. Invalid images fail the request. The Forman Brain manifest contract validates image declarations; its current text transport does not decode, upload or inspect an image.

    A future server adapter must verify actual bytes, orientation, encoded and expanded sizes before sending.

    A timeline of selected still frames is not native video processing.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn https://docs.typesafe.ai/images

    Answer or correct at the source

    What does its confidence number mean?

    Provider formula correction · reviewed 2026-10-16

    The current documentation specifies its formulas, superseding the September finding that the reviewed page does not specify its formula. Choice rescales the selected probability above an even split: (pMax - 1/n) / (1 - 1/n). Score compares probability-weighted distance from the modal level with uniform mean absolute deviation, clamped at zero. These distribution statistics are separate from correctness, ranking and permission. No application threshold is adopted here.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn https://docs.typesafe.ai/confidence

    Answer or correct at the source

    Does calibration establish that one answer is correct?

    Provider claim · local evidence unavailable

    The provider describes calibration across groups of predictions and explicitly distinguishes it from an individual answer being correct. Yawn has no retained Jev evaluation population or outcomes in this reference, so its suitability for a Yawn decision remains unverified.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn https://docs.typesafe.ai/concepts/system-one

    Answer or correct at the source

    What does a judgment retain?

    Bounded application implementation · no independent calibration result

    The Confidence panel retains the exact question and criteria, selected source IDs and hashes, source and binding revision, approved pinned model, full typed response and supplied distribution/confidence, token usage and one-call owner approval in the existing durable orientation run ledger. The raw authorized text request is retained there too. It never converts this output to an orientation command, completed goal or action grant. Human decisions, corrections and later outcomes remain separate.

    No probability vector or provider confidence is reconstructed from a rank score or prose. Noul has no extra confidence field.

    Retained means stored against this source basis, not accepted as canonical truth. A changed basis withholds the result as current; the same request key cannot start another call.

    Company purpose and saved-move reviews use the existing Forman Brain request and receipt contract. Saved reviews share yawn_orientation_model_runs; unsaved draft review has a five-minute process-local replay cache and no durable persistence. These are different admitted operations, never one another's receipts.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn src/domain/yawn-jev-judgment.ts src/domain/forman-brain-judgment.ts src/lib/company/company-jev-service.ts

    Answer or correct at the source

    Is Jev connected or running here?

    Runtime readiness and explicit owner-send required

    Inspecting this reference performs no Jev call. The verified Dave owner Confidence panel may issue an authenticated model-list GET on mount; that checks metadata authentication without testing inference. This panel requires a configured server key, the existing capture-execution flag, durable run storage and an explicit one-call press for inference. Company's separate purpose review requires its own explicit authenticated owner press and exact context. Both use the same server-only credential; neither starts by opening Story. Current readiness and the retained receipt establish what actually happened; the public source file cannot establish those live facts.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn

    Answer or correct at the source

    How does this connect to Company and Forman Brain?

    Separate consumers over existing source, approval and run-ledger seams

    Company uses the Forman Brain typed contract to compare its current purpose, questions and candidate moves. Its explicit Ask Jev control reviews that exact context with pinned jev-1.13.0. A saved review uses the established owner-scoped run ledger; a draft review is not persisted and duplicate protection is limited to its current server process. A returned suggestion never chooses or executes the move. The Confidence panel instead admits one owner-selected saved-text question with its own revision-bound receipt.

    Sources: YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn YAWNS/Product/FormanBrain/index.yawn YAWNS/Product/FormanBrain/jev-image-research.yawn src/lib/company/company-jev-service.ts src/app/api/yawn/jev/route.ts

    Answer or correct at the source

    Recorded model fields

    Product constraint · recorded
    Authorize the source model before projecting its title, questions, sources, history or ranking.
    YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn
    Product constraint · recorded
    Show only recorded judgments. Missing probabilities, calibration, thresholds and outcomes stay unavailable.
    YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn
    Product constraint · recorded
    Keep a stated forecast credence, a ranker score, a provider confidence statistic and observed correctness separate.
    YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn
    Product constraint · recorded
    Corrections return to the existing source and feedback workflow; this read view creates no saved correction.
    YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn
    Product constraint · recorded
    No Jev request, model spend, reservation or execution is authorized by inspecting confidence.
    YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn
    Why is this question here?

    Source order · no numerical ranking receipt supplied

    This view preserves the order supplied by its source adapter. Its position is not a probability or a measure of truth.

  4. What puts one next move above another?

    Decision coverage: unavailable

    Authored reference order; no empirical confidence or decision ranking is implied.

    No authorized decision frontier is available in this projection.

    Why is this question here?

    Source order · no numerical ranking receipt supplied

    This view preserves the order supplied by its source adapter. Its position is not a probability or a measure of truth.

    Open source model
  5. What corrections and results are retained?

    History coverage: partial

    This is a partial history, not a complete account of how the model was produced.

    1. sourced_reference · 2026-09-24T00:32:46Z · agent:openai:codex

      Authored a public source-derived reference and product constraints. This is no provider run or independent calibration result.

      YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn
    2. source_revision · 2026-10-17T01:42:38Z · agent:openai:codex

      Reviewed current provider docs and connected a bounded owner-approved interface. Configuration, deployment and an actual inference remain separate runtime facts.

      YAWNS/Product/YAWN_CONFIDENCE_JEV_REFERENCE_2026_09_23.yawn

    A history event is retained as its recorded kind. It does not by itself establish that a correction changed a later result.

    Why is this question here?

    Source order · no numerical ranking receipt supplied

    This view preserves the order supplied by its source adapter. Its position is not a probability or a measure of truth.

  6. How can I challenge or correct this model?

    Return to the existing source workflow

    Open the source model to answer its questions or use its existing correction and feedback controls. A correction is not saved by opening this inspection page. Return here and refresh sources to inspect the resulting revision and available history.

    Why is this question here?

    Source order · no numerical ranking receipt supplied

    This view preserves the order supplied by its source adapter. Its position is not a probability or a measure of truth.

    Correct at sourceJev reference and documented limits

    This part has not been completed.

    Cost not estimated.