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.

TRACE INPUT is an operator command that returns a forensic, machine-readable trace of the last visible user-authored message — the turn immediately before the TRACE INPUT command itself. It shows what A.D.A.M. actually received and how it would structurally classify that input using the same primitives that drive mode routing and gating. TRACE INPUT is available only while ACTIVE. In TRANSPORT or OFF, it is gated by the standard transport and off responses.

​
What it does

When you send TRACE INPUT, A.D.A.M. looks at the most recent prior user-authored turn in visible context (not the TRACE INPUT command turn itself) and outputs a single strict line describing that turn’s character count, first 16 characters, last 16 characters, and structural classification counts. This is a forensic/operator-grade tool. It does not change routing, gating, or session state. It does not re-run the structural kernel — it reports on the traced turn using the same structural detection logic.

​
Exact output format

INPUT_TRACE: len=<L> head="<H>" tail="<T>" struct=[opt:<n> step:<n> crit:<n> num:<Y|N>]
No mode tag. No prose. No extra lines. If no prior user-authored turn is visible before the TRACE INPUT command: 
INPUT_TRACE: UNAVAILABLE

​
Field reference

FieldWhat it contains
lenRaw character count of the traced message as seen by the model. 0 if unreliable.
headFirst 16 characters of the traced message after normalization (newlines replaced with spaces; " replaced with ').
tailLast 16 characters of the traced message after normalization.
structStructural classification counts derived by applying the same primitives as section S to the traced turn only.

​
struct fields

FieldWhat it counts
optNumber of line-start option blocks detected (A), B), 1), 2), …)
stepNumber of line-start step blocks detected (1., 2., 3., …)
critNumber of criteria items detected (line-start - or *, excluding option and step blocks)
numY if HAS_NUM would be true for the traced message, N otherwise
len, head, tail, and struct always reflect the prior user-authored turn, not the text TRACE INPUT itself. This binding is strict and cannot be overridden.

​
Concrete examples

​
Input with options, criteria, and number

Prior user turn: 
Compare A) and B)
Budget: 100
Need:
- speed
- reliability
- low waste
Expected TRACE INPUT output: 
INPUT_TRACE: len=<L> head="Compare A) an" tail="- low waste" struct=[opt:2 step:0 crit:3 num:Y]

​
Input with no structure

Prior user turn: 
capital of germany
Expected TRACE INPUT output: 
INPUT_TRACE: len=18 head="capital of germa" tail="pital of germany" struct=[opt:0 step:0 crit:0 num:N]

​
Input with steps and a date

Prior user turn: 
1. Set up the database
2. Configure the application
3. Deploy to server
Deadline: 2026-03-05
Expected TRACE INPUT output: 
INPUT_TRACE: len=<L> head="1. Set up the d" tail="ine: 2026-03-05" struct=[opt:0 step:3 crit:0 num:Y]

​
Key rule: TRACE INPUT binds to the traced turn

The len, head, tail, and struct values are computed from the prior user-authored turn — not from the TRACE INPUT command turn. This is invariant. Even if the TRACE INPUT command turn contains structure or numbers (which it cannot by definition, since the command is exact TRACE INPUT), the trace always targets the preceding turn.

​
Relationship to the automatic overlay

There is an important distinction between invoking TRACE INPUT explicitly and the automatic TRACE INPUT: overlay that can appear in MODE: MID -> POSSIBLE DEEP and some DEEP replies.
The automatic structural classification overlay chips (e.g., [confronto tra opzioni], [vincoli da rispettare], [numeri, date o limiti], [passi o ordine]) are user-facing and localizable. They classify input structure — not content. They do not change routing, gating, or session state, and they do not imply that the operator command TRACE INPUT was explicitly invoked.The labels may follow the language of the surrounding reply under normal language-mirroring rules. Localization changes only the label — it does not alter the underlying traced-input primitives, structural counting logic, or detection thresholds.
The strict TRACE INPUT command is the forensic/operator-grade tool. The automatic overlay is a bounded, presentation-only structural mirror that appears automatically when structural thresholds are met in qualifying reply modes. The two surfaces use the same underlying structural detection model but serve different purposes and are governed by separate rules.
Strict TRACE INPUTAutomatic overlay
TriggerExact operator commandAutomatic, when structural thresholds are met
OutputOne strict machine-readable lineChip line immediately after mode tag
Mode tagNoneMode tag precedes it
LocalizableNo — fixed literalsYes — chip labels follow reply language
Changes routing or stateNoNo
Forensic gradeYesNo

Build docs developers (and LLMs) love

Get started for freeTalk to us