home strategic briefing
spine sovereignty map motion map lexicon identity graph provenance physics engine
Screen 06 · Epistemic Constraint · Audit Spine

The Provenance Layer

The traceable lineage of every signal, weight, threshold, and capital decision inside the Revenue OS. If gravity governs what is allowed, provenance governs what is true. Not a standalone layer — a horizontal audit plane touching every component.

Definition
Provenance is the discipline of origin. Every signal has lineage. Every probability has decomposition. Every escalation has a threshold record. Every capital shift has an economic trace. Autonomy without provenance is opacity. Opacity without governance is risk.
Why Provenance Exists
AI-native GTM multiplies signal volume and execution speed
Without Provenance
Probability becomes opaque — scores exist but cannot be decomposed. Threshold drift goes unnoticed — gates shift without record. Vendor logic contaminates judgment — external weights bleed into internal decisions. Forecasting becomes narrative — no traceable path from prediction to outcome.
With Provenance
Every probability score can be decomposed into contributing signals and weights. Every routing decision can be reconstructed from threshold to execution. Every correction event can be explained with cohort, error, and delta. Every capital reallocation has traceable logic from SPI shift to budget movement.
Provenance Overlay Across Seven Layers
Horizontal audit plane — not a vertical layer
Spine Layer Primary Function Agent Provenance Domain Required Audit Fields
L01 Feedback Error correction, reweighting Learning Agent Model Provenance cohort_window error_threshold weight_delta version_change approver
L02 Orchestration Velocity, routing, escalation Qualification + Motion Routing Provenance threshold_active capacity_state agent_fired account_state territory
L03 Positioning Value translation Narrative Agent Model Provenance variant_id cluster_target test_window lift_measured
L04 Signal Dual probability engine Signal Agent Signal Provenance sources[] weights{} decay_λ icp_mod model_ver timestamp
L05 ICP Economic clustering ICP Agent Economic Provenance density_inputs{} receptivity_inputs{} cluster_version
L06 Gravity P&L constraints P&L Agent Economic Provenance spi_delta cac_assumption coverage_multiple margin_constraint
Vendor Inputs External telemetry — Vendor Provenance vendor_id signal_type raw_value normalization internal_weight
Five Provenance Domains
Each domain produces immutable audit records
1 Signal Provenance Layer 04 — Signal Interpretation
Every APS must carry metadata. No composite probability exists without ancestry. If a probability score cannot be reverse-engineered, it is disallowed.
sources[]Structural + behavioral
weights{}Coefficients at score time
decay_constantsλ values per signal
icp_modifiersCluster-level adjustments
model_versionLogic version ID
timestampScore calculation time
2 Model Provenance Layer 01 — Revenue Feedback
When correction triggers, record the full context. Adaptive calibration is bounded — provenance records the bounds.
cohort_window90-day rolling scope
cluster_idWhich segment triggered
error_thresholdWhat limit was breached
weight_deltaExact adjustment made
version_changeOld → new model version
approvernull if autonomous, ID if gated
3 Routing Provenance Layer 02 — Orchestration
When an account escalates, the decision chain must be reconstructable. Every routing action traceable from signal through gate to pod assignment.
threshold_activeGate criteria at decision time
capacity_statePod/territory load
agent_firedWhich agent triggered action
account_stateLifecycle at decision time
territory_assignedPod + rep mapping
motion_trackWhich lane + channel
4 Economic Provenance Layer 06 — P&L Gravity
When capital reallocates, trace the economic logic. Economic gravity cannot drift silently.
spi_deltaWhich segment SPI shifted
cac_assumptionWhat CAC threshold applied
coverage_multiplePipeline coverage at decision
margin_constraintWhich floor triggered
5 Vendor Provenance Boundary Sovereignty Map Alignment
Vendor telemetry is allowed. Vendor logic is not. Every vendor-fed signal must be labeled, normalized, internally weighted, and separated from decision authority.
vendor_idSource system identifier
signal_typeWhat category of data
raw_valueUnmodified vendor output
normalizationInternal transform applied
internal_weightWeight assigned by Signal Agent
Architectural Rule — Non-Negotiable
No autonomous action without provenance logging. Agents may optimize. Agents may not obscure. Every threshold adjustment, budget shift, segment reclassification, and probability recalibration must produce an immutable log entry. Provenance is non-optional.
Provenance vs Attribution
Your architecture already demotes attribution. Provenance formalizes the replacement.
Attribution (Demoted)
"What touched the deal?"
Descriptive. Historical credit allocation. Backward-looking. Feeds calibration only — does not govern budget or execution decisions.
Provenance (Structural)
"What changed the probability, and why?"
Structural. Forward-looking causality chain. Traceable from signal origin through weight application to decision output. Governs trust in the system.
Minimum Viable Provenance
Before any agent is allowed to execute
1
Score Decomposition
Any APS explainable to contributing signals + weights
2
Decision Chain
Every routing action traceable from signal → gate → pod
3
Version Registry
Model versions tracked. Weight changes logged with before/after
4
Vendor Labeling
Every external signal tagged with source and normalization
5
Immutable Store
Audit records append-only. No retroactive modification
Interview Framing
"Every probability score in our system is decomposable. Every routing decision is reconstructable. Every correction event is explainable. Provenance is what prevents an AI-native GTM from becoming another black box. It makes the system more transparent than the one it replaces."
Why This Matters Strategically
You are arguing for paradigm replacement. Paradigm replacement only succeeds if the new system is more transparent than the old one. Provenance is what prevents your architecture from becoming another black box. It makes dissent defensible. Once you formalize provenance, agent deployment becomes a compliance decision, not a creative one.