TRocket-Labs · oshorefueled · Jul 24, 2026 · Jul 13, 2026 · Jul 13, 2026 · Jul 13, 2026
diff --git a/AGENTS.md b/AGENTS.md
@@ -1,26 +1,28 @@
 # Repository Guidelines
 
-This repository implements VectorLint — a prompt‑driven, structured‑output content evaluator. Use this guide to navigate the codebase, run it locally, and contribute safely.
+This repository implements VectorLint — a prompt‑driven, structured‑output content reviewer. Use this guide to navigate the codebase, run it locally, and contribute safely.
 
 ## Project Structure & Module Organization
 
 - `src/`
-  - `index.ts` — CLI entry; orchestrates config, discovery, evaluation, reporting
+  - `index.ts` — CLI entry; orchestrates config, discovery, review, reporting
   - `boundaries/` — external data validation (config, CLI args, env vars, YAML, API responses)
   - `chunking/` — content chunking for large documents (recursive chunker, merger, utilities)
   - `cli/` — command definitions and CLI orchestration
   - `config/` — configuration loading and management
   - `errors/` — custom error types and validation errors
-  - `evaluators/` — evaluation logic (base evaluator, registry, specific evaluators)
+  - `executors/` — bounded single-call and target-paging review strategies
+  - `findings/` — finding verification, filtering, severity, and scoring
   - `output/` — TTY formatting (reporter, evidence location, line numbering)
   - `prompts/` — YAML frontmatter parsing, schema validation, directive loading
   - `providers/` — LLM abstractions (OpenAI, Anthropic, Azure, Gemini), request builder, provider factory
   - `scan/` — file discovery (fast‑glob) honoring config and exclusions
+  - `review/` — shared review contracts, budgets, boundaries, and request building
   - `schemas/` — Zod schemas for all external data (API responses, config, CLI, env)
   - `scoring/` — density-based score calculation
   - `types/` — TypeScript type definitions
 - `presets/` — bundled rule packs (e.g., `VectorLint/`)
-- `tests/` — Vitest specs for config, scanning, evaluation, providers
+- `tests/` — Vitest specs for config, scanning, review, providers
 
 ## Agent Behavior Guidelines
 
@@ -112,7 +114,7 @@ Rules must be organized into subdirectories (packs) within `RulesPath`.
 
 ### Zero-Config Mode
 
-If you just want to evaluate against a user instruction guide without specific rules:
+If you just want to review against a user instruction guide without specific rules:
 1. Create a `VECTORLINT.md` file with your user instruction content
 2. Run `vectorlint doc.md` — VectorLint creates a synthetic rule from your user instructions
 
@@ -131,19 +133,18 @@ If you just want to evaluate against a user instruction guide without specific r
 
 VectorLint assembles prompts in this order:
 
-1. **Directive** (`src/prompts/directive-loader.ts`) — Role definition, task, and evaluation instructions
+1. **Directive** (`src/prompts/directive-loader.ts`) — Role definition, task, and review instructions
 2. **User Instructions** (`VECTORLINT.md`) — Optional global style context
-3. **Rule** (the prompt body from the rule file) — Specific evaluation criteria
+3. **Rule** (the prompt body from the rule file) — Specific review criteria
 
-The content to evaluate is sent as a **user message** with line numbers prepended.
+The content to review is sent as a **user message** with line numbers prepended.
 
 ### Chunking
 
 For documents >600 words, VectorLint automatically chunks content:
 - Uses recursive splitting (paragraphs → lines → sentences → words)
-- Each chunk is evaluated separately
+- Each chunk is reviewed separately
 - Results are merged and deduplicated
-- Disable with `evaluateAs: document` in rule frontmatter
 
 ## Coding Style & Naming Conventions
 
@@ -180,11 +181,11 @@ For documents >600 words, VectorLint automatically chunks content:
 
 - Boundary validation: all external data (files, CLI, env, APIs) validated at system boundaries using Zod schemas
 - Type safety: strict TypeScript with no `any`; use `unknown` + schema validation for external data
-- Dependency inversion: depend on `LLMProvider` and `SearchProvider` interfaces; keep providers thin (transport only)
+- Dependency inversion: depend on `StructuredModelClient` and `ToolCallingModelClient`; keep providers thin (transport only)
 - Dependency injection: inject `RequestBuilder` via provider constructor to avoid coupling
-- Separation of concerns: rules define criteria; schemas enforce structure; CLI orchestrates; evaluators process; reporters format
+- Separation of concerns: rules define criteria; schemas enforce structure; CLI orchestrates; executors review; finding processors verify; reporters format
 - Separation of concerns: when a file starts combining contracts, orchestration, and utility logic, extract shared helpers and types into focused modules
-- Extensibility: add providers by implementing `LLMProvider` or `SearchProvider`; add evaluators via registry pattern
+- Extensibility: add providers by implementing the model-client contracts; add review strategies behind `ReviewExecutor`
 - Shared domain constants: avoid magic strings for core runtime concepts; define shared constants, enums, or types and import them where needed
 - Naming: choose domain-accurate names that reflect the real abstraction level; avoid use-case-specific terminology in shared runtime code
 - Logging: route runtime logging through an injected logger interface; keep concrete logger implementations behind the abstraction
@@ -206,7 +207,3 @@ VectorLint supports multiple output formats via the `--output` flag:
 - Anthropic: Claude models (Opus, Sonnet, Haiku)
 - Azure OpenAI: Azure-hosted OpenAI models
 - Google Gemini: Gemini Pro and other Gemini models
-
-### Search Providers
-
-- Perplexity: Sonar models with web search capabilities (used by technical-accuracy evaluator)
diff --git a/CONFIGURATION.md b/CONFIGURATION.md
@@ -15,7 +15,7 @@ VectorLint is configured via a `.vectorlint.ini` file in the root of your projec
 # Optional: Path to custom rules directory
 # If omitted, only preset rules (from RunRules) are used
 # RulesPath=.github/rules
-# Number of concurrent evaluations (Default: 4)
+# Number of concurrent reviews (Default: 4)
 Concurrency=4
 # Default severity for violations (Default: warning)
 DefaultSeverity=warning
@@ -34,8 +34,6 @@ GrammarChecker.strictness=7
 RunRules=Acme
 # Higher strictness for docs
 GrammarChecker.strictness=9
-# Require a higher score for technical accuracy
-TechnicalAccuracy.threshold=8.0
 
 # Marketing content - run "TechCorp" pack
 [content/marketing/**/*.md]
@@ -56,7 +54,7 @@ These settings control the application's core behavior.
 | Setting           | Type    | Default      | Description                                                        |
 | ----------------- | ------- | ------------ | ------------------------------------------------------------------ |
 | `RulesPath`       | string  | (none)       | Root directory for custom rule packs. If omitted, only presets are used. |
-| `Concurrency`     | integer | `4`          | Number of concurrent evaluations to run.                           |
+| `Concurrency`     | integer | `4`          | Number of concurrent reviews to run.                           |
 | `DefaultSeverity` | string  | `warning`    | Default severity level (`warning` or `error`) for reported issues. |
 
 ---
@@ -69,36 +67,23 @@ You can place a `VECTORLINT.md` file in your project root to define global style
 If no `.vectorlint.ini` exists, VectorLint will automatically:
 1. Detect `VECTORLINT.md`
 2. Create a synthetic "Style Guide Compliance" rule
-3. Evaluate your contents against it
+3. Review your contents against it
 
 ### Combined Mode
-If you have configured rules (via `.vectorlint.ini`), the content of `VECTORLINT.md` is **prepended** to the system prompt for every evaluation. This ensures your global style preferences (tone, terminology) are respected across all specific rules.
+If you have configured rules (via `.vectorlint.ini`), the content of `VECTORLINT.md` is **prepended** to the system prompt for every review. This ensures your global style preferences (tone, terminology) are respected across all specific rules.
 
 > **Note:** Keep `VECTORLINT.md` concise. VectorLint will emit a warning if the file exceeds ~4,000 tokens, as very large contexts can degrade performance and increase costs.
 
 ---
 
-## LLM & Search Providers
+## LLM Providers
 
-VectorLint relies on LLM and Search providers. These are configured globally in `~/.vectorlint/config.toml`, or project scope using a `.env` file (which takes precedence).
+VectorLint relies on an LLM provider. Configure it globally in `~/.vectorlint/config.toml`, or at project scope using a `.env` file (which takes precedence).
 
 You can generate these files using the `vectorlint init` command.
 
-### LLM Providers
-
 VectorLint supports multiple LLM providers. Set `LLM_PROVIDER` to your desired provider (e.g., `openai`, `anthropic`, `gemini`) and provide the corresponding API key.
 
-### Search Provider
-
-Some evaluators, such as **TechnicalAccuracy**, require access to current external knowledge to verify facts. VectorLint supports search providers to fetch this information.
-
-**Example configuration for Perplexity:**
-
-```bash
-SEARCH_PROVIDER=perplexity
-PERPLEXITY_API_KEY=pplx-...
-```
-
 ### Observability
 
 VectorLint can optionally emit AI execution telemetry to Langfuse. The first implementation is scoped to the Vercel AI SDK calls made through `VercelAIProvider`.
@@ -224,7 +209,7 @@ You can use named levels or direct numeric multipliers:
 [content/docs/**/*.md]
 RunRules=Acme
 GrammarChecker.strictness=strict
-TechnicalAccuracy.strictness=20
+Terminology.strictness=20
 ```
 
 ---

diff --git a/CREATING_RULES.md b/CREATING_RULES.md
@@ -1,6 +1,6 @@
 # Creating Rules for VectorLint
 
-A comprehensive guide to creating powerful, reusable content evaluations using VectorLint's prompt system.
+A comprehensive guide to creating powerful, reusable content reviews using VectorLint's prompt system.
 
 ## Table of Contents
 
@@ -36,9 +36,8 @@ Every rule is a Markdown file with two parts:
 ```markdown
 ---
 # YAML Frontmatter (Configuration)
-id: MyEval
-name: My Content Evaluator
-evaluator: base
+id: MyRule
+name: My Content Reviewer
 severity: error
 ---
 
@@ -57,9 +56,7 @@ project/
 │   └── rules/
 │       ├── Acme/                    ← Company style guide pack
 │       │   ├── grammar-checker.md
-│       │   ├── headline-evaluator.md
-│       │   └── Technical/           ← Nested organization supported
-│       │       └── technical-accuracy.md
+│       │   └── headline-reviewer.md
 │       └── TechCorp/                ← Another company's pack
 │           └── brand-voice.md
 └── .vectorlint.ini
@@ -78,7 +75,6 @@ score from the verified findings.
 
 ```markdown
 ---
-evaluator: base
 id: GrammarChecker
 name: Grammar Checker
 severity: error
@@ -117,7 +113,7 @@ Check this content for grammar issues, spelling errors, and punctuation mistakes
 
 The `target` field allows you to:
 
-1. **Specify which part** of content to evaluate (via regex)
+1. **Specify which part** of content to review (via regex)
 2. **Require certain content** to exist (e.g., "must have an H1 headline")
 3. **Provide helpful suggestions** when content is missing
 
@@ -136,13 +132,13 @@ target:
 
 **When `required: true`:**
 
-- If content matches → Evaluation proceeds normally
+- If content matches → Review proceeds normally
 - If no match → Immediate `error` with the suggestion message
 
 **When `required: false` or omitted:**
 
-- If content matches → Evaluate the matched content
-- If no match → Evaluate entire content
+- If content matches → Review the matched content
+- If no match → Review entire content
 
 ---
 
@@ -153,12 +149,10 @@ target:
 | Field         | Type          | Required | Description                                                    |
 | ------------- | ------------- | -------- | -------------------------------------------------------------- |
 | `specVersion` | string/number | No       | Rule specification version (use `1.0.0`)                       |
-| `evaluator`   | string        | No       | Evaluator type: `base`, `technical-accuracy` (default: `base`) |
 | `id`          | string        | **Yes**  | Unique identifier (used in error reporting)                    |
 | `name`        | string        | **Yes**  | Human-readable name                                            |
 | `severity`    | string        | No       | `error` or `warning` (default: `warning`)                      |
 | `strictness`  | number/string | No       | Density penalty: a positive number, `lenient`, `standard`, or `strict` |
-| `evaluateAs`  | string        | No       | `document` or `chunk` (default: `chunk`)                       |
 | `target`      | object        | No       | Content matching specification                                 |
 
 ---
@@ -178,7 +172,7 @@ Check if the headline is good.
 ✅ **Good:**
 
 ```markdown
-You are a headline evaluator for developer blog posts. Assess whether the headline:
+You are a headline reviewer for developer blog posts. Assess whether the headline:
 
 1. Clearly communicates a specific benefit
 2. Uses natural, conversational language (avoid buzzwords)
@@ -208,7 +202,6 @@ Help the LLM understand your domain:
 
 ```markdown
 ---
-evaluator: base
 id: GrammarChecker
 name: Grammar Checker
 severity: error
@@ -226,4 +219,4 @@ Report any errors found with specific examples.
 
 ---
 
-**Happy evaluating! 🚀**
+**Happy reviewing! 🚀**
diff --git a/README.md b/README.md
@@ -1,6 +1,6 @@
 # VectorLint: Prompt it, Lint it! [![npm version](https://img.shields.io/npm/v/vectorlint.svg)](https://www.npmjs.com/package/vectorlint) [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 
-VectorLint is a command-line tool that uses LLMs to find and score terminology, technical accuracy, and style issues that require contextual understanding.
+VectorLint is a command-line tool that uses LLMs to find and score terminology, consistency, and style issues that require contextual understanding.
 
 ![VectorLint Screenshot](./assets/VectorLint_screenshot.jpeg)
 
@@ -51,7 +51,7 @@ VectorLint scores your content using error density, enabling you to measure qual
 
 ## How VectorLint Reduces False Positives
 
-VectorLint uses a PAT (Pay A Tax) evaluation approach:
+VectorLint uses a PAT (Pay A Tax) review approach:
 
 1. **Candidate generation:** the model returns all potential violations with required gate-check fields (rule support, exact evidence, context support, plausible non-violation, and fix quality).
 2. **Deterministic surfacing:** VectorLint applies a strict filter and only surfaces violations that pass all required gates.

diff --git a/docs/VECTORLINT.md b/docs/VECTORLINT.md
@@ -1,7 +1,7 @@
 # VectorLint example styling rules
 
-{/* This file is prepended to the system prompt for every VectorLint evaluation 
-  This file defines global style instructions for all VectorLint evaluations.
+{/* This file is prepended to the system prompt for every VectorLint review
+  This file defines global style instructions for all VectorLint reviews.
   VectorLint prepends its contents to the system prompt for every rule it runs.
   Keep this file under ~800 tokens to avoid performance and cost issues.
   Adapt the rules below to match your team's style guide.

diff --git a/docs/best-practices.mdx b/docs/best-practices.mdx
@@ -13,7 +13,7 @@ After you run `VECTORLINT.md` against your content library and complete your fir
 
 ## Keep VECTORLINT.md under 800 tokens
 
-VectorLint emits a warning at 4,000 tokens, but a practical target is *much* lower. Under 800 tokens leaves headroom for rule-specific prompts to add context without the combined system prompt becoming unwieldy. Long context degrades LLM precision and increases API costs on every evaluation.
+VectorLint emits a warning at 4,000 tokens, but a practical target is *much* lower. Under 800 tokens leaves headroom for rule-specific prompts to add context without the combined system prompt becoming unwieldy. Long context degrades LLM precision and increases API costs on every review.
 
 If your `VECTORLINT.md` is growing, that's usually a sign that some rules are specific enough to belong in a dedicated rule pack file instead.
 
@@ -28,7 +28,7 @@ Check if the writing is clear.
 
 Write:
 ```
-You are a clarity evaluator for developer documentation. Flag sentences that:
+You are a clarity reviewer for developer documentation. Flag sentences that:
 1. Exceed 25 words
 2. Use passive voice where active voice is possible
 3. Contain filler phrases: "it is important to note", "please be aware", "in order to"
@@ -38,7 +38,7 @@ The second prompt gives the LLM a defined audience, measurable criteria, and spe
 
 ## Give rules domain context
 
-LLMs evaluate content relative to an implied standard. Make that standard explicit with a context block in your rule prompt. Tell the model who the audience is, what they value, and what good looks like for your specific content type.
+LLMs review content relative to an implied standard. Make that standard explicit with a context block in your rule prompt. Tell the model who the audience is, what they value, and what good looks like for your specific content type.
 
 ```markdown
 ## CONTEXT BANK
@@ -85,7 +85,7 @@ The goal is a workflow where every finding is worth reading. That takes iteratio
 
 ## Set a higher confidence threshold in CI than locally
 
-In CI, a false positive blocks a merge. Set `CONFIDENCE_THRESHOLD` higher in your CI environment than in local development so only the highest-confidence findings gate a merge. Lower-confidence candidates still surface locally where a writer can evaluate them in context.
+In CI, a false positive blocks a merge. Set `CONFIDENCE_THRESHOLD` higher in your CI environment than in local development so only the highest-confidence findings gate a merge. Lower-confidence candidates still surface locally where a writer can review them in context.
 
 ```bash
 # Local development — catch more, review in context
@@ -117,6 +117,6 @@ Skipping this step leads to rules that look correct on paper but produce noise i
 
 ## Next steps
 
-- [Tuning evaluation precision](/false-positive-tuning) — detailed guidance on CONFIDENCE_THRESHOLD and strictness
+- [Tuning review precision](/false-positive-tuning) — detailed guidance on CONFIDENCE_THRESHOLD and strictness
 - [CI Integration](/ci-integration) — set up content quality gates in your pipeline
 - [Customize style rules](/customize-style-rules) — write effective prompts for rule pack files
diff --git a/docs/ci-integration.mdx b/docs/ci-integration.mdx
@@ -155,6 +155,6 @@ Store API keys in your CI system's secret or environment variable manager — ne
 
 ## Next steps
 
-- [Tuning evaluation precision](/false-positive-tuning) — set `CONFIDENCE_THRESHOLD` and strictness for CI
+- [Tuning review precision](/false-positive-tuning) — set `CONFIDENCE_THRESHOLD` and strictness for CI
 - [Project Configuration](/project-config) — configure which files CI checks
 - [CLI reference](/cli-reference) — full command and exit code reference
diff --git a/docs/cli-reference.mdx b/docs/cli-reference.mdx
@@ -89,8 +89,6 @@ VectorLint reads configuration from environment variables in addition to config
 | `ANTHROPIC_API_KEY` | — | API key for Anthropic |
 | `GEMINI_API_KEY` | — | API key for Google Gemini |
 | `AZURE_OPENAI_API_KEY` | — | API key for Azure OpenAI |
-| `SEARCH_PROVIDER` | — | Search provider for `technical-accuracy` evaluator: `perplexity` |
-| `PERPLEXITY_API_KEY` | — | API key for Perplexity search |
 | `CONFIDENCE_THRESHOLD` | `0.75` | PAT pipeline confidence gate. Lower surfaces more findings; higher surfaces fewer. |
 
 For the full environment variable reference see [Environment variables](/env-variables).
@@ -99,5 +97,5 @@ For the full environment variable reference see [Environment variables](/env-var
 
 - [Installation](/installation) — install VectorLint globally or run with npx
 - [Configuration](/configuration) — configure VectorLint for your project
-- [Tuning evaluation precision](/false-positive-tuning) — adjust `CONFIDENCE_THRESHOLD` and strictness
+- [Tuning review precision](/false-positive-tuning) — adjust `CONFIDENCE_THRESHOLD` and strictness
 - [CI Integration](/ci-integration) — use exit codes to gate merges on content quality