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.

Manual override lets you take direct control of A.D.A.M.’s response depth for a single turn. Starting a message with exactly LOW, MID, or DEEP followed by boundary whitespace forces that mode for the reply — regardless of what the automatic structural routing would have chosen. This is the fastest way to get a short answer when A.D.A.M. would otherwise go deeper, or to force full decision support without waiting for gating.

​
Syntax

LOW <rest of message>
MID <rest of message>
DEEP <rest of message>
The override token must be the first token of the message. The rest of the message follows after a single space (boundary whitespace). Everything after the space is the actual request content.

​
Examples

Force a minimal, direct answer: 
LOW what is the capital of France
Force a structured but not deep answer: 
MID explain the difference between TCP and UDP
Force full decision support immediately, bypassing gating: 
DEEP compare these options:
A) PostgreSQL
B) MySQL
C) SQLite
Budget: under $100/month
Needs:
- managed hosting
- high availability
- JSONB support

​
Boundary strictness

The override token must be followed by boundary whitespace (a space or line break). A comma, punctuation, or any other non-whitespace character immediately after the token breaks the boundary and the override is not recognized.
InputOverride valid?Why
MID o che bello✅ YesSpace follows token
MID, o che bello❌ NoComma breaks boundary
DEEP (alone, no trailing content)✅ YesValid override, but no trailing source for overlay
deep explain this❌ NoToken is lowercase — case-sensitive
DEEPER explanation❌ NoDEEPER is not a recognized token
Override matching is case-sensitive. low, mid, and deep in lowercase are not recognized as override tokens and will be treated as normal message content.

​
Priority

Manual override sits at priority 5 in A.D.A.M.’s priority ladder:
PriorityRule
1REMOUNT_REPLAY_DETECTED
2TRANSPORT_GATE
3OFF_GATE
4ACTIVE_CONTROL_COMMANDS
5MANUAL_OVERRIDE
6STRUCTURAL_KERNEL
7STATE_DEFAULT
8OUTPUT_CONTRACT
This means manual override wins over structural routing and state defaults, but is subordinate to active control commands (like ADAM PING, ADAM OFF, etc.).

​
DEEP override and AUDIT

Forcing DEEP via manual override follows all normal DEEP output rules:
  • Full decision support response body
  • AUDIT footer appended (since AUDIT_ON is always true in DEEP)
  • Sparse Local Provenance signals on eligible top-level blocks where conditions are met
No gating consent is required. The Switch to DEEP? (yes/no) line is skipped entirely — you are already in DEEP.

​
DEEP override and the structural classification overlay

When DEEP is entered via manual override with trailing content, A.D.A.M. emits the automatic structural classification overlay as the second line of the reply. The overlay source turn is derived from the current user message after stripping the leading DEEP token. This normalization — removing the DEEP prefix before deriving the overlay — is overlay-only. It applies only to the presentation layer and does not alter:
  • The raw current user turn
  • The strict TRACE INPUT output (which always traces the raw turn)
  • The structural kernel computation
When DEEP is used alone (no trailing content), there is no source turn for the overlay, so the overlay is omitted.

​
No-inertia and override

A.D.A.M. recomputes the mode on every user message. A forced DEEP on one turn does not carry forward to the next. The next turn is evaluated from scratch via automatic routing unless you explicitly override again.
There is no “stay in DEEP” state after a manual override. Every turn starts fresh.

​
One-turn inheritance exception

There is one narrow exception to no-inertia:
A message with no whitespace that ends with ? inherits the previous exact LOW, MID, or DEEP for one turn only.
This lets single-word follow-up questions (like Why? or Really?) stay at the same depth as the preceding answer without requiring you to repeat the override token. MODE: MID -> POSSIBLE DEEP is never inherited under this rule — only the three exact modes (LOW, MID, DEEP).

​
Override reference table

OverrideExample inputExpected reply depth
LOWLOW what year was Python first releasedMinimal, direct answer — no structure overhead
MIDMID pros and cons of using a monorepoStructured answer with compact organization
DEEPDEEP compare A) Redis B) Memcached — latency under 1ms, cost <$50/moFull decision support with AUDIT footer and provenance signals on eligible blocks
Use LOW to cut through verbose responses on simple factual questions. Use DEEP when you already know the request is structurally complex and don’t want to go through the gating consent step.

Build docs developers (and LLMs) love

Get started for freeTalk to us