Concepts

Playbooks

A playbook is a proven method for a familiar kind of question — sizing a market, reviewing a literature, comparing vendors. Pick one when you start a piece of Deep Research and it shapes both the plan and how every researcher works.

Why playbooks exist

Some questions go wrong in predictable ways. A market size quoted from one analyst, with no definition, is a number nobody can check. A vendor comparison that measures each vendor on different things is not a comparison. A literature review that reads abstracts instead of studies weighs a blog post like a trial. A playbook writes the method down once, so every run of that kind of question follows it — rather than depending on how the question happened to be phrased.

The three playbooks

Playbook	Use it for	What it makes the research do
Market sizing	"How big is the market for X, and who leads it?"	State the market definition first (products, segments, geography, revenue or volume, year). Collect about three published top-down estimates, each with who produced it and how they defined it. Cross-check bottom-up from primary data (units × price). Add historical growth and attributed forecasts, and the leaders by share. If top-down and bottom-up differ by more than a factor of two, that gap is itself a finding.
Literature review	"What does the evidence say about X?"	Find a survey, systematic review or meta-analysis first. Read the strongest studies it cites in the original. Look for newer work and for critiques, failed replications and retractions. Record each study's design, size, finding, year and whether it is peer-reviewed, and weigh them accordingly — a large randomised trial outweighs many small observational studies.
Vendor comparison	"Compare A, B and C for our use"	Fix one criteria matrix before searching — capabilities, pricing model, deployment, certifications, integrations, support — and measure every vendor on the same list. Each vendor's own documentation first, dated; then independent checks such as analyst evaluations, benchmarks and certification registries. "Not published" is recorded as a finding, marketing claims are not, and no vendor is ranked on a criterion that could not be found for all of them.

Using one

On the Research page, choose a playbook in the Playbook picker beside Sources and Depth; None is the default. Over the API, pass it on POST /research:

{
  "question": "How large is the EV charging market in India, and who are the leading operators?",
  "sources": "both",
  "depth": "standard",
  "playbook": "market-sizing"
}

The values are market-sizing, literature-review and vendor-comparison. A playbook combines with any depth and any sources setting; it changes the method, not how much is spent.

How a playbook takes effect

A playbook acts at two points in a run:

The plan
The planner is told to follow the playbook, so the sub-questions it proposes cover what the method needs — for market sizing, a definition, top-down estimates, a bottom-up check and growth. You see and approve that plan before anything is searched, and you can still edit it.
Every researcher
Each researcher's brief names the playbook, and the researcher loads it with load_skill before it starts, alongside the general research-method and source-triage skills. The method therefore governs how each sub-question is searched and what is recorded — not just how the plan is worded.

The report is still written only from recorded notes and verified sentence by sentence, whatever the playbook. See Deep Research.

Playbooks are skills

Each playbook is an ordinary skill in the organisation ring — playbook-market-sizing, playbook-literature-review and playbook-vendor-comparison, tagged research and playbook — seeded with the built-in skill pack. That means you can open one in the skills catalog, read exactly what it asks for, and see which agent uses it (the hidden research-worker).

To adapt a method to your organisation — your preferred analyst sources, the criteria your procurement team always scores — publish a new version of the skill. Researchers load the current version on their next run, and the platform does not overwrite a playbook you have changed. Roll back like any other skill.

The list of playbooks is fixed You can change what the three playbooks say, but not add a fourth: the Research page and the API accept only these three names. A new skill tagged playbook is an ordinary skill and will not appear in the picker. The planner's one-line summary of each playbook is also built in, so a large change to a playbook's scope may not be reflected in how the plan is split.

Playbooks, skills and depth Depth decides how many sub-questions and how closely sources are read. A playbook decides which sub-questions and what a good answer to each contains. Skills are the mechanism: a playbook is one kind of skill, loaded on demand like any other.