API and agent tools

Discovery. Lists every available tool with its parameters.

The compact state to decide from: the answer /api/agent?view=state gives.

A summary: current conditions, action, margin (the confidence field), record date. Optional weekly, monthly, quarterly or yearly view. The MCP tool of the same name returns the full brief.

How long these conditions have held, how long they have lasted on average, and what usually comes next. With no conditions given it uses today's.

Weeks until a score crosses its change line at a pace you give, or, with no arguments, at today's scores and pace. Scenario modeling for agents.

Median SaaS EV/Revenue multiple across 14 public companies.

Coverage map of which months have signal history available.

Discovery manifest. Where an agent starts, the agent endpoints and MCP tools, the supported cadences, and the revisable-delegation conformance level with the third-party verifier that can check it. Start here if you are wiring up an agent.

Stable JSON endpoint for the current guidance. Returns the conditions, funding pressure, direction, margin (confidence), week, summary, url, and version. Type contract at lib/types.ts.

The state to decide from, about a tenth the size of the full brief: the conditions and the margin from their change lines, the data's freshness (automation.ready), each score against its pause, resume and change lines with the moves behind it, and the guidance. Weekly only. Check automation.ready before acting without a person looking.

The full structured brief, with the block an agent hands on. Supports weekly, monthly, quarterly, yearly.

One published week, unchanged since it went out, with its permalink. A week not on the record answers 404 and lists the weeks that are.

A past quarter or year from the archive (2024-q1, 2023), with its permalink. A period not in the archive answers 404 and lists the ones that are.

This week's guidance as a ready-to-send request for TypeSafe's Jev decision model (typesafe/jev on Cloudflare Workers AI). Put a short description of your plan in request.state.plan and send the request yourself. Jev answers whether the plan goes against the guidance, which decision it touches, how hard it is to undo, and whether it is safer after conditions change. Whether never sees the plan. Check automationReady before you act on an answer: it is false when this week's reading is not live or not current.

This week's guidance in one JSON object: the action, each area's action with its pause and resume lines, the decisions to revisit, and the lines that would change the guidance.

The two headline numbers on their own: funding pressure, risk appetite, both deltas, the distance to the line where conditions change, and the score that would cross first.

Weekly and monthly summaries side by side, with an alignment status naming any constraints where the two cadences disagree.

SaaS valuation multiples, with the source that answered. Returns 503 when no market data source answers, rather than an empty basket.

Paginated history of alerts sent when conditions changed. Read access is open; only the Monday run writes.

This week's guidance shaped for one tool: Slack blocks (the default), a Notion page, or a Linear issue. The Slack channel and the Notion database it names are placeholders; set your own before you send it.

Past guidance in NDJSON, shaped by the Revisable Delegation record schema — one belief record per week of published guidance. format=csv returns the same entries as a spreadsheet. Scoped to the caller: it reads your own decision memory, so it needs an x-whether-client-id header and answers 401 without one. No "Try it" link for that reason.

The week's guidance preformatted for Slack — action, pause line, resume line. Post it to a channel without reformatting.

The brief as a PNG card for decks and threads, 1200×800 by default. layout draws it landscape (1200×675), square (1080×1080) or portrait (1080×1350), and theme=light draws it on a light ground. month=3&year=2024 draws that month-end's reading, and stage (seed, series-a, series-b or bootstrapped) calibrates the action to a company stage. ticker and sector add a label and leave the reading unchanged.

The screenshot of the homepage's guidance taken each Monday, as a PNG. date is a reading's date, and latest, the default, is the newest. Each is kept about two months; a date with none stored answers 404.

This week's conditions as an SVG badge for a README, a data room or an investor page, in the conditions' color. It says so when the reading is a saved copy.

The guidance as a widescreen meeting slide — hand-written SVG, so it stays sharp at projector size and the labels are still editable text in Figma or Keynote. Accepts seed, series-a, series-b, or growth.

The same slide as a file: a one-slide native PPTX with every box editable, or a one-page widescreen PDF for a pre-read. Served as an attachment. Takes the same stage values as the SVG. With `week`, the slide is the frozen brief as published — no stage, and a brief from before scores were stored prints no scores.

The brief as a snippet a newsletter build step can insert at send time: ready-to-paste HTML (the default), plain text, JSON, or page, a whole document for an iframe. mode=call-strip shortens it to one line. stage and runway (a band: critical, tight, balanced or abundant; or a number of months) calibrate the guidance to one company, and ticker names it.

The current brief in Markdown format. Instant, structured, cacheable.

Agent discovery document. Linked from <link rel="llm"> in the page head.

The answer pages, guides and terms in one Markdown file, for a model that reads the site's text in one fetch.

RSS feed for the weekly brief. Subscribe in any RSS reader.

The same weeks as a calendar to subscribe to: each Monday of the last six months is an all-day event with that week's conditions and guidance. Subscribe by URL in any calendar app.

What whether status reads. asOf reads a past month; view=expanded adds what would change the guidance and the signals that moved.

One row per month-end, newest first, going back weeks weeks (at most 52, the default). Each row carries that month-end's own margin and trend.

What whether scenario reads: the conditions the scores you give would read as, against this week's. Either score may be left out.

What whether changes, call, regime and diagnostics print, in JSON. Each takes asOf.

This week's guidance and the numbers behind it. Add --as-of 2026-03 to read any past month.

The state an agent decides from, in JSON: the answer /api/agent?view=state gives.

The same week rewritten for a specific audience.

Export the brief to json, csv, markdown, pdf, or a full briefing pack.

Poll for a change in the guidance and print only what changed.

Model a hypothetical and see which guidance it produces.

What would have to move, and by how much, to change this week's guidance.