feat: Signal Score — AI-assisted application pre-screening

.env.example

@@ -0,0 +1,20 @@
+# Environment variables for Awesomebits
+# Copy to .env for local development (gitignored)
+#
+# The README documents .env as the standard location.
+# For extra safety, dotenv-rails also loads .env.local
+# (which takes precedence) — use that for personal API keys.
+
+# AWS (S3 uploads)
+# AWS_ACCESS_KEY_ID=XXX
+# AWS_SECRET_ACCESS_KEY=YYY
+# AWS_BUCKET=your-bucket-name
+
+# Database (defaults work for local dev)
+# DB_NAME=awesomefoundation_development
+# POSTGRES_HOST=localhost
+# POSTGRES_USER=
+# POSTGRES_PASSWORD=
+
+# Spam detection
+# SPAM_REGEXP=

.env.local.example

@@ -0,0 +1,9 @@
+# Personal API keys — takes precedence over .env
+# Copy to .env.local and fill in your keys (gitignored)
+#
+# dotenv-rails loads .env.local after .env, so values here
+# override anything in .env. Use this for secrets you don't
+# want to accidentally commit.
+
+# Signal Score (LLM scoring via Anthropic Batch API)
+# ANTHROPIC_API_KEY=sk-ant-...

.gitignore

@@ -1,7 +1,10 @@
 .ruby-gemset
 .bundle
 .env
+.env.local
 .envrc
+.scratch/
+context/memory/.basic-memory/
 db/*.sqlite3
 log/*.log
 tmp/

CLAUDE.md

@@ -0,0 +1,113 @@
+# CLAUDE.md — Instructions for Claude Code Sessions
+
+This file is read automatically at the start of every Claude Code session in this workspace.
+
+## Project
+
+**Awesomebits** — the web application for the Awesome Foundation, a global micro-granting community that gives $1,000 grants to "awesome" projects. This is a fork (`divideby0/awesomebits`) of the upstream repo (`awesomefoundation/awesomebits`).
+
+We're building **Signal Score** — an AI-assisted application pre-screening system that helps trustees prioritize grant applications by surfacing quality signals. See GitHub issue #590.
+
+## Project Structure
+
+```
+.
+├── app/                  # Rails application
+│   ├── extras/           # SpamChecker, SpamClassifier
+│   ├── models/           # ActiveRecord models (Project, Chapter, etc.)
+│   └── ...
+├── context/
+│   └── memory/           # basic-memory knowledge base for Signal Score
+├── scripts/
+│   └── signal-score/     # Signal Score tooling (Ruby)
+├── .scratch/             # Local data, never committed
+│   └── data/             # Parquet, CSV, DuckDB files
+├── CLAUDE.md             # This file
+└── ...
+```
+
+## Tech Stack
+
+- **Ruby on Rails** — existing app (Ruby 3.3.6, Rails 7.2.3)
+- **PostgreSQL** — production database
+- **Ruby** — Signal Score scripts (same toolchain, future Rails integration)
+- **DuckDB** — local analytical queries on historical grant data
+
+## Secrets & Configuration
+
+- **`.env.local`** — personal API keys and secrets (gitignored). Never use `.env` for secrets.
+- `dotenv-rails` is in the Gemfile and loads `.env.local` automatically.
+- Required keys:
+  ```
+  ANTHROPIC_API_KEY=sk-ant-...
+  ```
+
+## Git & Commits
+
+### Commit Lint
+- **Header max 50 characters** — this is strict, plan for it
+- **Body lines max 72 characters**
+- **Format:** `type(scope): description`
+- **Types:** feat, fix, refactor, chore, docs, test, perf, ci
+- **Valid scopes:** signal-score, spam, data, scripts, context, infra, docs, deps, repo
+- **References required:** Include `Refs: #591` or `Closes: #N` in the body
+- **Subject must be lowercase** — no sentence-case, start-case, pascal-case, or upper-case
+
+### Branch Convention
+- Feature branches: `feat/<issue-number>-<name>` (e.g. `feat/0591-signal-score-scripts`)
+- Always branch from `master` (upstream default)
+
+### Git Remotes
+- `origin` → `awesomefoundation/awesomebits` (upstream, read-only for us)
+- `fork` → `divideby0/awesomebits` (our fork, push here)
+- Push to `fork`, PR against `origin`
+
+## Data
+
+### Historical Grant Data
+Historical application data is exported from the Awesome Foundation production database and stored in a privileged Google Drive folder. Contact @divideby0 for access.
+
+### Local Data Setup
+Data files live in `.scratch/data/` (gitignored). Expected files:
+- `projects.csv` → `projects.parquet` → loaded into `awesomebits.duckdb`
+- `chapters.csv` → `chapters.parquet` → loaded into `awesomebits.duckdb`
+- Additional: `comments.csv`, sample sets, validation scores
+
+### DuckDB
+```bash
+duckdb .scratch/data/awesomebits.duckdb
+```
+
+Use for analytical queries on historical data. Never connect to production.
+
+## Signal Score Architecture
+
+### Scoring Pipeline
+1. Application text → LLM batch API (Anthropic)
+2. Structured JSON output with Trust Equation dimensions
+3. Composite score 0.0–1.0 + feature breakdown + flags
+
+### Trust Equation
+`T = (Credibility + Reliability + Intimacy) / (1 + Self-Interest)`
+
+### Key Files
+- `scripts/signal-score/score_grants.rb` — batch scoring via Anthropic API
+- `scripts/signal-score/import_data.rb` — CSV → Parquet → DuckDB
+- `context/memory/` — research notes, analysis, pattern discovery
+
+## Context: Existing Spam Detection
+
+Two existing systems (bot detection only, not content analysis):
+- `app/extras/spam_checker.rb` — blocklist + identical fields
+- `app/extras/spam_classifier.rb` — behavioral JS metadata analysis
+
+Signal Score is complementary — it analyzes content, not form behavior.
+
+## Notifying Evie
+
+When you finish a task, get stuck, or need feedback:
+```bash
+openclaw agent --agent main --message "[CC: <brief task name>] <your message>" --timeout 30
+```
+
+Keep messages concise — they'll be read on a phone.

Gemfile

@@ -78,6 +78,7 @@ group :test do
   gem 'shoulda-matchers', '~> 3'
   gem 'timecop'
   gem 'turnip'
+  gem 'webmock'
 end
 
 group :staging, :production do

Gemfile.lock

@@ -152,6 +152,9 @@ GEM
     concurrent-ruby (1.3.4)
     connection_pool (2.5.0)
     content_disposition (1.0.0)
+    crack (1.0.1)
+      bigdecimal
+      rexml
     crass (1.0.6)
     csv (3.3.3)
     cucumber-gherkin (28.0.0)
@@ -200,6 +203,7 @@ GEM
       activerecord (>= 4.0.0)
     globalid (1.2.1)
       activesupport (>= 6.1)
+    hashdiff (1.2.1)
     high_voltage (4.0.0)
     honeypot-captcha (1.0.1)
     htmlentities (4.3.4)
@@ -475,6 +479,10 @@ GEM
       activemodel (>= 6.0.0)
       bindex (>= 0.4.0)
       railties (>= 6.0.0)
+    webmock (3.26.1)
+      addressable (>= 2.8.0)
+      crack (>= 0.3.2)
+      hashdiff (>= 0.4.0, < 2.0.0)
     webrick (1.9.1)
     websocket (1.2.11)
     websocket-driver (0.8.0)
@@ -560,6 +568,7 @@ DEPENDENCIES
   tus-server (~> 2.3)
   uppy-s3_multipart (~> 1.2)
   web-console
+  webmock
   will_paginate (~> 3.3.1)
   xmlrpc
 

README.md

@@ -62,6 +62,44 @@ AWS_SECRET_ACCESS_KEY=YYY
 AWS_BUCKET=your-bucket-name
 ```
 
+### Signal Score (AI-Assisted Pre-Screening)
+
+Signal Score uses the Anthropic API to pre-screen grant applications, helping trustees
+prioritize their review. Scores are advisory only and never auto-hide applications.
+
+**Setup:**
+
+1. Get an Anthropic API key from [console.anthropic.com](https://console.anthropic.com)
+2. Add it to `.env.local` (create this file if it doesn't exist — it's gitignored):
+
+```shell
+ANTHROPIC_API_KEY=sk-ant-...
+```
+
+3. Enable scoring for a chapter by inserting a config row:
+
+```sql
+INSERT INTO signal_score_configs (chapter_id, scoring_config, prompt_overrides, category_config, enabled, created_at, updated_at)
+VALUES (NULL, '{"model": "claude-haiku-4-5-20251001"}'::jsonb, '{}'::jsonb, '{}'::jsonb, true, NOW(), NOW());
+```
+
+Set `chapter_id` to NULL for a global default, or a specific chapter ID for per-chapter config.
+
+**How it works:**
+- New applications are scored asynchronously via `ScoreGrantJob` (SuckerPunch)
+- Each score costs ~$0.01 via the Anthropic Haiku model (~2 seconds per app)
+- Results are stored in `projects.metadata['signal_score']` as JSONB
+- Trustees can filter by score threshold and sort by score on the dashboard
+- Score badges are hidden by default — trustees opt in via "Show scores" checkbox
+- The scoring rubric uses a Trust Equation framework: `T = (C + R + I) / (1 + S)`
+
+**Dashboard controls:**
+- **Show scores** — toggle score badges on individual applications
+- **Sort** — Latest, Earliest, Score highest, Score lowest, Random (stable hash)
+- **Filter** — All apps, Hide low signal (default), Borderline+, Solid+, Strong only
+
+**Cost at scale:** ~$0.01/app. A chapter receiving 50 apps/month = ~$0.50/month.
+
 Subdomains
 ----------
 

app/assets/javascripts/application-score-filter.js

@@ -0,0 +1,46 @@
+/* Score filter dropdowns — each .score-filter-wrapper is independent */
+
+$('a.score-selection').on('click', function(event) {
+  event.preventDefault();
+  const $wrapper = $(this).closest('.score-filter-wrapper');
+  const $selector = $wrapper.find('.score-selector');
+  const isOpen = $selector.hasClass('expanded');
+
+  // Close all dropdowns first
+  $('.score-selector').removeClass('expanded');
+  $('a.score-selection').removeClass('expanded');
+
+  if (!isOpen) {
+    $selector.addClass('expanded');
+    $(this).addClass('expanded');
+  }
+});
+
+$(document).click(function(e) {
+  if (!$(e.target).closest('.score-filter-wrapper').length) {
+    $('.score-selector').removeClass('expanded');
+    $('a.score-selection').removeClass('expanded');
+  }
+});
+
+/* Show/hide score badges — client-side toggle with localStorage persistence */
+(function() {
+  var STORAGE_KEY = 'awesomebits-show-scores';
+  var $checkbox = $('#show-scores');
+
+  // Restore saved state on page load
+  if (localStorage.getItem(STORAGE_KEY) === 'true') {
+    $checkbox.prop('checked', true);
+    $('.signal-score').show();
+  }
+
+  $checkbox.on('change', function() {
+    var checked = $(this).is(':checked');
+    localStorage.setItem(STORAGE_KEY, checked);
+    if (checked) {
+      $('.signal-score').show();
+    } else {
+      $('.signal-score').hide();
+    }
+  });
+})();