Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/xxyoudeadpunkxx/a.d.a.m.-adaptive-depth-and-mode/llms.txt

Use this file to discover all available pages before exploring further.

A.D.A.M.’s structural routing is automatic, but you can override it at any time by starting your message with LOW, MID, or DEEP. Manual override wins over structural routing — it sits at position 5 in the priority ladder, above STRUCTURAL_KERNEL and STATE_DEFAULT. No gating, no waiting — the mode is forced immediately.

​
Syntax

The rules are strict and short:
  • The override token must be the first token of your message
  • It must be followed by a boundary space — or be the entire message
  • It is case-sensitive: LOW, MID, DEEP exactly — no lowercase, no mixed case
  • The override is prefix-based: everything after the token and its boundary space is treated as the message content
Valid overrides: 
LOW capital of germany
MID explain JWT tokens
DEEP compare these three options:
DEEP
MID
Not valid overrides: 
MID, explain JWT tokens
mid explain JWT tokens
Low capital of germany
DEEP:compare these options
A comma, colon, or any non-space character immediately after the token breaks override detection. MID, explain JWT tokens is not a manual override — the comma prevents the boundary match, and the message is routed normally by the structural kernel.
Boundary examples from the spec:
MessageOverride valid?Reason
MID o che bello✅ YesSpace after token
MID, o che bello❌ NoComma after token
DEEP compare options:✅ YesSpace after token
mid explain JWT❌ NoLowercase
DEEP (alone)✅ YesToken with no trailing content

​
When to Use Each Override

LOW

Fast, minimal answer.Use LOW when you want a direct, no-frills answer and the router might escalate the message to MID or DEEP because of its shape.Example: LOW what is the capital of Japan? → Tokyo.

MID

Default structured answer.Use MID when you want structured, rigorous output but your message is structurally simple and wouldn’t normally trigger anything beyond AUTO_LIGHT.Example: MID explain the difference between TCP and UDP

DEEP

Full decision support, no gating.Use DEEP when your request is semantically complex but structurally simple — the structural kernel routes by shape, not meaning. Using the override skips the gating question entirely.Example: DEEP explain the tradeoffs of eventual consistency

​
Override Behavior in DEEP

When you use the DEEP manual override:
  • The protocol enters full decision support mode immediately — no Switch to DEEP? (yes/no) question
  • AUDIT is always ON — the reply always ends with a 4-line AUDIT block
  • C5 sparse local provenance signals may appear on eligible top-level blocks (column-1 numbered or bullet items) in the response body
  • You have already consented to DEEP by using the override — there is no separate consent step
Structural overlay in manual DEEP:
  • If your message is DEEP with no trailing content: no automatic structural overlay is emitted
  • If your message is DEEP <content>: the structural overlay is derived from <content> after stripping the leading DEEP token — so chips reflect the shape of what you actually asked
Example: 
DEEP I need help choosing between three database engines for a high-read workload with strict latency SLAs
Because this message has no option blocks, step blocks, criteria bullets, or numeric signals in its shape, no structural overlay chips would be emitted — even though the content is clearly complex. This is exactly the scenario where the manual DEEP override adds value: you get full decision support even when the structural kernel would have routed it as AUTO_STANDARD.

​
The No-Inertia Rule

A.D.A.M. recomputes mode on every message. A previous DEEP reply does not keep subsequent messages in DEEP. After a DEEP session, a simple factoid question will be routed as LOW or MID — as if the DEEP session never happened. One-turn exception: A message with no whitespace that ends with ? inherits the previous exact mode (LOW, MID, or DEEP) for one turn only. This exception never applies to MODE: MID -> POSSIBLE DEEP. 
DEEP compare these frameworks in terms of performance, ecosystem, and hiring market

→ MODE: DEEP reply

why?

→ inherits DEEP for this one turn (no whitespace, ends with ?)

what about TypeScript?

→ re-evaluated from scratch (has whitespace)
The one-turn exception applies to the previous exact mode, not to the general direction of the conversation. If the previous reply was MODE: MID -> POSSIBLE DEEP, that mode is never inherited.
Manual override DEEP still respects the full output contract. AUDIT is attached, C5 signals may appear on eligible blocks, and publish-boundary validation still runs. The override changes which mode is forced — it does not bypass the protocol’s output rules, strict invariant checks, or fail-closed behavior.

Build docs developers (and LLMs) love

Get started for freeTalk to us