TraceStack
An end-to-end accountability pipeline with two pre-pipeline discovery instruments and five pipeline stages. Every obligation traced, every contribution attributed, every handoff audited. The chain from authoritative text to operational action — mapped, analyzed, and proved.
The Problem
Organizations using AI to assist with regulatory enforcement, claims processing, prior authorization, procurement, and compliance review face a growing accountability gap. AI tools generate checklists, flag violations, extract obligations, and prioritize decisions. But no system verifies the chain between the authoritative source and the operational action.
When a decision is challenged, the organization needs to demonstrate: what source was used, what was extracted from it, who or what contributed to the decision, and whether the chain from source to action held together. Today, that chain is reconstructed manually, inconsistently, and only when challenged.
Before the chain can even be verified, the source document must be structurally understood — what entities and relationships does it contain, and what structural gaps might exist that no extraction process will catch? These are discovery questions. They precede verification. No existing platform asks them.
California's 2026 AI liability, privacy, and procurement developments create a clear accountability direction: agencies and vendors using AI in enforcement-adjacent workflows will need defensible records showing what the AI produced, what humans reviewed, and how regulatory obligations were translated into operational decisions.
What TraceStack Is
TraceStack is a discovery-to-proof system for regulated AI workflows. Two optional discovery instruments map and analyze source structure before the pipeline begins. Five pipeline stages prove that the chain from authoritative text to operational action was structured, attributable, reconstructible, and audited.
Existing platforms manage inspections, permits, claims, and decisions. TraceStack maps the landscape of what's in the source, surfaces what might be missing, then proves the record behind those actions was lawfully and traceably formed from authoritative text.
What TraceStack proves
What structure the source contains. What unstated structure would improve coherence. Which authority was used. What obligations were extracted. What evidence was required. What record was created. Who or what contributed. What version was active. What changed. What broke. Whether the final record is promotable.
What TraceStack does not prove
That the legal, clinical, contractual, or regulatory interpretation is correct. Content correctness remains the responsibility of validated extraction, subject-matter review, and the governing institution.
Discovery Instruments
Before the accountability pipeline begins, two optional instruments perform structural discovery on the source material. They are not pipeline stages. They do not produce evidentiary artifacts. They produce readings that humans use to understand the landscape before verification starts.
Topology maps what is stated. Inkling surfaces what is not. Both are optional. Both feed the pipeline. The progression is sequential — Topology produces the structure, Inkling reads the structure and surfaces gaps — but an operator can skip Inkling and proceed directly from Topology to RBAT.
Topology maps the entities, relationships, and structural claims present in source material through guided elicitation. Each relationship is graded High, Medium, or Low confidence based on how explicitly it was stated. A verification pass catches structural gaps. The output is an unverified structural map — not evidence, not interpretation.
Produces: TOPOLOGY — UNVERIFIED STRUCTURE
Method adapted from Gene Bellinger's Aha! Mystery protocol (Systems Thinking World).
Inkling performs coherence analysis on a verified Topology export. It surfaces structural gaps, missing relationships, and latent patterns that no extraction process would catch because they were never stated in the source material. It does not modify the topology. It does not make causal claims. It does not generate signals from dialogue or document text — only from topology structure.
Produces: INKLING — STRUCTURAL COHERENCE SIGNAL
"Required" always means "required for structural coherence within the current topology," not required by law, policy, or good practice.
Coherence analysis approach informed by the Apperception Engine (Evans et al., 2020).
Five Pipeline Stages
The chain from authoritative text to operational action has five structurally distinct failure points. Each requires a different kind of verification. No single tool addresses all five.
RBAT tests whether obligation boundaries align with a declared structural rule beyond chance, clustering, density, and simpler competing rules. If the source is structurally incoherent, the pipeline halts before anyone builds infrastructure around it.
Source structure supports extraction · Boundary claims outperform baselines · Extraction should proceed, warn, or halt
DDRP extracts obligation-to-evidence mappings deterministically. Each record includes source coordinate, obligation type, actor, evidence requirement, and threshold. Same source plus same rules equals same output.
What obligation was extracted · Where it came from in the source · What evidence satisfies or violates it
DAS stores obligations, evidence requirements, decisions, citations, and histories as linked records. Obligations link to sources. Evidence links to obligations. Decisions link to evidence. Actions link to decisions.
Records are relationally linked · Decisions have an evidence basis · A later reviewer can reconstruct the chain
CAAP records whether an artifact was AI-generated, human-generated, AI-generated and human-reviewed, or human-generated with AI assistance. Attribution at record level, not pipeline-run level.
What AI contributed · What humans reviewed or changed · Who is accountable for each promoted record
PAL audits the connections between stages: hash consistency, flag propagation, source traceability, attribution completeness, coordinate consistency, and temporal integrity. It does not re-run the pipeline. It verifies the handoffs.
Final record is promotable · Warnings were consumed or overridden · Chain is INTACT, DEGRADED, BROKEN, or QUARANTINED
What Existing Platforms Do Not Do
We have not identified an existing environmental-health, claims, or procurement platform that begins at source-document structural discovery, performs deterministic obligation extraction, preserves relational obligation/evidence/citation traceability, records record-level AI/human attribution, and audits whole-chain integrity.
| Capability | Accela / Tyler / GovOwl | TraceStack |
|---|---|---|
| Inspection management | Yes | Not in scope |
| Permit tracking | Yes | Not in scope |
| Source-document structural mapping | No | Topology |
| Structural coherence analysis | No | Inkling |
| Source-document structural coherence test | No | RBAT |
| Deterministic obligation extraction with source coordinates | No | DDRP |
| Relational obligation/evidence/citation substrate | No | DAS |
| Record-level AI/human attribution | No | CAAP |
| Whole-chain integrity audit | No | PAL |
TraceStack does not compete with these platforms. It provides the discovery and accountability layer underneath them. The organization keeps its existing system. TraceStack sits beneath and provides the verified chain.
Licensing Model
TraceStack is positioned as a licensing play to existing platform vendors, not as a direct-to-agency product. One licensing deal with a platform like Accela or Tyler reaches hundreds of agencies without direct sales overhead.
IP Protection
- Methodology published under CC BY-NC 4.0, establishing public priority and restricting commercial use under the license
- Implementation details, integration architecture, validators, and operational tooling intended to remain proprietary
- DOI-timestamped prior art on all pipeline components
- Licensing terms prohibit sublicensing, attribution stripping, or standalone competition
Publication Map
The following publications establish the intellectual foundation and prior art for the TraceStack pipeline and its discovery instruments.
Live Demonstrations
Technical Review and Implementation Discussion
TraceStack is published, timestamped, and priority-established. The methodology is citable, inspectable, and designed for evaluation against existing regulated workflows.