docs
    contracts: Changelog
    v2.30.0.md

    Changelog

    2.30.0

    • Add fixed recovery reasons distinguishing an ineligible original review from a blocking unpublished preparation that requires its owner's action.
    • Add an exact prior-review reporting recovery admission and safe public result. The native policy effect and original activity record have separate completion states.
    • Add bounded original-request discovery and UUID preparation reads, with native timestamps and separate preparation and policy action identities.
    • Preserve native preparation creation timestamps, including numeric offsets and six fractional digits, in both original read shapes.

    2.29.0

    • Preserve native concurrency timestamps, including numeric timezone offsets and all six fractional digits, in both delegated claim and original policy recovery requests.
    • Add an original preparation lookup contract for inspecting a saved request without starting or resuming its work.
    • Add explicit unpublished-preparation retirement request and receipt contracts, preserving bounded retained and unrecorded inventory.
    • Add bounded Google campaign diagnostic reports with separate section coverage, nullable metrics and untrusted provider text.
    • Add explicit diagnostics request and original action payload parsers, with complete ad group criterion identity for keyword text.
    • Add the delegated-command entry with bounded owner-controlled execution policies, safe capability observations, four named Resonance effect families and truthful native agent attribution.

    2.28.0

    • Add an optional signed reporting deadline for late attribution reads after a campaign's original flight. The server derives the deadline from approved attribution while finance and write authority keep their existing expiry.
    • Add a standing operating policy to Resonance's existing mandate. The policy names the agent and verified principal, and limits Google Search changes to specific reviewed targeting and ad copy.
    • Add the Google Search content update selector and command profile. Preparation names one existing ad. The command carries its exact approved text changes, current policy receipt and prior content fingerprint.
    • Key an operating policy approval to the accepted campaign and its exact predecessor receipt so that returning to earlier terms requires a fresh approval.

    2.27.0

    • Add the booking-intake entry: what a booking may ask a guest about their business. A booking type's step is off, optional or required (parseBookingBusinessStep reads anything unknown as off), and checkBookingBusinessAnswers is the one check the booking widget, the app's booking page and the API all run, with the guest-facing words in BOOKING_BUSINESS_ERROR_TEXT. normalizeBookingCompany removes control and format characters (bidi overrides, zero-width marks) and lone surrogates and never truncates; normalizeGuestWebsite returns a public http or https address with no query, fragment, port or credentials, keeps the path, and refuses IP addresses, the literal local names localhost, .local and .internal, and other schemes (it runs no DNS, so anything that fetches the address must check it itself, and the url must be HTML-escaped wherever it's rendered). A required company is refused for every source except portal with verifiedMember: true, which the API sets from its authenticated context and never from the request body; a missing or unknown source is held to the answers. bookingBusinessDetails is the one reader for the answers on every surface a host sees, and BOOKING_INTAKE_JSON_SCHEMA describes the three wire fields for an agent. The browser-safe part is its own module so the widget doesn't bundle the schema or the reader. stripWww moved into the sites leaf module and is still exported from sites.
    • Let a qualified-lead goal count contacts who book. The reviewed objective and the native qualification witness can each name up to 100 eligibleBookingTypeIds beside the one managed form, and the two lists must match exactly, in any order or letter case. A witness that adds or drops a booking type, a duplicate or empty list, and booking types on any other goal source are refused. A goal without the field serializes and digests exactly as before.
    • Add the resonance-confirmation entry: the one wire contract for the emailed code behind a budget approved on a person's word. It holds the two routes, the closed request and answer parsers, the six-key on_their_word evidence parser, the refusal table with its status codes, the code numbers (six digits, ten minutes, five tries, three an hour) and the marker that identifies the confirmation mail (subject tag and header). The shape of stored evidence is read here; whether a code was really redeemed stays the database's word. The receipt's expiresAt is the real deadline (the shorter of ten minutes and the prepared plan's own expiry), a prepared plan with under two minutes left is refused as PLAN_EXPIRING before any code is spent, a person is sent at most 3 codes an hour per project, 6 an hour and 15 a day across projects, and attemptsLeft travels under data (the one field the server's error filter forwards); a wrong code with no tries left reads as the lockout. The rate limits are RESONANCE_CONFIRMATION_RATE_HOURLY (3 an hour for a project, 6 an hour for a person) and RESONANCE_CONFIRMATION_RATE_DAILY (15 a day); the single RESONANCE_CONFIRMATION_RATE stays in the table because the first confirmation migration still answers with it. RESONANCE_CONFIRMATION_PERSON_LOCKED (429) is the answer when a person's agent has had three codes locked in a day. The rate-limit answers carry how long to wait as retryAfterSeconds under error.data (1 to 86,400), which confirmationFailureOf returns beside the failure kinds rate_limited_daily and person_locked, and confirmationWaitWords words it ("about 13 minutes", rounded up). CONFIRMATION_LOCKS_PER_DAY and CONFIRMATION_RETRY_MAX_SECONDS are the numbers.
    • Add the quote-screen entry: the one screen for the quote a person's yes carries. A quote over 1,000 characters is refused before anything reads it, then provider keys, tokens, logins, passwords, PINs, card numbers (Luhn plus a brand prefix), IBANs (mod-97) and routing, account and social security numbers are refused, while campaign names written with words, abbreviations, all-caps or separators pass. Unseparated names of 24 or more characters are refused by design and listed as known limits.

    2.26.0

    • Add positioning previews for exact Meta, Google Search and Broadcast draft targets. Current source witnesses and an independent consistency assessment support the owner's application checks.
    • Add the property-interest entry with closed event and site report contracts. Exposure-normalized action rates, missing-session events and accepted tour inquiry evidence retain separate fields. The request matcher checks the exact site, window and explicit listing set.

    2.25.0

    • Add guided Meta dataset selection and setup contracts, including exact verified dataset readback and a shared dataset name validator.
    • Add audience setup review for one saved rule and an explicit opt-in for future eligible visits. Keep the verified audience receipt separate from the revocable visit grant.
    • Add combined readiness and paginated recovery contracts. Accepted visit counts and provider audience readiness remain separate observations.

    2.24.0 (2026-11-02)

    • Align UUID JSON Schema patterns with the existing runtime acceptance of lowercase, uppercase and mixed spelling in call preparation, Resonance and experiment measurement. Keep UUID versions 1–8 and existing variant restrictions, with original identity text preserved.

    • Add the dependency-free sales-call-prep entry for bounded call guides, captured objective/style, immutable version and explicit use/correction receipts. Keep resolved context and native provenance separate from model content. Add private observational learning summaries and deterministic server hash-input serializers without changing existing ledger approval controls.

    2.23.0

    • Add optional per-channel providerReporting.metaQueries to the existing report selector and resolved grant. Each explicit query fixes its attribution window and conversion report time. Original UUID text, list order and omitted legacy fields stay unchanged; actual provider membership, original review and admission remain native checks.
    • Add the closed ResonanceMetaRetargetingReadiness response parser, type and JSON schema. It keeps saved configuration facts separate from an observed audience check, preserves unknown size and size freshness, and carries no launch authority. The API classifier and consumers share one exported verdict and reason vocabulary.

    2.22.0

    • Add one additive set of rules for retargeting several properties from one campaign. A Meta channel can now carry an optional channelGroup, property, tier, primaryText and convertersAcknowledged. Channels in one group must share their site, account, dataset, Page, categories, countries, objective and optimization, each property keeps its own destination, and tiers of one property can't overlap in time. A grouped campaign's allocations can't exceed its total less its reserve, and a special category in a group requires provider-managed age. A definition that uses none of these fields parses and serializes exactly as before.
    • Add checkChannelGroups, splitBudget, suggestPropertyWeights and expandPropertyChannels. The split conserves the total in whole provider quanta, the suggestion keeps every share between a floor and a ceiling, and the expander builds new property channels from the group's first channel with deterministic ids. It never sets a creative, a converter acknowledgement or an approval, and expanding twice is a no-op.
    • Add an optional minDaysSinceVisit recency band to an audience reference and to the audience draft, checkAudienceBand, and the exported META_WEBSITE_AUDIENCE_MAX_RETENTION_DAYS (180), so the editor, the API and the contract share one number. Meta audiences hold at most ten rules, and Google audiences take no band.
    • Add metaExclusionPolicy and META_SAC_CUSTOM_AUDIENCE_EXCLUSION_VERIFIED, which ships false: targeting-level exclusion under a special category stays refused until a reviewed release proves it. Add META_CALL_TO_ACTIONS, the one list the editor and compiler validate against.
    • Raise the public observation page bound from 300 to 500 metrics, because one provider row can now expand to five (spend, impressions, clicks, reach and results). The bound of 100 native rows is unchanged.
    • Add the provider metrics reach and results, the counting unit provider_unique_reach (reach only, never summed across channels), and deriveProviderRatios, the one place frequency and cost per result are computed. Zero, missing and mixed inputs are unknown with a reason, never zero.
    • Add three optional fields to the Resonance measurement configuration: metaEligible on a page resource rule (absent means false), and browserWitness and tagManagerAttested on a Meta destination. Configurations that omit them parse as before and mean no Meta browser loading.

    2.21.0

    • Add optional, closed page.experiment authoring to managed Funnel drafts and definitions, with shared headline/body bounds and variant checks. Keep authored selection separate from the existing native-derived publication manifest. Plain documents stay compatible; older strict parsers reject extended documents, so consumers must adopt together before exposing authoring.

    • Refuse observed provider metrics paired with unestablished support while retaining unknown per-metric values on complete query snapshots.

    • Add an optional original read-only provider-reporting grant and a separate closed draft selector. Bind server-resolved scope, agent, principal and paid channels into the existing mandate review/digest while preserving old absent-field bytes. Native identity, original review and current admission remain separate authority checks.

    • Add closed public observation item, support and page parsers with explicit native identity alignment, mandatory nonfinal provider flags and truthful first-party/unknown metrics. Preserve native-row pagination and the generic metric and optimizer contracts.

    • Add one explicit SQL pre-read clock policy and evidence branch for conditional reported-cost allocation. Preserve immutable report provenance and independent outcome maturity without cross-clock freshness assumptions. Existing mature, finite and default profiles stay compatible; original native authority, expiry and funding remain separate requirements.

    2.20.0

    • Publish the resonance, resonance-constants and signal-experiments entries with matching CommonJS, ESM and TypeScript surfaces. Add the documentation manifest used by sonor.dev. These validation contracts grant no campaign approval or spending authority.

    • Bind current Meta configuration review to the existing semantic definition digest plus a nondecreasing audit revision. Add required native currentDefinitionDigest metadata so metadata-only saves can preserve multiple unchanged channel configurations. Original snapshots stay immutable; exact prepared-plan, approval and native root fences remain unchanged.

    • Add pure native measurement bootstrap save/current/snapshot/response contracts and one deterministic page projection using existing semantic digest bytes. Native page resources retain UTC millisecond revisions; rule revisions use full existing evidence digests. Project/site CAS, actual protected writer and one read-only native freshness resolver remain separate adoption requirements.

    • Add pure Meta configuration save/snapshot/view/response contracts, exact existing-audience or admitted-setup mappings, immutable friendly review material, shared digest/projection helpers, minimal optional mandate feed selection and post-acceptance original feed review. Preserve the raw existing mandate-draft response and no-feed request bytes. Saved profile references require actual persisted operation IDs and integer configuration revisions; parsing grants no native capability or approval.

    • Add closed explicit provider_managed Meta age intent alongside unchanged absent, omitted and bounded behavior. Actual first same-root PAUSED creator observation and unchanged activation comparison remain native owner work; existing consumers must explicitly support the new branch before adoption.

    • Add an optional original Meta website-visit feed grant and a closed internal deliver_website_visit command through the existing Resonance parser and digests. Preserve omitted legacy fields and require explicit native feed authority; parsing doesn't authenticate a producer or permit manual origination.

    • Add same-occurrence reported path and Meta cookie observation shapes, one fixed SDK observation policy and canonical semantic configuration/profile digest helpers. Add original-time checks with an inclusive five-minute local freshness bound and zero future allowance. These are Sonor rules, not Meta acceptance guarantees; native capture, consent, admission and delivery remain separate.

    • Represent explicit omitted or bounded original Meta age selection. Absent fields retain legacy bytes and don't choose defaults. Actual policy, compiler and normalized readback support remain adapter responsibilities.

    • Add a closed Google qualified-conversion delivery command and explicit original mandate operation. Reference one original Google channel and a persisted conversion ID. Optional plan, mandate and time context checks public grant consistency and the original prepared definition revision; private facts, native receipts, current consent and final admission remain server-owned. Existing launch grants never expand automatically.

    • Add closed finite Meta candidate policy, lifecycle evidence and one-action decision variants through the existing Resonance entry. Reference original candidate channel IDs, preserve mature allocation, and bind later graph evidence without a digest cycle. Fixed exploration explicitly has no performance-decision policy. Optional actual evidence/plan arguments check cross-object scope and action consistency; native funding and admission remain server-owned.

    • Add optional, closed clickIdentifiers to page-view measurement occurrences. Accept exactly one bounded gclid, gbraid or wbraid; other occurrence kinds reject it. Export its parser, type and JSON Schema. Identifiers describe observed lineage and grant no conversion or provider authority.

    • Add strict Google configuration draft, snapshot, view and review contracts, one canonical projection and a digest-input helper. Prepared Google Search previews require either actual existing-goal evidence or bounded one-action setup intent for creation after the same mandate; other channels use explicit null. Require the explicit historical qualified_at_observation_v1 provider signal profile, distinct from current Sonor outcome counts. Draft declarations stay bound to their campaign and creative scope; parsing grants no provider or approval authority.

    • Add strict command-preparation request and response parsers. Preparation reports existing mandate or current owner approval requirements and never grants authority. Add @sonordev/contracts/resonance-constants for browser wire literals without campaign schemas.

    • Add canonical Signal experiment create/update parsers and bounded Resonance audience/experiment management commands. Add reviewed public source-marker fields and the explicit first-party last-touch attribution model. These parsers validate wire data and grant no approval or provider authority.

    • Add @sonordev/contracts/resonance: strict campaign, account capability, prepared-plan, mandate lookup, command, receipt, measurement and consent boundaries for connected campaign operations. Money uses exact decimal-string micros with checked provider-unit conversion. Parsing validates data and scope; it never grants advertising or spending authority.

    • Retain the 2.19 review-photo presentation exports and all existing contracts.

    2.19.0

    • Add an optional per-review photo presentation to @sonordev/contracts/broadcast-layouts: REVIEW_PHOTO_PRESENTATIONS, ReviewPhotoPresentation and isReviewPhotoPresentation support the explicit background choice. Input consumers handle an omitted choice by preserving the saved value and null by clearing it to automatic. Existing automatic layouts, composition seeds and variation draws are unchanged.
    • Keep the decision-role entry and all other published contracts behavior unchanged.

    2.18.0

    A minor release: one new entry, @sonordev/contracts/decision-roles. Every other entry is unchanged.

    • What a job title says about a buying decision. DECISION_ROLES is the four parts a person can play (owner, decision_maker, influencer, gatekeeper), strongest first, and decisionRoleForTitle(title) returns the one a title names, or null when it names none. The role comes from the title alone, by eight rules read in order with the first match deciding, so "Vice President of Sales" is a decision maker, "Executive Assistant to the CEO" is a gatekeeper and "Marketing Coordinator" is an influencer. "Principal" is an owner unless it names a job, so "Principal Engineer" has no role. A title is read whole (it isn't split on "&" or "/") and only its first 200 characters count. DecisionRole is the type. Anything that finds the people on a company's website can import it, so two scans of the same page can't give one person two roles.