Whether
How it works
Whether scores two things from public U.S. market data every Monday:
- Funding pressure (0–100): how hard it is for U.S. companies to raise and borrow money. Higher is worse.
- Risk appetite (0–100): how willing investors and buyers are to back new products, hires and growth. Higher is better.
Together the two scores place the week in one of four conditions: Sunny, Cloudy, Stormy or Freezing. Interest rates, the yield curve, the pace of rate rises, stock-market volatility and the Chicago Fed financial conditions index move the scores, because the record holds each of them at every month-end since 2012. Inflation, labor, the high-yield credit spread and startup-market inputs are shown beside them and move nothing.
Each decision area gets an action for this week, a pause line that holds it back and a resume line, set apart from the pause line so one noisy week cannot flip it. Three rows from the brief for the week of Oct 19, 2026, in Cloudy conditions:
| Area | This week | Pause line | Resume line |
|---|
| Hiring | Limit hiring to backfills and mission-critical roles tied to delivery risk | Funding pressure > 76 or risk appetite < 44 | Funding pressure < 66 and risk appetite > 54 |
| Spending | Put a payback figure on every discretionary line before approving it | Funding pressure > 76 | Funding pressure < 66 and the payback figures still hold |
| Fundraising | If you plan to raise, prepare the materials now so the timing stays yours | Risk appetite < 44 or funding pressure > 76 | Risk appetite > 54 and funding pressure < 66 |
The live brief at / fits the actions to a company's stage and cash runway. Every number traces to its formula and source on /trust.
The product
| |
|---|
| |
Decide (/operate): this week's rule for each area, the work to stop and start, and a runway planner. | Why it changed (/signals): each reading since the last brief, measured against the lines it governs. |
| |
Trust (/trust): how each score is built, where each feed comes from and when it last updated. | Record (/history): conditions by month since 2012, every published brief, and how past calls held up. |
| |
Share (/share): the week's guidance as a Slack post, email, meeting slide, ticket or investor update. | On a phone: the same brief at 430px. |
Captured from whether.work on Oct 21, 2026. docs/assets/README.md says how to recapture them.
Quick start
Try it without installing
Every endpoint is public and needs no key or account:
curl https://whether.work/brief.md
That prints this week's brief as Markdown: the conditions, the guidance, and each decision with its pause and resume lines. For the same reading as about 3KB of JSON, with each score against its lines:
curl "https://whether.work/api/agent?view=state"
Run locally
git clone https://github.com/zz-plant/whether.git
cd whether
bun install
bun run dev
Open http://localhost:3000. Without network access the brief falls back to the snapshots in data/ and says so.
Use the CLI
bun run cli:whether -- status
bun run cli:whether -- brief slack
bun run cli:whether -- watch --changes-only
status prints this week's conditions and call, brief slack prints the Slack post, and watch refreshes until a score moves. Pass --api-base-url https://whether.work to read production. docs/cli-command-reference.md lists every command.
Connect an agent
In Claude Code:
claude mcp add --transport http whether https://whether.work/api/mcp
In Cursor or VS Code, one click: Add to Cursor · Add to VS Code
Any other MCP client takes the same URL:
{
"mcpServers": {
"whether": {
"url": "https://whether.work/api/mcp"
}
}
}
Or run the server locally over stdio with bun run mcp:whether. Agents should call get_state first. The tools, the manifest at /.well-known/whether-agent.json and /llms.txt are all generated from lib/agent/tools.ts; docs/agents/agent-interface.md covers the rest.
The REST endpoints, their parameters and the ones that need a header are in docs/developer/api-reference.md; whether.work/api-docs is the public version.
Embed this week's call
Paste two lines into any page, such as a team wiki, an investor portal or an internal dashboard:
<div data-whether-embed="brief"></div>
<script async src="https://whether.work/embed.js"></script>
brief draws the guidance and each area's action; call-strip draws a single line. The block updates when the brief does and links back to the full reading. The builder at whether.work/share adds a company stage and previews the result.
Under the hood
- One registry, every agent surface. Each tool is defined once in
lib/agent/tools.ts. The HTTP and stdio MCP servers, the plain-GET shim, the manifest and /llms.txt all read from it, and tests/agentDiscoveryParity.test.ts fails if any of them drifts.
- An engine that reads no data and writes no words.
lib/regimeEngineScoring.ts turns inputs into codes. Every sentence a reader sees comes from lib/report/, and an architecture test rejects any engine import of copy, React or data files. A change to scoring bumps ENGINE_VERSION.
- Fallback that says so. If a feed fails or goes stale, the brief draws from the snapshots in
data/ and labels the reading with its date and the reason.
- A record you can audit. Conditions are scored at every month-end since 2012 by the same engine the live brief runs, so a past call can be checked against what followed.
- Fast, strict gates. The 2,500+ tests run in about 14 seconds on Bun. The pre-push gate rejects new
as casts, new lint suppressions and exports with no caller.
Development
| Command | What it runs | Time |
|---|
bun run dev | Dev server with hot reload (Vinext on Cloudflare Workers) | — |
bun run test <path> | One test file | ~0.2s |
bun run test | The full suite, 2,500+ tests | ~14s |
bun run typecheck | TypeScript 7 (native) | ~5s |
bun run lint | oxlint, warnings denied | ~1s |
bun run build | The production Worker bundle | — |
bun run check | Lint, types, tests, assets, audit and landmark checks, concurrently | ~50s |
check also runs on git push through .husky/pre-push, and that hook is the gate: the GitHub workflows run only by hand. CLAUDE.md lists every command and what each costs; AGENTS.md holds the rules a change has to meet.
Codebase map
whether/
├── app/ # Routes and UI
│ ├── page.tsx # The weekly brief
│ ├── operate/ # Decide: rules per area, runway planner, scenarios
│ ├── signals/ # Why it changed: the evidence behind the scores
│ ├── trust/ # Sources, freshness and formulas
│ ├── history/ # The record: conditions by month, published briefs
│ ├── share/ # Every copy, file and image format of the brief
│ └── api/ # JSON API, MCP server, images and cron handlers
├── lib/
│ ├── regimeEngine.ts # Scores the week and names its condition
│ ├── macroSnapshot.ts # Treasury and FRED ingestion, with fallback
│ ├── report/ # Brief assembly and every reader-facing sentence
│ ├── export/ # Slack, email, slide, deck and investor-update builders
│ ├── alert/ # Alerts between Mondays, by email and Slack
│ └── agent/ # The tools every agent surface reads
├── data/ # Fallback snapshots
├── docs/ # Product specs, editorial rules and agent guides
└── tests/ # The suite
Place in the revisable delegation loop
Whether holds the act step in the STD-07 Revisable Delegation Record loop. It turns changed conditions into weekly operating guidance, and the operator acts. It authorizes nothing and takes no action itself.
- Export: a reader's saved decisions in STD-07 shape at
GET /api/decision-memory?format=records, as newline-delimited JSON, SHA-256 chained, oldest first. The endpoint returns only the requesting owner's records.
- Conformance: Level 2 (append-only, hashed, chained), declared in
/.well-known/whether-agent.json.
- Siblings: Ambit (capability and authorization), Refract (discrepancy), NextConsensus (belief), Ethotechnics (record shape and vocabulary).
Documentation
