traces.solutions
§ Library

Kindred ideas

One relational map of TRACE — its own domain instances alongside independent external work that arrives at the same conclusions. Every entry is read through a single lens: the layers, principles, and metrics it engages, where TRACE adds measurable rigour, and what it should borrow back.

The engages chips reference the same L1–L4, T-R-A-C-E + parsimony, and trust-metric definitions used across this site.

Framework
TRACE itself — the anchor every other entry maps to.

TRACE

Framework
Trustworthy, Reasoned, Accountable, Context-bound, Escalated AI
Zabolotnii et al., 2026

A four-layer architecture (L1 deterministic core · L2a/L2b right-sized learned components · L3 tiered orchestration · L4 human supervision) that makes AI behaviour in operationally critical domains measurable, auditable, and safe by design. Six design principles (five in the acronym plus Model Parsimony) and seventeen trust metrics turn trust properties into measured quantities.

layers
L1L2aL2bL3L4
principles
TRACEMP
metrics
RCR · RCI · UTC · CRP · CFI · IPSR · EP · TCC · FPA · RBI · OvR · SNR · ETC · CE · ABC · OSI · CPR
Anchor of the library. Paper 1 (framework synthesis) — arXiv:2605.03838 (submitted May 2026); this site is the companion.
Instance
A domain deployment of TRACE.

Clinical decision support

Instance · Clinical decision support
Physician as final authority (Instance A) · foundational
Zabolotnii, Holinko, Antonenko

The foundational instance. Rule-oriented clinical logic (L1, FUTURE-AI-aligned traceability); risk scores, vital-sign time series and lab classifiers plus clinical-note NER (L2a); an LLM validator that checks free-text notes for coherence against the patient's anamnesis (L2b); routine cases via calibrated L2a, borderline → L2b, joint high-risk ∧ high-confidence or L2a/L2b inconsistency → mandatory clinician handoff (L3); physician as final reviewer, aligned with the TRIAD human–AI framework (L4).

layers
L1L2aL2bL3L4
principles
TRACEMP
metrics
ETC · OvR · CE · CFI · CRP · EP · FPA · RBI · ABC · OSI
Contributes
  • ·Empirical anchor for the framework — the first validated instantiation.
  • ·Concrete mapping of TRACE onto external clinical-AI standards: FUTURE-AI traceability and the TRIAD human–AI collaboration framework.
Open / to reconcile
  • ·Paper 0 under review — the metrological validation is written up, not yet peer-accepted.
Foundational instance. Paper 0 — IEEE Instrumentation & Measurement Magazine special issue "A Measure of Trust in Healthcare"; under review (Sep 2026).

Industrial multi-domain — upstream oil & gas

Instance · Industrial multi-domain
Driller / supervisor / HSE escalation (Instance B) · patent pending
Shcherban (lead), Zabolotnii

A foundational instance running three sub-domains at once (technology, operations, administrative). A physics-informed control core (L1); classical ML dominant over MWD/LWD and equipment signals — anomaly detection, predictive maintenance, residual-life (L2a); LLM validators for incident narratives and contract diff (L2b); cost-tiered per-sub-domain routing with sustained-anomaly and SLA-breach escalation (L3); driller / supervisor / HSE officer and lawyer / procurement reviewers under DNV-RP-0671 governance (L4). The dominant layer shifts with the type of evidence.

layers
L1L2aL2bL3L4
principles
TRACEMP
metrics
RCR · RCI · EP · TCC · FPA · OvR · RBI · ABC · OSI · ETC · CPR
Contributes
  • ·An adaptive non-Gaussian statistical core placed ahead of any ML stage (model parsimony in practice), validated on the open Utah FORGE geothermal dataset.
  • ·Independent industrial origin (2024–2026) that structurally motivated the TRACE formalisation.
  • ·Strongest validation of Model Parsimony — L2a-dominant across structured-signal sub-domains.
  • ·Sub-domain × layer matrix showing the dominant layer shifting by evidence type.
  • ·Where CPR gets its first empirical evaluation on a real platform (Paper 2).
Open / to reconcile
  • ·Domain-specific methodology and operational statistics are reported separately (Paper 2, Shcherban-led) — not all public yet.
Foundational instance. Patent pending — UA u 2025 04038; U.S. Copyright Office deposit (Mar 2026). Paper 2 planned Q3 2026.

Judicial decision support

Instance · Judicial decision support
Supreme Court of Ukraine · Legal Positions Database (Instance C)
Zabolotnii, with the Supreme Court of Ukraine, funded by Expertise France

A v2 research roadmap: seven applied directions plus a cross-cutting EU-integration dimension. TRACE's value concentrates in L3, reframed as a machine-checkable normative act of the Court rather than a technical config. Direction 4.7 makes L4 oversight empirically measurable (telemetry + injected-error red-teaming + NASA-TLX).

layers
L1L2aL2bL3L4
principles
TREC
metrics
ETC · OvR · EP · CRP · CFI · UTC · RCI · IPSR · RBI · SNR · ABC · CE · OSI · FPA
Contributes
  • ·L3 as a machine-checkable normative act — the strongest L3 formalisation across all instances.
  • ·Direction 4.7 operationalises L4 metrics (OvR, RBI, SNR): meaningful-oversight measurement against rubber-stamping.
  • ·Explicit out-of-scope catalogue: predictive justice, automated procedural decisions, training-data contamination.
Open / to reconcile
  • ·Seventeen metrics under-used in the draft — addressed via the metrics-mapping companion.
  • ·Staged autonomy (ABC) absent; natural fit: pilot → category expansion with accumulated stability data.
Internal TRACE instance. Target venues per roadmap: JURIX 2027, Artificial Intelligence and Law, Law/Innovation/Technology.

Customs & government assistant

Instance · Customs & government
State Customs Service of Ukraine · tender architecture (4th domain)
Zabolotnii, 2026 (tender engineering package)

The most operationally concrete instance: named models (GliNER, XGBoost, e5-large, bge-reranker, GPT-5 Mini/Nano), a T1–T5 tier-ladder delivery, and bottom-up TCO. Fully aligned with current canon — it uses the L2a/L2b split, CPR, and the six-principle set as defined in trace.js, rather than extending them.

layers
L1L2aL2bL3L4
principles
TRACEMP
metrics
RCR · UTC · CRP · CE · EP · TCC · RBI · OvR · ETC · OSI · CPR
Contributes
  • ·Anti-pattern catalogue (§9): LLM-by-default, TAO-conflation, confidence-only escalation, cosmetic L4, learned L1, stateless L3, L2→L4 direct — TRACE's negative space, reusable in canonical docs.
  • ·Tier-ladder T1→T5: the clearest concrete operationalisation of Staged Autonomy / ABC (authority earned through accumulated stability data, incremental delivery).
  • ·Task-by-task L2a-vs-L2b selection table — a worked template for the right-sized-learned principle.
  • ·Honest CPR economics (Annex B v2): at GPT-5 prices CPR is a parsimony / overhead diagnostic, not a 'cheaper-than-baseline' slogan.
Open / to reconcile
  • ·Surfaces a 4th domain (customs / government) absent from the site's DOMAINS — clinical, upstream, legal only.
  • ·Annex B v1's 'CPR ≈ 0.10 = 10× cheaper' framing was retracted in v2; keep the quality + compliance-avoidance framing.
Applied / commercial instance (tender bid). Not a peer-reviewed source.
Kindred idea
Independent external work that corroborates or complements TRACE.

Orchestrated AI — neuro-symbolic enterprise governance

Kindred idea
Trisotech · Knowledge Worker Co-pilot
Denis Gagné & Tom — webinar "Retake Enterprise AI in 2026"

Three architectures — prompt-centric → integration-led → orchestrated — where only model-driven orchestration 'scales towards governance'. A neuro-symbolic split puts business rules in versioned BPMN/CMMN/DMN (not prompts), keeps a mandatory human in the loop, and treats the model as replaceable (bring-your-own-AI). Independent industry corroboration of TRACE's central thesis.

layers
L1L2bL3L4
principles
TR
metrics
ETC · OvR
Contributes
  • ·BPMN/CMMN/DMN as an executable substrate for L1/L3 — TRACE specifies layer properties; this offers a concrete carrier.
  • ·'Iceberg / shadow-processes' narrative — a strong opener for talks.
  • ·'Case = strategic map, BPMN = tactical routes' metaphor for explaining L3 orchestration.
Where TRACE adds rigour
  • ·Governance is qualitative — no quantified metrics; TRACE's seventeen measure what they assert.
  • ·No staged autonomy (ABC) — authority is static, not earned from stability data.
  • ·Bounded context as a safety envelope is absent — the case file grows unbounded for months (a CFI risk, not a guardrail).
  • ·A single escalation role vs TRACE's differentiated final-authority roles per domain.
Cite as industry corroboration only — NOT a methodological primary source (commercial, sellable-trust framing).