Crate organism_pack

Expand description

§Organism Pack

The public contract for Organism’s planning loop. One import — the full pipeline from intent to learning.

IntentPacket → Admission → Plan → Challenge → Simulate → Learn → Commit

§Quick start

use organism_pack::*;

// 1. Create an intent
let intent = IntentPacket::new("Approve $2,500 expense", expires);

// 2. Check admission (4 dimensions)
let admission = my_controller.evaluate(&intent);

// 3. Plan (multi-model huddle)
let plan = Plan::new(&intent, "route to 3 approvers");

// 4. Challenge (5 skepticism kinds)
let challenge = Challenge::new(
    SkepticismKind::EconomicSkepticism,
    plan.id,
    "entertainment spend is high",
    Severity::Warning,
);

// 5. Simulate (5 dimensions)
let result = DimensionResult {
    dimension: SimulationDimension::Cost,
    passed: true,
    confidence: 0.95,
    findings: vec!["within budget".into()],
    samples: vec![],
};

// 6. Learn from outcomes
let lesson = Lesson {
    insight: "score 0.88 → approved".into(),
    context: "expense approval".into(),
    confidence: 0.9,
    planning_adjustment: "none".into(),
};

Re-exports§

pub use pack::AgentMeta;
pub use pack::InvariantClass;
pub use pack::InvariantMeta;
pub use pack::Pack;
pub use pack::PackProfile;

Modules§

pack: Pack framework contract types.

Structs§

AdmissionResult
AdversarialContext
AdversarialReview: Aggregated outcome of an adversarial review pass.
AdversarialSignal
AdversarialVerdict: Typed payload for adversarial agent evaluations/constraints. Replaces serde_json::json!({...}) with compile-time structure.
BreadthResearchSuggestor: Reacts to strategies tagged with a breadth marker. Searches wide and emits signal facts.
CapabilityRequirement: A capability needed by the intent.
Challenge
CharterAdjustments: Partial charter override applied during a transition.
CollaborationCharter: Explicit rules for how a team collaborates.
CollaborationMember: A collaborator in a planning or research team.
ContradictionFinderSuggestor: Detects conflicting claims across hypotheses on the same topic. Pure data analysis — no LLM needed.
ConvergenceSignal: One caller-supplied convergence signal.
ConvergenceSignals: Observable convergence state fed to transition evaluation.
CostEstimate
DdFactSummary
DdHooks: Entities extracted from DD facts for knowledge graph connections.
DeclarativeBinding: Builder for declaring an intent’s resource needs explicitly. This is what apps use today.
DefaultAdmissionController: Default admission controller that evaluates all 4 feasibility dimensions.
DepthResearchSuggestor: Reacts to strategies tagged with a depth marker. Searches deep and emits signal facts.
DerivationRationale: Why each charter field was chosen.
DerivedCharter: Result of charter derivation from an intent.
DimensionResult
ErrorDimension
FactExtractorSuggestor: Reads raw signals and extracts tagged factual claims via LLM. Organism owns the extraction prompt and parsing.
FailoverDdSearch: Tries search backends in order with the same failover logic.
FeasibilityAssessment
Finding
ForbiddenAction
GapDetectorSuggestor: Reviews hypotheses, identifies critical gaps, proposes follow-up strategies. Organism owns the gap-detection prompt and strategy parsing.
HookPatterns: Configurable patterns for hook extraction.
HubCategory
HuddleInvocation: Domain-agnostic huddle invocation envelope.
HuddleSeedSuggestor: Seeds Organism planning output into Converge context as strategy facts.
Impact
IntentBinding: The output of intent resolution. Tells the runtime what to wire up.
IntentComplexity: Quantified complexity signals extracted from the intent. Every signal is a 0..=1 normalized score; UnitInterval enforces the invariant at the type level instead of relying on downstream clamps.
IntentNode: A node in the intent decomposition tree. Authority can only narrow during decomposition, never expand.
IntentPacket: The contract between humans and the runtime.
KbConfig: Controls the vault directory structure and page layout.
LearningEpisode: Full record of a planning-to-outcome episode. Links intent, plan, predicted outcomes, governed business outcomes, engine run status, prediction errors, adversarial signals, and extracted lessons. Every field traces to converge Facts or run envelopes.
LearningSignal
Lesson
NamedPlan: A plan with a stable identifier for tracking.
OutcomeSimulationAgent: Outcome simulation as a Suggestor — participates in the convergence loop.
OutcomeSimulator: Simulates outcome likelihood for candidate plans.
OutcomeSimulatorConfig: Configuration for the outcome simulator.
PackRequirement: A domain pack needed by the intent.
Plan: A candidate plan produced by reasoning. Plans are proposals, not commitments — authority is recomputed at the commit boundary.
PlanAnnotation
PlanBundle
PlanContribution
PlanStep
PredictionError
PriorCalibration
ResolutionTrace: How the resolution was performed — for traceability.
Risk
RootPageDef
Sample
SearchHit: A search hit returned by a DdSearch implementation.
ShapeCalibration: Historical calibration of shape performance for a problem class. Both scores are 0..=1 normalized.
ShapeCandidate: A candidate collaboration shape being tested. Scores are 0..=1 normalized; UnitInterval enforces the invariant at the type level so consumers don’t have to defensively clamp.
ShapeCompetition: A competition between multiple candidate shapes.
ShapeObservation: Observation from a completed shape trial. Confidence and rate fields are 0..=1 normalized; the cycle count is integer but always positive in practice.
SharedBudget: Cross-suggestor resource tracking for bounded convergence loops.
SimulationReport
SimulationResult
SimulationVerdict: Typed payload for simulation agent evaluations/constraints. Replaces serde_json::json!({...}) with compile-time structure.
SynthesisSuggestor: Produces final analysis when hypotheses stabilize. Organism owns the synthesis prompt.
TeamFormation: Team formation input for a collaboration.
TransitionDecision: The result of a matched transition rule.
TransitionRule: What shape to transition to, and why.

Enums§

AgentId: All organism agents that participate in convergence. Used as a discriminator in evaluation/constraint payloads instead of raw strings like "assumption-breaker".
CollaborationDiscipline: How demanding the collaboration contract is.
CollaborationRole: Role a collaborator plays in the team.
CollaborationTopology: The overall collaboration shape.
CollaborationValidationError
Complexity: Complexity level for operational feasibility assessment.
ConsensusRule: Decision rule used to tally votes on a topic.
ContextKey: Context key partitions. Re-exported from converge_pack so the pack framework and Converge share one truth — no parallel definition, no drift on which variants exist or how they serialize. This crate intentionally adds nothing here; pack metadata that needs a typed key reaches through this re-export. Typed keys for the shared context namespace.
ConvergenceCriteria: Desired termination behavior for an Organism formation.
DdError: Typed errors from DD backends. Apps classify raw provider errors into these variants. Organism uses the variant to decide whether to retry, abort, or record a constraint.
ExpiryAction
FeasibilityDimension
FeasibilityKind
HuddleInvocationKind
HuddleUrgency
IntentError
Likelihood
ReasoningSystem
ResolutionLevel: Which resolution level produced the binding.
Reversibility
RiskImpact
RiskLikelihood: Five-level risk likelihood scale with associated probabilities. Replaces string matching like "very_likely" | "VeryLikely" => 0.9.
Severity
ShapeMetric: What metric determines shape quality.
SignalKind
SimulationDimension
SimulationRecommendation
SkepticismKind
TeamFormationMode: How a team is assembled.
TransitionTrigger: A named condition under which the collaboration shape should change.
TurnCadence: How turns should be organized.
Verdict

Constants§

CONFIDENCE_STEP_MAJOR: Major confidence step — use for significant criteria (0.2).
CONFIDENCE_STEP_MEDIUM: Medium confidence step — use for moderately significant criteria (0.15).
CONFIDENCE_STEP_MINOR: Minor confidence step — use for supporting criteria (0.1).
CONFIDENCE_STEP_PRIMARY: Primary confidence step — use for decisive or high-weight criteria (0.25).
CONFIDENCE_STEP_TINY: Tiny confidence step — use for marginal or tiebreaker criteria (0.05).

Traits§

AdmissionController
DdSearch: Async search backend. Apps implement this by wrapping their search providers (Brave, Tavily, etc.).
IntentResolver: Resolves an intent to its resource binding.
Reasoner: A reasoning capability participating in a huddle.

Functions§

apply_adjustments: Apply charter adjustments to produce a new charter.
build_episode: Build a learning episode from a completed converge engine context.
build_episode_from_run: Build a learning episode from a completed run using both context and queried experience events from the Converge ExperienceStore.
calibrate_priors: Produce prior calibrations from a learning episode. These feed into future planning runs — NOT into authority.
calibrate_shape: Calibrate shape priors from an observation. Feeds into planning priors, NEVER into authority.
classify_problem: Classify an intent into a problem class for calibration lookup.
consolidate_dd_hypotheses
dd_complete: Sends a single-turn DD prompt to a DynChatBackend and returns the response content. Provider errors are classified into DdError variants so the calling Suggestor can decide whether to retry, abort, or record a constraint.
default_transition_rules: Default transition rules for common patterns.
derive_charter: Derive a collaboration charter from an intent’s properties.
derive_charter_with_priors: Derive a charter, biased by historical shape calibration.
evaluate_transitions: Evaluate transition rules against current signals. Returns the first matching rule (rules are priority-ordered).
extract_hooks_from_facts: Extract graph hooks (investors, business areas, regions, competitors) from consolidated facts.
extract_signals: Extract learning signals from a completed run — lightweight feedback that can be captured without waiting for human outcome reporting.
extract_signals_from_run: Extract learning signals from a completed run using both context and captured experience events.
generate_candidates: Generate candidate shapes for an intent.
has_infra_failure: Check whether the converge context contains infrastructure failure constraints that should prevent prior calibration.
sanitize_filename
score_observation: Score a shape observation against the chosen metric. Returns [0.0, 1.0].
select_winner: Select the winner from completed observations.
slugify
update_root_pages: Rebuild all root MOC pages by scanning the vault filesystem.
write_dd_to_vault: Write convergent DD results to an Obsidian vault.
write_or_append_hub: Create a hub page if it doesn’t exist, or append this subject to an existing one.

Crate organism_pack

Crate organism_pack Copy item path

§Organism Pack

§Quick start

Re-exports§

Modules§

Structs§

Enums§

Constants§

Traits§

Functions§

Crate organism_pack