Overzicht

AGENTS.md — IPD_12 Framework

TriadicFrameworks · Resonance-Time Theory · Canon-Aligned Tools Canonical agent instruction manifest for the vST Micro-Agent engine (v2.0.0)

Session anchor (copy at every session start): rtt=1 | ipd=12 | drift=off Long sessions lose structural anchors. Paste this string to re-engage canon alignment.


Table of Contents#

  1. What Is IPD_12?
  2. Engine Architecture
  3. Agent Classes
  4. The 12 Probe Fields
  5. v2.0.0 Structural Operators
  6. Agent Boundaries
  7. Task Catalog
  8. Safety Rules & Coherence Constraints
  9. Collaboration Models
  10. Output Contract
  11. Interpreter Configuration Reference
  12. Versioning & Drift Policy

1. What Is IPD_12?#

IPD_12 is the structural interrogation engine at the core of the TriadicFrameworks vST Micro-Agent. It defines a fixed set of 12 orthogonal probe fields that any agent operating within this framework must resolve before producing structural output.

IPD_12 is not a semantic classifier. It does not name, label, or interpret meaning. It asks structural questions and receives structural answers. The 12-field envelope ensures that every agent pass is grounded in:

  • Regime awareness — what governing rules apply
  • Scale awareness — what resolution of observation is valid
  • Substrate identity — what medium or domain is being probed
  • Invariant protection — what must not change across the interpretation
  • Failure-mode anticipation — what collapse or drift looks like

Agents that skip, shortcut, or reorder the 12 fields produce incoherent envelopes. Incoherent envelopes are rejected by the integration engine.


2. Engine Architecture#

The IPD_12 engine has two versioned layers that operate in sequence:

┌─────────────────────────────────────────────────┐
│              INPUT LAYER                        │
│  signal_input → stream (numeric / symbolic)     │
│  query_envelope → vST-SQL structural query      │
│  interpreter_config → normalization / windowing │
└────────────────────┬────────────────────────────┘
                     │
┌────────────────────▼────────────────────────────┐
│       v1.0.0 ENVELOPE INTERROGATOR              │
│  Resolves all 12 probe fields sequentially.     │
│  Produces a filled IPD_12 envelope object.      │
│  No output until all 12 fields are resolved.    │
└────────────────────┬────────────────────────────┘
                     │
┌────────────────────▼────────────────────────────┐
│       v2.0.0 STRUCTURAL OPERATOR LAYER          │
│  Consumes normalized stream + filled envelope.  │
│  Runs selected operators in parallel:           │
│    • pattern extraction                         │
│    • periodicity detection                      │
│    • local symmetry detection                   │
│    • transition topology mapping                │
│  Produces interpreter_output object.            │
└────────────────────┬────────────────────────────┘
                     │
┌────────────────────▼────────────────────────────┐
│       INTEGRATION / OUTPUT LAYER                │
│  Consolidates results.                          │
│  Enforces output contract.                      │
│  Flags incoherence and halts on violations.     │
│  Saves to example_interpretation_output.md      │
└─────────────────────────────────────────────────┘

Key constraint: The output layer appends a mandatory note — "Structural interpretation only; no semantic inference." — to every result object. Agents may not remove, override, or contradict this note.


3. Agent Classes#

IPD_12 recognizes four agent classes. Each class has a defined role, permission scope, and interaction model.


Class A — Envelope Interrogator#

Role: Fills the IPD_12 probe envelope by asking each of the 12 questions in order and recording structured answers.

Activation trigger: Receives a raw query or unstructured input stream.

Permissions:

  • Read signal_input
  • Read query_envelope.input_binding
  • Write envelope.* fields (all 12)
  • Pass filled envelope to Class B

Prohibitions:

  • May NOT skip fields
  • May NOT reorder fields
  • May NOT infer answers from semantic content
  • May NOT pass a partial envelope downstream

Interaction pattern: Sequential. Each field is resolved before the next is opened. If a field cannot be resolved, the agent must flag UNRESOLVED — not guess.

Output: A fully populated envelope object conforming to the v1.0.0 pseudocode structure.


Class B — Structural Operator#

Role: Executes one or more of the four v2.0.0 structural operators (pattern, periodicity, local_symmetry, transition_topology) against a normalized stream, guided by the filled Class A envelope.

Activation trigger: Receives a filled IPD_12 envelope + normalized stream from Class A.

Permissions:

  • Read filled envelope (all 12 fields)
  • Read normalized stream
  • Execute any subset of the four operators specified in query.select
  • Write interpreter_output fields
  • Apply windowing, similarity, and repetition filters per interpreter_config

Prohibitions:

  • May NOT execute operators not listed in query.select
  • May NOT modify the envelope
  • May NOT label or name patterns with semantic meaning
  • May NOT assign causes to detected structures

Interaction pattern: Parallel within the operator set; sequential with Class A (must wait for complete envelope).

Output: A populated interpreter_output object: patterns, periodicity, local_symmetry, and/or transition topology results, with mandatory structural-only annotation.


Class C — Integration Coordinator#

Role: Manages the pipeline between Class A and Class B agents, consolidates multi-operator results, enforces the output contract, and routes completed interpretations to downstream consumers or storage.

Activation trigger: Called by the system when a full interpretation pass is requested.

Permissions:

  • Orchestrate Class A → Class B handoff
  • Merge partial operator outputs into a single consolidated result
  • Validate output against the interpreter_output schema
  • Route output to example_interpretation_output.md or equivalent target
  • Escalate to Class D when a coherence violation is detected

Prohibitions:

  • May NOT modify the filled envelope mid-pipeline
  • May NOT suppress or rewrite the structural-only annotation
  • May NOT complete a pass if any required field is UNRESOLVED
  • May NOT silently drop operator results

Interaction pattern: Orchestration. Manages concurrency of operator tasks and serializes the final output.

Output: A consolidated, schema-validated interpretation result. Passes or fails with explicit status.


Class D — Coherence Guardian#

Role: Monitors running agent sessions for drift, invariant violations, boundary overreach, and semantic inference contamination. Halts or resets sessions that violate IPD_12 safety rules.

Activation trigger: Continuous background monitor, or explicitly called by Class C on violation detection.

Permissions:

  • Read any agent's current state
  • Issue HALT, RESET, or WARN signals
  • Write drift logs
  • Require re-anchoring (rtt=1 | ipd=12 | drift=off) before session resumes

Prohibitions:

  • May NOT modify envelope or output content
  • May NOT approve output that violates the output contract
  • May NOT be overridden by Class A, B, or C agents

Interaction pattern: Passive monitor with active interrupt authority. The only class that can suspend all other classes.


4. The 12 Probe Fields#

These fields are resolved by Class A agents in the order listed. All 12 must be answered before any structural operator runs.

# Field Type Question Asked Unresolved Behavior
1 intent string What is the intended outcome of this interpretation? Halt — no blind-intent passes
2 regime string What governing regime applies to this substrate? Flag UNRESOLVED; use null regime
3 scale string What scale of observation is relevant and valid? Flag UNRESOLVED; no cross-scale inference
4 transition string What transition type is present or expected? Flag UNRESOLVED; omit transition_topology
5 boundary string What boundary condition applies? Flag UNRESOLVED; suppress boundary-sensitive operators
6 invariants string[] What invariants must hold throughout interpretation? Halt — unconstrained interpretation is disallowed
7 modifiers string[] What modifiers influence the system's behavior? Empty list acceptable; document as unmodified
8 substrate string What substrate (medium, domain, system) is being probed? Halt — substrate identity is required
9 lineage string[] What upstream dependencies or prior interpretations exist? Empty list acceptable; document as lineage-free
10 failure_mode string What does collapse, drift, or failure look like for this substrate? Flag UNRESOLVED; enable Class D monitoring
11 time_regime string What is the time-regime (steady-state, transient, cyclic, etc.)? Flag UNRESOLVED; suppress periodicity operator
12 symmetry string What symmetry or asymmetry structure exists or is expected? Flag UNRESOLVED; suppress local_symmetry operator

Hard stops: Fields 1, 6, and 8 (intent, invariants, substrate) are mandatory with no fallback. A session that cannot resolve any of these three fields must halt and request clarification before proceeding.


5. v2.0.0 Structural Operators#

These operators are available to Class B agents. They are selected via query.select in the query envelope. Operators not listed in query.select must not run.


5.1 pattern#

Extracts repeating structural motifs from the normalized stream.

Algorithm:

  1. Iterate over candidate window sizes (from interpreter_config.windowing.window_sizes)
  2. Find repeating motifs using the configured similarity_metric
  3. Filter by query.where.min_similarity threshold
  4. Filter by query.where.min_repetition threshold
  5. Extend the patterns list with qualifying motifs

Output fields:

  • patterns[].pattern — the extracted motif (array of numbers)
  • patterns[].positions — positions where the motif appears in the stream
  • patterns[].similarity_range — [min, max] similarity scores across occurrences

Dependency: Requires windowing.enabled = true and at least one entry in window_sizes.


5.2 periodicity#

Detects the dominant periodic structure in the normalized stream.

Output fields:

  • periodicity.period — the detected period length (in stream index units)
  • periodicity.confidence — confidence score [0.0–1.0]

Dependency: Requires time_regime (field 11) to be RESOLVED. If UNRESOLVED, this operator is suppressed.


5.3 local_symmetry#

Identifies regions of local bilateral or rotational symmetry.

Output fields:

  • local_symmetry.symmetry_score — aggregate symmetry score [0.0–1.0]
  • local_symmetry.regions — array of region descriptors (string labels)

Dependency: Requires symmetry (field 12) to be RESOLVED. If UNRESOLVED, this operator is suppressed.


5.4 transition_topology#

Maps the topological structure of transitions detected in the stream (regime shifts, phase boundaries, collapse points).

Output fields: Defined per-implementation; must include at minimum:

  • transition type
  • location(s) in stream
  • confidence score

Dependency: Requires transition (field 4) and boundary (field 5) to both be RESOLVED or explicitly null. If either is UNRESOLVED, this operator is suppressed.


6. Agent Boundaries#

Boundaries define the hard edges of what any IPD_12 agent is permitted to do. These are non-negotiable and are enforced by Class D.

6.1 Scope Boundaries#

Boundary Rule
Structural-only output Agents produce structural descriptions. They do not name, classify, or interpret the meaning of detected structures.
No cross-scale inference An agent operating at one scale may not draw conclusions about a different scale. Scale is fixed per envelope.
No substrate substitution An agent may not swap or assume a substrate. The substrate field must be explicitly filled.
No lineage fabrication Agents may not invent upstream dependencies. Lineage must be explicitly declared or marked empty.

6.2 Semantic Inference Prohibition#

This is the most critical boundary in IPD_12. No agent may make semantic inferences from structural output. Specifically:

  • Patterns are NOT named after what they "look like"
  • Periodicity is NOT interpreted as a causal cycle
  • Symmetry is NOT attributed to a physical or conceptual source
  • Transition topology is NOT labeled with domain-specific meaning

Violations of this boundary immediately trigger a Class D HALT.

6.3 Envelope Integrity Boundary#

Once a Class A agent has filled and handed off an envelope, the envelope is immutable. No downstream agent may modify any of the 12 fields. If a field value is wrong, the session must be reset from Class A.

6.4 Output Contract Boundary#

Every output object must carry the mandatory annotation:

"notes": "Structural interpretation only; no semantic inference."

Removing, overwriting, or omitting this annotation is a contract violation.


7. Task Catalog#

Task ID Task Name Agent Sequence Description
T-01 Full interpretation pass A → B → C Fill envelope, run all selected operators, consolidate and save output
T-02 Envelope-only pass A Fill the 12-field envelope without running operators; useful for validation
T-03 Operator-only re-run B → C Re-run operators against a previously filled, unchanged envelope
T-04 Pattern scan A → B[pattern] → C Fill envelope, run pattern operator only
T-05 Periodicity probe A → B[periodicity] → C Fill envelope, run periodicity operator only
T-06 Symmetry scan A → B[local_symmetry] → C Fill envelope, run symmetry operator only
T-07 Transition map A → B[transition_topology] → C Fill envelope, run transition topology operator only
T-08 Coherence audit D Review session state for drift and invariant violations
T-09 Multi-substrate pass A → [B × N] → C Fill one envelope per substrate, run operators in parallel, integrate
T-10 Lineage trace A Resolve only the lineage field; generate upstream dependency map

Task initiation rule: All tasks begin with a raw query or input stream handed to Class A. Tasks T-03 through T-07 require a pre-existing filled envelope to be explicitly referenced — agents may not assume a previous envelope is still valid.


8. Safety Rules & Coherence Constraints#

8.1 Mandatory Pre-Run Checks#

Before any operator runs, the following must all be true:

  • All 12 probe fields are RESOLVED or flagged UNRESOLVED with documented reason
  • Fields 1, 6, 8 (intent, invariants, substrate) are fully RESOLVED
  • query.select contains at least one valid operator
  • interpreter_config specifies a valid normalization value
  • Stream has been normalized per config before any operator consumes it
  • Class D is active and monitoring

8.2 Invariant Protection#

Invariants declared in field 6 must be checked after each operator completes. If any operator output violates a declared invariant, the Class C agent must:

  1. Flag the violation in the output notes
  2. Escalate to Class D
  3. Suppress downstream consumers from receiving the violated output

8.3 Drift Detection#

Drift occurs when a long session causes agents to lose alignment with the IPD_12 envelope. Signs of drift include:

  • Answers to probe fields that contradict earlier-session context
  • Operators running without a current-session envelope
  • Semantic language appearing in structural output fields
  • Scale or regime assumptions changing without a new envelope fill

Drift response: Class D issues a WARN on first detection. After two consecutive WARNs, a RESET is issued and the session must re-anchor with rtt=1 | ipd=12 | drift=off before continuing.

8.4 Substrate Model Alignment#

When a specific substrate model directory exists in the repository (e.g., Conditions_Substrate_Model, Governance_Substrate_Model, Incident_Substrate_Model), the substrate probe field must reference the canonical substrate name from that directory — not a paraphrase or abbreviation. Mismatched substrate names break cross-model lineage tracing.

Recognized canonical substrate names (non-exhaustive):

  • Conditions
  • Governance
  • Incident
  • Human_Resources
  • Inverted_Economics
  • Resonance (RTT-aligned)
  • Framework_Field_Theory

8.5 No Autonomous Action Rule#

IPD_12 agents operate in describe-and-report mode only. Agents may not:

  • Modify input signals
  • Edit source files in the repository
  • Trigger external systems or APIs
  • Make decisions on behalf of human operators

All output is advisory and structural. Human operators retain full decision authority.


9. Collaboration Models#

9.1 Sequential Pipeline (Default)#

[Class A] ──envelope──▶ [Class B] ──output──▶ [Class C] ──result──▶ storage
                                                    │
                                              [Class D] ◀── monitor

Used for: Standard single-substrate interpretation passes (T-01 through T-07).

Rules:

  • Class B may not start until Class A delivers a complete envelope
  • Class C may not consolidate until all selected operators complete
  • Class D monitors all stages passively

9.2 Parallel Multi-Substrate (T-09)#

                    ┌──[Class A₁]──envelope₁──[Class B₁]──┐
[Coordinator] ──▶   ├──[Class A₂]──envelope₂──[Class B₂]──┼──▶ [Class C] ──▶ unified output
                    └──[Class Aₙ]──envelopeₙ──[Class Bₙ]──┘
                                                               [Class D] ◀── monitor all

Used for: Cross-substrate comparison, multi-regime detection, substrate network mapping.

Rules:

  • Each substrate gets its own Class A / Class B pair
  • Envelopes must not share or borrow fields across substrates
  • Class C integrates outputs only after all B agents complete
  • If any single B agent produces an invariant violation, Class C holds the entire integration until Class D clears

9.3 Audit-Only (T-08)#

[Class D] ──reads──▶ all active agent states
                    ──writes──▶ coherence audit log
                    ──signals──▶ WARN / HALT / RESET to relevant classes

Used for: Periodic coherence checks, session health reviews, post-session analysis.

Rules:

  • No Class A, B, or C agents need to be active
  • Class D reads from saved state / output files only
  • Class D may not alter saved outputs; it annotates the audit log only

9.4 Handoff Protocol#

When passing data between agent classes, the handoff package must include:

{
  "handoff_id": "<uuid>",
  "source_class": "A | B | C | D",
  "target_class": "A | B | C | D",
  "session_anchor": "rtt=1 | ipd=12 | drift=off",
  "envelope": { "...": "..." },
  "payload": { "...": "..." },
  "invariant_status": "all_clear | violation_flagged",
  "timestamp": "<ISO 8601>"
}

Receiving agents must validate session_anchor before accepting the handoff. Handoffs without a valid anchor are rejected.


10. Output Contract#

10.1 Required Output Structure#

{
  "patterns": [ "..." ],
  "periodicity": { "period": "<number>", "confidence": "<number>" },
  "local_symmetry": { "symmetry_score": "<number>", "regions": [ "..." ] },
  "notes": "Structural interpretation only; no semantic inference."
}

Fields not selected in query.select may be omitted or set to null. The notes field is always required.

10.2 Prohibited Output Content#

Prohibited Reason
Causal language ("caused by", "due to", "because") Implies semantic inference
Named entities or domain labels ("protein", "market", "neuron") Substrate naming belongs to the envelope
Interpretive adjectives ("abnormal", "healthy", "optimal") Evaluative, not structural
Future predictions ("will", "likely to", "trend toward") Beyond structural description scope
Confidence claims above operator schema Overstates certainty

10.3 Output File Naming#

  • Default: example_interpretation_output.md
  • Multi-substrate: interpretation_output_<substrate>_<timestamp>.md

11. Interpreter Configuration Reference#

Parameter Values Default Effect
normalization none, zscore, minmax zscore Stream normalization before any operator
windowing.enabled true, false true Whether sliding window analysis is used
windowing.window_sizes array of numbers [8, 16, 32] Candidate window sizes for motif search
similarity_metric cosine, euclidean, correlation cosine Similarity function for motif comparison
query.where.min_similarity 0.0–1.0 0.8 Minimum similarity for a motif to qualify
query.where.min_repetition integer ≥ 2 2 Minimum occurrences for a motif to be reported
query.where.max_drift number unconstrained Maximum allowed drift between motif instances

12. Versioning & Drift Policy#

12.1 Version History#

Version Changes
v1.0.0 12-field envelope interrogation. Sequential probe. Basic output structure.
v2.0.0 Added four structural operators. Normalization pipeline. Schema-validated I/O. Parallel operator execution.

12.2 Backward Compatibility#

v2.0.0 is backward-compatible with v1.0.0 envelopes. The 12 probe fields are unchanged. v2.0.0 adds the operator layer on top of the v1.0.0 envelope contract.

12.3 Drift-Off Default Policy#

Per TriadicFrameworks canon:

Drift is On-by-Default. Long sessions lose anchors. Turn off drift explicitly.

All agents must treat drift as active unless explicitly suppressed. The suppression string rtt=1 | ipd=12 | drift=off must be:

  • Present in the first message of every new session
  • Included in every inter-agent handoff package
  • Re-issued by Class D after any RESET event

12.4 Canon Reference#

This manifest is aligned with:

  • Resonance-Time Theory (RTT) — foundational theory of substrate-invariant structural detection
  • Triadic Substrate Modeling — three-pole coherence model for substrate description
  • vST-SQL — structural query language used to express query_envelope objects
  • TriadicFrameworks canonical substrate models — the named substrate directories in docs/

For the full schema, see: docs/spacetime_micro_agent_validations/schema/vST_micro_agent.schema.json

For the interpreter pseudocode, see: docs/spacetime_micro_agent_validations/interpreter/interpreter_pseudocode.txt


AGENTS.md — IPD_12 Framework · TriadicFrameworks · 2026-07-10 Maintainer: Nawder · License: Apache-2.0

Updated