@opacedev/astro-ai-content-checker 0.3.1

package/node_modules/@opacedev/ai-content-checker-contracts/src/generated/method-result.schema.d.ts

@@ -0,0 +1,39 @@
+/* eslint-disable */
+/**
+ * This file was automatically generated by json-schema-to-typescript.
+ * DO NOT MODIFY IT BY HAND. Instead, modify the source JSONSchema file,
+ * and run json-schema-to-typescript to regenerate this file.
+ */
+
+export type HttpsSchemasOpaceAgencyContentIntegrityV1MethodResultSchemaJson = {
+  [k: string]: unknown;
+} & {
+  id: string;
+  category: "detector" | "watermark" | "provenance" | "unicode" | "pattern" | "fidelity";
+  provider_or_method: string;
+  version: string;
+  status: "pass" | "attention" | "fail" | "inconclusive" | "unsupported" | "not_configured" | "not_run" | "error";
+  availability?: "available" | "restricted" | "not_available";
+  native_outcome?: string;
+  score: number | null;
+  score_scale?: {
+    [k: string]: unknown;
+  } | null;
+  threshold: {
+    [k: string]: unknown;
+  } | null;
+  segments: {
+    [k: string]: unknown;
+  }[];
+  evidence: {
+    [k: string]: unknown;
+  }[];
+  /**
+   * @minItems 1
+   */
+  limitations: [string, ...string[]];
+  started_at: string;
+  completed_at: string;
+  privacy_route: "browser" | "wordpress_local" | "local_service" | "hub_provider" | "commercial_byok";
+  [k: string]: unknown;
+};

package/node_modules/@opacedev/ai-content-checker-contracts/src/generated/pattern-finding.schema.d.ts

@@ -0,0 +1,26 @@
+/* eslint-disable */
+/**
+ * This file was automatically generated by json-schema-to-typescript.
+ * DO NOT MODIFY IT BY HAND. Instead, modify the source JSONSchema file,
+ * and run json-schema-to-typescript to regenerate this file.
+ */
+
+export interface HttpsSchemasOpaceAgencyContentIntegrityV1PatternFindingSchemaJson {
+  rule_id: string;
+  rule_version: string;
+  severity: "note" | "low" | "medium" | "high";
+  message: string;
+  suggestion?: string;
+  span: {
+    start_utf16: number;
+    end_utf16: number;
+    start_codepoint: number;
+    end_codepoint: number;
+    [k: string]: unknown;
+  };
+  matched_text_hash: string;
+  evidence: {
+    [k: string]: unknown;
+  };
+  [k: string]: unknown;
+}

package/node_modules/@opacedev/ai-content-checker-contracts/src/generated/protected-span.schema.d.ts

@@ -0,0 +1,35 @@
+/* eslint-disable */
+/**
+ * This file was automatically generated by json-schema-to-typescript.
+ * DO NOT MODIFY IT BY HAND. Instead, modify the source JSONSchema file,
+ * and run json-schema-to-typescript to regenerate this file.
+ */
+
+export interface HttpsSchemasOpaceAgencyContentIntegrityV1ProtectedSpanSchemaJson {
+  id: string;
+  kind:
+    | "name"
+    | "organisation"
+    | "number"
+    | "date"
+    | "time"
+    | "currency"
+    | "unit"
+    | "url"
+    | "email"
+    | "quote"
+    | "citation"
+    | "code"
+    | "user_selected";
+  text: string;
+  start_utf16: number;
+  end_utf16: number;
+  start_codepoint: number;
+  end_codepoint: number;
+  normalised_value?: unknown;
+  policy: "exact" | "equivalent_format" | "must_remain_supported" | "user_review";
+  source: "deterministic" | "ner" | "citation_parser" | "user";
+  confidence?: number | null;
+  content_hash: string;
+  [k: string]: unknown;
+}

package/node_modules/@opacedev/ai-content-checker-contracts/src/generated/rewrite-request.schema.d.ts

@@ -0,0 +1,65 @@
+/* eslint-disable */
+/**
+ * This file was automatically generated by json-schema-to-typescript.
+ * DO NOT MODIFY IT BY HAND. Instead, modify the source JSONSchema file,
+ * and run json-schema-to-typescript to regenerate this file.
+ */
+
+export interface HttpsSchemasOpaceAgencyContentIntegrityV1RewriteRequestSchemaJson {
+  schema_version: "1.0";
+  contract_version: string;
+  request_id: string;
+  created_at: string;
+  source: Source;
+  protected_spans: HttpsSchemasOpaceAgencyContentIntegrityV1ProtectedSpanSchemaJson[];
+  candidate_count: number;
+  /**
+   * @minItems 1
+   */
+  allowed_routes: [
+    "browser" | "wordpress_local" | "local_service" | "hub_provider" | "commercial_byok",
+    ...("browser" | "wordpress_local" | "local_service" | "hub_provider" | "commercial_byok")[]
+  ];
+  checks: string[];
+  gate_policy: "strict" | "manual_review";
+  style_brief?: string;
+  [k: string]: unknown;
+}
+export interface Source {
+  content: string;
+  content_type: "plain_text" | "html" | "markdown";
+  language: string;
+  title?: string;
+  source_uri?: string;
+  content_hash?: string;
+  normalised_hash?: string;
+  [k: string]: unknown;
+}
+export interface HttpsSchemasOpaceAgencyContentIntegrityV1ProtectedSpanSchemaJson {
+  id: string;
+  kind:
+    | "name"
+    | "organisation"
+    | "number"
+    | "date"
+    | "time"
+    | "currency"
+    | "unit"
+    | "url"
+    | "email"
+    | "quote"
+    | "citation"
+    | "code"
+    | "user_selected";
+  text: string;
+  start_utf16: number;
+  end_utf16: number;
+  start_codepoint: number;
+  end_codepoint: number;
+  normalised_value?: unknown;
+  policy: "exact" | "equivalent_format" | "must_remain_supported" | "user_review";
+  source: "deterministic" | "ner" | "citation_parser" | "user";
+  confidence?: number | null;
+  content_hash: string;
+  [k: string]: unknown;
+}

package/node_modules/@opacedev/ai-content-checker-contracts/src/index.ts

@@ -0,0 +1,63 @@
+export const SCHEMA_VERSION = "1.0" as const;
+export const CONTRACT_VERSION = "1.0.0" as const;
+export const WORDPRESS_REST_NAMESPACE = "oaci/v1" as const;
+export const WORDPRESS_ADMIN_PAGE = "admin.php?page=oaci-lab" as const;
+export const LOCAL_ENGINE_ORIGIN = "http://127.0.0.1:8741" as const;
+export const PUBLIC_API_IDENTITY = "Opace\\ContentIntegrity\\Integration\\PublicApi::instance()" as const;
+export const READY_HOOK = "oaci_ready" as const;
+export const PHP_PUBLIC_API_METHODS = ["version", "is_compatible", "capabilities", "register_source_adapter", "create_session", "get_session", "approve", "get_approved_output", "mark_applied", "get_receipt", "asset_handles"] as const;
+export const JS_MOUNT_CONTRACT = {
+  global: "OpaceContentIntegrity",
+  apiVersion: "1.0",
+  requiredOptions: ["surface", "sourceRef", "sourceHash", "getContent", "onApproved", "onClose"],
+  returns: ["destroy", "refresh", "getState"],
+  events: ["oaci:statechange", "oaci:approved", "oaci:error"]
+} as const;
+
+export const METHOD_STATUSES = ["pass", "attention", "fail", "inconclusive", "unsupported", "not_configured", "not_run", "error"] as const;
+export type MethodStatus = (typeof METHOD_STATUSES)[number];
+
+export const PRIVACY_ROUTES = ["browser", "wordpress_local", "local_service", "hub_provider", "commercial_byok"] as const;
+export type PrivacyRoute = (typeof PRIVACY_ROUTES)[number];
+
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1EnvelopeSchemaJson as ContractEnvelope } from "./generated/envelope.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1MethodResultSchemaJson as MethodResult } from "./generated/method-result.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1AnalysisRequestSchemaJson as AnalysisRequest } from "./generated/analysis-request.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1AnalysisResultSchemaJson as AnalysisResult } from "./generated/analysis-result.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1ProtectedSpanSchemaJson as ProtectedSpan } from "./generated/protected-span.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1PatternFindingSchemaJson as PatternFinding } from "./generated/pattern-finding.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1GateResultSchemaJson as GateResult } from "./generated/gate-result.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1CandidateSchemaJson as Candidate } from "./generated/candidate.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1RewriteRequestSchemaJson as RewriteRequest } from "./generated/rewrite-request.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1JobSchemaJson as Job } from "./generated/job.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1CapabilitiesSchemaJson as Capabilities } from "./generated/capabilities.schema.js";
+export type { HttpsSchemasOpaceAgencyContentIntegrityV1IntegrityReceiptSchemaJson as IntegrityReceipt } from "./generated/integrity-receipt.schema.js";
+export type { OpaceAIContentIntegrityCheckerResult as CheckerResult } from "./generated/checker-result.schema.js";
+
+export interface AnthropicUnsupportedMethodResultV1 {
+  id: "watermark.anthropic";
+  category: "watermark";
+  provider_or_method: string;
+  version: "unavailable-2026-08-26";
+  status: "unsupported";
+  availability: "not_available";
+  native_outcome?: "not_available";
+  score: null;
+  threshold: null;
+  segments: Record<string, unknown>[];
+  evidence: Record<string, unknown>[];
+  limitations: [string, ...string[]];
+  started_at: string;
+  completed_at: string;
+  privacy_route: PrivacyRoute;
+  [key: string]: unknown;
+}
+
+export interface JsMountOptionsV1 {
+  surface: string;
+  sourceRef: string;
+  sourceHash: string;
+  getContent(): string | Promise<string>;
+  onApproved(payload: unknown): void;
+  onClose(): void;
+}

package/node_modules/@opacedev/ai-content-checker-contracts/src/semantic-validation.mjs

@@ -0,0 +1,32 @@
+const HASH_PATTERN = /^sha256:[a-f0-9]{64}$/;
+
+function result(valid, errors = []) {
+  return { valid, errors };
+}
+
+export function validateProtectedSpan(span) {
+  const errors = [];
+  if (span.end_utf16 <= span.start_utf16) errors.push("utf16_range_empty_or_reversed");
+  if (span.end_codepoint <= span.start_codepoint) errors.push("codepoint_range_empty_or_reversed");
+  if (typeof span.text === "string") {
+    if (span.end_utf16 - span.start_utf16 !== span.text.length) errors.push("utf16_length_mismatch");
+    if (span.end_codepoint - span.start_codepoint !== [...span.text].length) errors.push("codepoint_length_mismatch");
+  }
+  if (!HASH_PATTERN.test(span.content_hash ?? "")) errors.push("source_hash_invalid");
+  return result(errors.length === 0, errors);
+}
+
+export function validatePatternFinding(finding) {
+  const errors = [];
+  const span = finding.span ?? {};
+  if (span.end_utf16 <= span.start_utf16) errors.push("utf16_range_empty_or_reversed");
+  if (span.end_codepoint <= span.start_codepoint) errors.push("codepoint_range_empty_or_reversed");
+  if (!HASH_PATTERN.test(finding.matched_text_hash ?? "")) errors.push("matched_text_hash_invalid");
+  return result(errors.length === 0, errors);
+}
+
+export function validateContractSemantics(schemaName, value) {
+  if (schemaName === "protected-span.schema.json") return validateProtectedSpan(value);
+  if (schemaName === "pattern-finding.schema.json") return validatePatternFinding(value);
+  return result(true);
+}

package/node_modules/@opacedev/ai-content-checker-core/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Opace Ltd
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/node_modules/@opacedev/ai-content-checker-core/README.md

@@ -0,0 +1,186 @@
+# @opacedev/ai-content-checker-core
+
+![Free Opace AI Content Checker, Detector and Watermark Tools](https://raw.githubusercontent.com/OpaceDigitalAgency/opace-ai-content-checker-detector/main/docs/assets/opace-ai-content-checker-detector-hero-v3.png)
+
+Deterministic, offline content-integrity engine: invisible-Unicode and homoglyph forensics, named writing-signal rules with an editorial score, protected spans, safe-fix previews, diffs, gates and RFC 8785 hash-only receipts. This is the single analysis implementation used by every Opace surface (web checker, WordPress plugin, Chrome extension, Astro integration, CLI).
+
+- Consumes `@opacedev/ai-content-checker-contracts` contract `1.0.0`
+- Browser/SSR-safe ESM, transport-free by design: no network call, telemetry or postinstall download
+- Unicode carriers: 38 rules over 415 code points, with context exemptions for emoji, cursive/Indic scripts and French typography (`unicode:2026.08.2`)
+- Homoglyphs: 60 Cyrillic/Greek confusables with a mixed-script gate
+- Writing signals: 113 weighted categories (116 named rules, including 7 rhythm and cadence stylometrics and 3 chat-export formatting rules, measured, not calibrated to a clean sheet: on 1,200 human long-form documents these rules flag 24.8% of them, and any claim that they produce no false positives is withdrawn — see the measured rates below) at `en-signals:2026.08.6`, with a raise-only evidence-based escalation policy recorded in an explanatory `escalation` result field. 95 of the 116 fire on real documents: one is recorded inactive and twenty dormant, with the measured per-rule inventory in `tests/battery/rule-liveness.json`
+- Protected spans: 12 kinds including name, organisation and citation
+- `watermark.anthropic` remains visibly `unsupported` until an official interface exists; the browser checker runs a separate live known-key scan (`@opace/watermark-lab`) rather than presenting that boundary as a check
+- Licence: MIT for the Opace-authored package
+
+A pass applies only to its named disclosed check and does not prove authorship or detector clearance. The full capability inventory is in the repository's [capability register](../../docs/CAPABILITIES.md); every test total and evaluation behind it is indexed in the [evidence index](../../docs/EVIDENCE-INDEX.md).
+
+> Release state: a 0.1.0 npm candidate is prepared locally but is not published. The install command below applies only after owner-approved publication.
+
+## Install
+
+```sh
+npm install @opacedev/ai-content-checker-core
+```
+
+Targets ESM on modern browsers and Node.js 20 or newer.
+
+## API
+
+```js
+import {
+  inspect,                    // full analysis over an AnalysisRequest
+  computeEditorialSignals,    // 0-100 writing-signal score for plain text
+  inspectSignalsV2,           // raw signal findings
+  inspectUnicode,             // carrier/homoglyph findings for a string
+  extractProtectedSpans,      // 12 protected-span kinds
+  previewSafeFixes, diff, validateCandidate,
+  buildReceipt, verifyReceipt,
+  presentCycle5Result, composeCheckerAxes,
+  buildContentFreeSharePayload, assertCheckerResultInvariants,
+  listMethods, registerPatternPack,
+  projectVisibleText, prefixedSha256, sha256Hex, utf8Bytes,
+  UNICODE_RULES_VERSION, EN_SIGNALS_PATTERN_VERSION,
+} from "@opacedev/ai-content-checker-core";
+```
+
+The checker-result helpers are presentation/contract logic, not a bundled detector. They freeze the
+Cycle-5 identity and five `signal-*` level ids, format all scores once per run, preserve the three
+independent axes and reject model/route/section/export contradictions. A full-checker adapter supplies
+the measured model result; without one, `notAssessedAiPattern()` is the only honest AI axis.
+
+### Full inspection
+
+`inspect(request)` takes a contract `AnalysisRequest` and returns a frozen `AnalysisResult` with per-method statuses, evidence, limitations and hashes. Verified example (output from the built package):
+
+```js
+const result = await inspect({
+  schema_version: "1.0",
+  contract_version: "1.0.0",
+  request_id: "req_readme_demo",
+  created_at: new Date().toISOString(),
+  source: { content: "In today's rapidly evolving landscape, Opace quoted £1,250.​",
+            content_type: "plain_text", language: "en-GB" },
+  checks: ["unicode.invisible", "style.patterns", "watermark.anthropic"],
+  privacy: { allowed_routes: ["browser"], save_receipt: false, retain_content: false }
+});
+
+result.summary;
+// { pass: 0, attention: 2, fail: 0, inconclusive: 0, unsupported: 1, not_configured: 0, not_run: 0, error: 0 }
+result.methods.map(m => [m.id, m.status]);
+// [["unicode.invisible","attention"], ["style.patterns","attention"], ["watermark.anthropic","unsupported"]]
+result.protected_spans.map(s => s.kind);
+// ["currency", "number"]
+```
+
+The zero-width space hidden in the sample content is what turns `unicode.invisible` to `attention`. Statuses come from the fixed vocabulary `pass | attention | fail | inconclusive | unsupported | not_configured | not_run | error`; unrun checks are never collapsed into a pass, and a zero-finding pattern run carries a scope note stating it is not evidence of human authorship.
+
+### Editorial signals score
+
+```js
+const signals = computeEditorialSignals(draftText);
+// { score: 23, classification: "mixed_signals", probabilities, confidence: "low",
+//   categoriesHit, findingCount, wordCount, version: "en-signals:2026.08.6",
+//   status: "scored", escalation: null, description }
+```
+
+Classification is trinary (`human_like` / `mixed_signals` / `ai_like`), false-negative-biased by design, and probability-consistent: the label is always the argmax of the published probabilities. The evidence-based escalation policy (artefact floor, citation co-occurrence, artefact-plus-score, formatting floor, formatting cluster, furniture gate, finding breadth) can only raise a classification and, when it does, `escalation` names the policy and reason. Never present the score as authorship detection: clean, well-prompted AI prose scores low on any rule tier, which is why the result vocabulary says "no strong AI-style signals" rather than "human".
+
+Measured on the fresh 5,558-document long-form corpus (`services/local-engine/research/longform-corpus/`), this rules tier detects **45.1%** of AI writing and wrongly flags **24.8%** of human writing. It flags roughly one human document in four, which is why it is editorial suggestion only and contributes nothing to any AI verdict. The older provider-eval aggregate is superseded and must not be quoted: its human side was 169 documents, 76% of them encyclopaedic and question-and-answer text, so its rate described that corpus and not published prose. See `docs/CAPABILITIES.md` §7.2 and §7.3.
+
+### Unicode findings
+
+```js
+inspectUnicode("claim​ed");
+// [{ code_point: "U+200B", severity: "medium", fix: "remove", span: {...},
+//    matched_text_hash: "sha256:...", limitations: [...] }]
+```
+
+## Test and integration boundary
+
+```sh
+npm ci
+npm test
+npm run pack:check
+```
+
+Use the typed exports rather than copying algorithms into a consumer; browser, Astro, CLI and extension adapters must produce the same stable fields for the same projected input, and the repository's cross-surface battery enforces this. Consumers remain responsible for explicit user consent before inspecting content and for keeping content out of logs and receipts.
+
+## Attribution
+
+This engine was built on existing open-source work by deliberate choice, so the projects it
+stands on are named here rather than only in a licence file.
+
+- [avoid-ai-writing](https://github.com/conorbronsdon/avoid-ai-writing) — MIT, Conor Bronsdon and contributors. 46 of the 51 v2 writing-pattern rule categories, the stylometric methods, the weights and the classifier logic, adapted to TypeScript, plus Cyrillic and Greek lookalike map data. One upstream bug was fixed in the port.
+- [watermarks-remover](https://github.com/guillaumemeyer/watermarks-remover) — MIT, Guillaume Meyer. The invisible-character and space-substitute carrier tables and the explicit-carrier inspection model. Table data only; no upstream code is distributed.
+- [Unicode Consortium character data](https://www.unicode.org/Public/UCD/latest/) — Unicode licence. The 415-code-point carrier inventory across 38 rules and the 60-entry confusable set.
+- [antislop-sampler](https://github.com/sam-paech/antislop-sampler) — Apache-2.0, Sam Paech. Fiction phrase and over-represented name data behind the `fiction-slop-phrase` and `fiction-promptonym` rules.
+- [slop-forensics](https://github.com/sam-paech/slop-forensics) — MIT, Sam Paech. Per-model observations corroborating the fiction-lane rules.
+- [SLOP_Detector](https://github.com/SicariusSicariiStuff/SLOP_Detector) — Apache-2.0, SicariusSicariiStuff. The graded penalty-class weighting approach.
+- [slop-gate](https://github.com/hwajongpark/slop-gate) — MIT, hwajongpark. Promotional-register and buzz-phrase pattern data.
+- [anti-ai-writing](https://github.com/avectats7/anti-ai-writing) — MIT, avectats7. Buzz-phrase and weak-verb observation data.
+- [anti-slop](https://github.com/kjmagnan1s/anti-slop) — MIT, kjmagnan1s. Faux-insight phrase data and the protect-list and context-profile design.
+- [claude-slop-detector](https://github.com/aplaceforallmystuff/claude-slop-detector) — MIT, aplaceforallmystuff. Staccato-fragment and tripled-negation observations.
+- [Wikipedia, *Signs of AI writing*](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) — CC BY-SA 4.0. Editorial guidance independently re-expressed with no verbatim excerpts, credited as the licence requires.
+- [Project Gutenberg](https://www.gutenberg.org) public-domain texts, all published before 1929 — the embedded human-prose reference corpus behind the rhythm and register signals.
+- [canonicalize](https://github.com/cyberphone/json-canonicalization) — Apache-2.0. RFC 8785 canonicalisation.
+- Published academic findings by Liang et al. (ICML 2024), Kobak et al. (*Science Advances* 2025), Juzek & Ward (COLING 2025), Reinhart et al. (PNAS 2025), Geng & Trotta (2024) and Pew Research (2026), used as rule thresholds and lexicon facts.
+
+Several well-known detector repositories were cloned and read during the research phase and are
+credited as exactly that — read, not used. Nothing in this package derives from `fast-detect-gpt`,
+`Binoculars`, `RADAR`, `DIPPER`, `ai-detector-bench`, `BIRA`, `SIRA` or `MarkLLM`.
+
+Full records, with versions, snapshot commits and file-level destinations:
+[THIRD_PARTY_NOTICES.md](https://github.com/OpaceDigitalAgency/opace-ai-content-checker-detector/blob/main/THIRD_PARTY_NOTICES.md) ·
+[DEPENDENCY-LEDGER.md](https://github.com/OpaceDigitalAgency/opace-ai-content-checker-detector/blob/main/docs/legal/DEPENDENCY-LEDGER.md).
+
+## Troubleshooting and links
+
+- **A check reports `unsupported`:** preserve that result. Do not replace it with a different method or a pass.
+- **Browser and server output differ:** confirm both adapters supplied the same projected UTF-8 text and use the same contract/core versions (`UNICODE_RULES_VERSION`, `EN_SIGNALS_PATTERN_VERSION` are exported for exactly this).
+- **A safe fix changes protected content:** reject the candidate and inspect the protected-span gate before applying any text.
+
+Report security concerns through the repository [security policy](https://github.com/OpaceDigitalAgency/opace-ai-content-checker-detector/blob/main/SECURITY.md). For non-sensitive help, use [Opace AI Content Checker & Detector support](https://opace.agency/get-in-touch/). Changes follow the repository [contribution guide](../../CONTRIBUTING.md) and [changelog](../../CHANGELOG.md).
+
+[Opace AI Content Checker & Detector](https://opace.agency/tools/ai/content-verification-integrity/) · [Browser checker](https://opace.agency/tools/ai/content-verification-integrity/checker/) · [Capability register](../../docs/CAPABILITIES.md) · [Opace](https://opace.agency/) · [Opace Digital Agency on GitHub](https://github.com/OpaceDigitalAgency) · [Related contracts package](../contracts/README.md)
+
+## Where this is weakest, measured
+
+This package ships the deterministic character forensics and editorial writing rules. It does
+**not** contain the trained model that produces an AI reading; Cycle 5 runs only in the web
+checker. The model results are disclosed here so package users can distinguish the two systems.
+
+**The writing rules are editorial feedback, not detection.** On 922 machine and 1,200 held-out
+human long-form documents, they detect 45.1% of machine writing while flagging **24.8% of human
+writing** — one human document in four. `computeEditorialSignals` returns a writing score, never
+an authorship reading, and must not be presented as one.
+
+The deployed Cycle 5 model (`tier3-cycle5-v1`, margin rule
+`max(m1, m2+0.34) >= 3.571`) was measured separately through both live runtimes on the full
+5,558-document long-form evaluation corpus. That corpus is not wholly independent: 654/922 AI
+documents are independent of every Cycle 2 split and 268 are not; 11/4,636 human documents
+overlap. The separate topic-matched held-out slice is the independent evasion measurement.
+
+| Cycle 5 measurement | result | denominator and runtime |
+|---|---:|---|
+| AI documents flagged | **97.8%** | 902/922, EU server fp32 |
+| human documents wrongly flagged | **0.99%** | 46/4,636, EU server fp32 |
+| AI documents flagged | **97.6%** | 900/922, browser int8 |
+| human documents wrongly flagged | **1.57%** | 73/4,636, browser int8 |
+| topic-matched held-out AI flagged | **86.9%** | 153/176, server evaluation route |
+| structured human partners wrongly flagged | **0.2%** | 1/418, server evaluation route |
+| human fiction wrongly flagged | **3.1% / 3.5%** | 7/227 server / 8/227 browser |
+| held-out 100-word AI flagged | **76.8%** | 43/56, server evaluation route |
+| heavy AI edits of human originals flagged | **28.5%** | 39/137, server evaluation view |
+| academic human documents wrongly flagged | **0.8%** | 15/1,992, server evaluation view |
+
+The 100-word cell is small, fiction remains higher-risk than the overall human set, and treating
+heavy AI edits as machine-assisted is a product boundary rather than proof of authorship. Do not
+use a result for an academic misconduct decision about one student.
+
+*Historical, not current:* Cycle 2 (`tier3-cycle2-v1`, live 28 August – 1 September 2026) produced
+the earlier 883/922 server, 889/922 browser, 45/4,636 server-human and 90/4,636 browser-human
+figures. Its fiction and length rows are retired and must not be mixed with Cycle 5.
+
+Complete list with sources: [Honest limitations](https://github.com/OpaceDigitalAgency/opace-ai-content-checker-detector#honest-limitations).
+The same figures plotted, with the 50% acceptance floor drawn in: [result charts](https://github.com/OpaceDigitalAgency/opace-ai-content-checker-detector#what-it-measures-and-where-it-fails).