https://www.asion.ai/assets/frameworks-D0D-Ml72.js

1var e=[{slug:`prism`,paperSlug:`prism-readiness-diagnostic`,name:`PRISM`,tagline:`AI readiness, scored across five dimensions before a single agent is built.`,fullTitle:`PRISM — AI Readiness Scoring`,category:`Assessment`,accent:`#1E5BFF`,summary:`PRISM is ASION's structured diagnostic for enterprise AI readiness. It scores an organization across five dimensions — Data, Infrastructure, Policy, Skills, and Use Cases — on a reproducible 0-100 scale, then returns the ranked gaps that must be closed before deployment is worth attempting.`,whenToUse:[`Before committing capital to an AI program, platform, or multi-year vendor contract.`,`When executive sponsors and operating teams disagree on whether the organization is "ready" and need a shared, evidence-backed answer.`,`After a stalled pilot, to diagnose which of the five dimensions actually failed — not just the symptom.`,`At annual strategy cycles to re-baseline readiness against last year's score and measure real progress.`],outputs:[`A reproducible PRISM score (0-100) per dimension with a composite readiness index.`,`A ranked remediation map — the specific gaps, owners, and sequencing to close them.`,`An evidence appendix citing the data, documents, and interviews behind each score.`,`A re-score protocol your team owns, so quarterly baselines are produced without ASION on the engagement.`],phases:[{name:`Scoping & data intake`,detail:`A short working session defines the functions in scope and the evidence ASION will collect. Data custodians are named; access is granted on a read-only, time-boxed basis.`},{name:`Five-dimension assessment`,detail:`Parallel tracks score Data (lineage, quality, consent), Infrastructure (compute, networking, identity), Policy (governance, risk appetite, regulatory posture), Skills (operating literacy, not headcount), and Use Cases (business value, feasibility, dependencies).`},{name:`Evidence review & calibration`,detail:`Every score is traced to source evidence. Functions under assessment review the draft scores, challenge them, and the calibration is logged. No score is finalized without a named client owner signing off.`},{name:`Remediation map & handover`,detail:`ASION delivers the score, the ranked gaps, and the re-score protocol. Your team is certified to run the next PRISM cycle internally — the framework is yours from that point forward.`}],metrics:[{metric:`Data-dimension score`,definition:`Composite score for lineage, quality, consent, and retrieval readiness across the assessed data estate.`,source:`Data catalog audit + sampling review`,cadence:`At PRISM cycle; baseline then quarterly`},{metric:`Infrastructure-dimension score`,definition:`Compute, networking, identity, and observability readiness for AI workloads in the target environments.`,source:`Platform inventory + control testing`,cadence:`At PRISM cycle`},{metric:`Policy-dimension score`,definition:`Maturity of governance, risk appetite, disclosure posture, and regulatory mapping for AI use.`,source:`Policy inventory + stakeholder interviews`,cadence:`At PRISM cycle; re-scored on regulatory change`},{metric:`Skills-dimension score`,definition:`Operating literacy across named owners, reviewers, and escalation paths — not raw headcount.`,source:`Role-mapped capability interviews`,cadence:`At PRISM cycle; re-scored post-ELEVATE`},{metric:`Use-case-dimension score`,definition:`Portfolio-level feasibility and business-value readiness for the candidate AI use cases in scope.`,source:`PATHFINDER catalog + ROI models`,cadence:`At PRISM cycle; re-scored per portfolio change`}],faq:[{q:`What is PRISM?`,a:`PRISM is ASION's structured AI readiness assessment. It scores an organization across five dimensions — Data, Infrastructure, Policy, Skills, and Use Cases — on a reproducible 0-100 scale and returns a ranked remediation map. It is the first framework applied in most ASION consulting engagements because the results determine which subsequent frameworks are worth running.`},{q:`When does the client own PRISM?`,a:`The client owns the PRISM framework at the end of the engagement. The score, the evidence, the remediation map, and the re-score protocol are transferred with a written handover. ASION does not retain the data or the operating copy of the scoring workbook — your team runs future cycles without us.`},{q:`How does PRISM integrate with PATHFINDER?`,a:`The Use-Case dimension of PRISM and the PATHFINDER catalog share a single source of truth. Candidate use cases surfaced during PRISM feed directly into PATHFINDER's impact-versus-feasibility scoring, and PATHFINDER's ranked shortlist updates the Use-Case score on the next PRISM cycle.`},{q:`How long does a PRISM assessment take?`,a:`A standard PRISM cycle runs four to six weeks end to end for a single business unit, and six to nine weeks across a multi-function estate. Calendar time is driven by evidence access and stakeholder interview availability, not analysis throughput.`},{q:`Can we skip PRISM if we already have an AI strategy?`,a:`Sometimes — if you already have a recent, evidence-backed readiness baseline with named remediation owners and you can produce it. In practice, most "AI strategies" are roadmaps without readiness evidence, and skipping PRISM means subsequent frameworks get built on assumptions. When in doubt, run an accelerated PRISM first.`},{q:`What are PRISM's common failure modes?`,a:`The three we see most: (1) scoring is done without access to real evidence, so results are self-reports; (2) Skills is conflated with headcount instead of operating literacy; (3) the remediation map is published without named owners and dates, so nothing moves between baselines. All three are addressable in scoping.`}],relatedBlogSlugs:[`ai-readiness-assessment-enterprise`,`data-readiness-enterprise-ai`,`why-enterprise-ai-pilots-fail`]},{slug:`pathfinder`,paperSlug:`pathfinder-minimum-viable-agent`,name:`PATHFINDER`,tagline:`A 25-case catalog, scored on impact and feasibility, narrowed to the five with real ROI.`,fullTitle:`PATHFINDER — Use-Case Prioritization`,category:`Planning`,accent:`#4FA3FF`,summary:`PATHFINDER converts a messy backlog of "AI ideas" into a ranked, defensible shortlist. It starts with a 25-case catalog scored on impact and feasibility, filters against data, governance, and operating constraints, and returns the top five with fully modeled ROI — not demos, but business cases ready for capital allocation.`,whenToUse:[`When the AI backlog has grown past what any single sponsor can evaluate and decisions are stalling.`,`Before committing a platform, team, or budget cycle to an agent build program.`,`When functional leaders each have a favored use case and no shared basis for c

1omparison exists.`,`After PRISM, to translate readiness into the specific sequence of use cases the organization should pursue.`],outputs:[`A 25-case catalog with impact and feasibility scores and the evidence behind each.`,`A ranked top-five shortlist with ROI models including cost, uplift, risk, and payback window.`,`Kill-reason memos for every case that did not advance — so the answer to "why not this one?" is written, not argued.`,`A scoring rubric your team owns for future use-case intake, so new ideas are triaged against the same bar.`],phases:[{name:`Workflow scan`,detail:`ASION and the client jointly surface ten to twenty-five candidate workflows from interviews, backlogs, and portfolio reviews. No case is excluded at this stage — the inventory is the artifact.`},{name:`Impact × feasibility scoring`,detail:`Each case is scored on business impact (hours saved, revenue protected, risk reduced) and feasibility (data availability, governance posture, integration cost). Scores are calibrated against recent comparable engagements.`},{name:`Viability filters`,detail:`Cases are filtered against data, regulatory, and operating constraints. Anything that fails a hard constraint is removed with a dated kill-reason memo so the decision survives reorganizations.`},{name:`ROI modeling on the top five`,detail:`The five highest-ranked cases receive full ROI models — adoption curve, cost of ownership, escalation budget, and sensitivity analysis. Each is ready to present to capital allocation without further work.`}],metrics:[{metric:`Impact score (0-10)`,definition:`Annualized business value of the use case expressed as hours saved, revenue protected, or risk reduced.`,source:`Functional interviews + finance validation`,cadence:`At PATHFINDER cycle; refreshed annually`},{metric:`Feasibility score (0-10)`,definition:`Composite of data readiness, integration complexity, governance posture, and operating capacity.`,source:`PRISM dimension scores + engineering review`,cadence:`At PATHFINDER cycle`},{metric:`ROI payback window`,definition:`Months from deployment to cumulative break-even including ELEVATE and PULSE operating cost.`,source:`Modeled ROI workbook`,cadence:`Per shortlisted use case`},{metric:`Kill-reason coverage`,definition:`Share of unselected cases with a written dated memo citing the filter that removed them.`,source:`PATHFINDER catalog`,cadence:`At cycle close`},{metric:`Intake-to-decision latency`,definition:`Median days between a new case entering the catalog and a ranked decision being recorded.`,source:`Catalog change log`,cadence:`Continuous after handover`}],faq:[{q:`What is PATHFINDER?`,a:`PATHFINDER is ASION's use-case prioritization framework. It takes a 25-case catalog, scores each case on impact and feasibility, filters against data and governance constraints, and produces a ranked top-five shortlist with fully modeled ROI. It replaces the debate about "which use case first" with an auditable written answer.`},{q:`When does the client own PATHFINDER?`,a:`The client owns PATHFINDER at the end of the cycle. The catalog, the scoring rubric, the ROI models, and the kill-reason memos are transferred. New use-case intake is run by the client's own team against the same rubric, with ASION as an optional reviewer.`},{q:`How does PATHFINDER integrate with PRISM?`,a:`PATHFINDER consumes PRISM's Data, Infrastructure, and Policy dimension scores as direct feasibility inputs — a use case cannot score well on feasibility if the underlying dimensions have not been remediated. Conversely, PATHFINDER's ranked shortlist updates the Use-Case dimension on the next PRISM cycle.`},{q:`How long does PATHFINDER take?`,a:`A PATHFINDER cycle typically runs three to five weeks. The workflow scan and scoring each take about a week; viability filtering is a working session; ROI modeling on the top five takes one to two weeks with finance partnership.`},{q:`Can we skip PATHFINDER if we already know our top use case?`,a:`Rarely advisable. Most "we already know" conclusions originate from a single sponsor's preference and have not been tested against impact-versus-feasibility scoring. Even if the conclusion holds, the artifacts — the scoring rubric, the kill-reason memos — are what survive the sponsor's next rotation.`},{q:`What are PATHFINDER's common failure modes?`,a:`Three recurring failures: (1) impact is scored without finance validation, so the numbers collapse under scrutiny; (2) feasibility is scored before PRISM, so data and governance constraints are guessed; (3) the top five is published without kill-reason memos for the rest, so the prior year's conclusions are re-litigated every planning cycle.`}],relatedBlogSlugs:[`enterprise-ai-roi-measurement`,`ai-cost-optimization-enterprise`,`why-enterprise-ai-pilots-fail`]},{slug:`beacon`,paperSlug:`beacon-stakeholder-alignment`,name:`BEACON`,tagline:`Stakeholder alignment and vendor vetting before a single line of agent code is written.`,fullTitle:`BEACON — Stakeholder Alignment & Vendor Vetting`,category:`Alignment`,accent:`#3B7BE5`,summary:`BEACON aligns executive sponsors, functional owners, and technical delivery on 

1scope, success criteria, and exit conditions before build starts — and applies the same structured review to vendor selection. The artifact is a signed engagement charter and a RACI the organization will actually use.`,whenToUse:[`Before agent development starts, to prevent the mid-build scope drift that stalls most deployments.`,`When a vendor shortlist needs to be narrowed against a rubric the organization can defend to audit.`,`When an engagement has multiple sponsors and "done" means different things to each of them.`,`After a scope change that materially moves success criteria — BEACON is re-run, not patched.`],outputs:[`A signed engagement charter covering outcomes, risk appetite, and exit conditions.`,`A RACI matrix naming humans by role for every lifecycle stage, with delegation rules.`,`A vendor-vetting scorecard with weighted criteria and a signed recommendation memo.`,`A written conflict-resolution protocol so scope disagreements escalate on a known path.`],phases:[{name:`Executive briefing`,detail:`A structured session with named executive sponsors captures outcome expectations, risk appetite, and the explicit boundary between acceptable and unacceptable failure modes.`},{name:`Functional RACI workshop`,detail:`Functional owners, reviewers, and escalation paths are mapped end to end. Ambiguous ownership is resolved 

1in session; where it cannot be, the decision is escalated with a named deadline.`},{name:`Vendor vetting (if applicable)`,detail:`Candidate vendors are scored against a weighted rubric covering model-agnosticism, governance maturity, exit terms, and deployment references. The recommendation is a signed memo, not a deck.`},{name:`Charter sign-off & conflict protocol`,detail:`The engagement charter is signed by every named sponsor and owner. A written conflict-resolution protocol defines how scope disagreements escalate so that mid-build drift has an owner.`}],metrics:[{metric:`Charter sign-off coverage`,definition:`Share of named sponsors and owners who have signed the engagement charter before build starts.`,source:`Engagement governance log`,cadence:`At charter close; re-verified on scope change`},{metric:`RACI completeness`,definition:`Share of lifecycle stages with a named Accountable and Responsible role assigned.`,source:`RACI matrix`,cadence:`At charter close`},{metric:`Vendor scorecard weighting`,definition:`Documented weights applied to each rubric dimension in vendor vetting.`,source:`Vendor-vetting workbook`,cadence:`Per vendor selection`},{metric:`Scope-change escalation latency`,definition:`Median hours between a scope change surfacing and a recorded decision on the conflict-resolution path.`,source:`Conflict-resolution log`,cadence:`Continuous during engagement`},{metric:`Exit-condition specificity`,definition:`Share of exit conditions expressed as measurable, dated criteria rather than subjective narrative.`,source:`Engagement charter`,cadence:`At charter close`}],faq:[{q:`What is BEACON?`,a:`BEACON is ASION's stakeholder-alignment and vendor-vetting framework. It produces a signed engagement charter, a working RACI, a vendor scorecard where applicable, and a written conflict-resolution protocol. Its purpose is to make scope and ownership explicit before build begins.`},{q:`When does the client own BEACON?`,a:`The client owns BEACON's artifacts immediately — the charter, the RACI, and the vendor rubric are client documents from the moment they are signed. The framework itself, including the workshop templates and rubric weights, transfers at engagement close so internal teams can run BEACON cycles without ASION.`},{q:`How does BEACON integrate with PATHFINDER?`,a:`BEACON is run on the top-ranked PATHFINDER use case before build begins. Its sponsors and owners are the same humans who signed off on the PATHFINDER ROI model — BEACON converts their agreement on the case into a dated, specific agreement on scope and exit.`},{q:`How long does BEACON take?`,a:`A BEACON cycle is typically two to four weeks. The executive briefing and RACI workshop each take one working session; vendor vetting adds one to two weeks depending on the shortlist size; charter sign-off is a calendar-driven final step.`},{q:`Can we skip BEACON if stakeholders already agree?`,a:`The question itself is usually the answer. "Stakeholders already agree" is almost always true in the abstract and false in specifics — scope, success criteria, and exit conditions. BEACON takes a few weeks and prevents the months of mid-build drift that "we already agree" invariably produces.`},{q:`What are BEACON's common failure modes?`,a:`Three failures recur: (1) the charter is signed but exit conditions are narrative rather than dated criteria; (2) RACI is completed for build but not for operations, so ownership evaporates at handover; (3) the conflict-resolution protocol is documented but never invoked, and scope disagreements leak into unstructured email.`}],relatedBlogSlugs:[`ai-vendor-selection-framework`,`ai-change-management-enterprise`,`ai-center-of-excellence`]},{slug:`elevate`,paperSlug:`elevate-progressive-autonomy`,name:`ELEVATE`,tagline:`Progressive autonomy and certified capability — your team runs what ASION builds.`,fullTitle:`ELEVATE — Progressive Autonomy & Capability Building`,category:`Capability`,accent:`#1542C7`,summary:`ELEVATE moves agents from assistive to supervised to autonomous operation with documented promotion criteria at each transition, while certifying the client practitioners who own them. Across ASION consulting engagements, staged autonomy reduces incident rate by 62 percent versus direct autonomous deployment.`,whenToUse:[`Before granting an agent production write access on a business-critical workflow.`,`When the deploying team has never operated an autonomous agent in production before.`,`When regulatory posture requires named human oversight with documented promotion evidence.`,`After an incident or demotion event, to reset the tier and re-establish promotion criteria.`],outputs:[`A per-agent tier ledger — assistive, supervised, or autonomous — with signed promotion evidence.`,`Dated promotion criteria for each transition, owned by a named client operator.`,`A certified-practitioner roster — the humans qualified to promote, demote, and operate each agent.`,`A demotion-and-revocation protocol so autonomy can be safely walked back without taking the agent offline.`],phases:[{name:`Tier-criteria definition`,detail:`Before the agent runs, the promotion criteria for assistive-to-supervised and supervised-to-autonomous are written, signed, and dated. The criteria are evidence-based — volume, error rate, escalation trend — not elapsed time.`}

1,{name:`Assistive tier`,detail:`The agent proposes; a human executes. Every proposal is logged with accept, modify, or reject. Assistive is held until the accept-rate trend clears the threshold defined in tier-criteria.`},{name:`Supervised tier`,detail:`The agent executes, but writes on production systems require human approval. This is where most operating discipline is defined. Compressing this tier is the single strongest predictor of autonomous-tier incidents.`},{name:`Autonomous tier + practitioner certification`,detail:`The agent executes within policy with audit-trail oversight. In parallel, named practitioners are certified against the operating criteria and take ownership. Demotion remains available as a normal operating event.`}],metrics:[{metric:`Current tier`,definition:`The active autonomy tier for the agent — assistive, supervised, or autonomous.`,source:`ELEVATE tier ledger`,cadence:`Continuous; recorded per transition`},{metric:`Time-in-tier`,definition:`Days the agent has operated at the current tier since the last promotion or demotion event.`,source:`Tier ledger`,cadence:`Continuous`},{metric:`Supervised-tier error rate`,definition:`Share of supervised-tier actions that required correction or rejection by the reviewing human.`,source:`Supervised-tier operations log`,cadence:`Weekly during supervised tier`},{metric:`Promotion-criteria conformance`,definition:`Share of documented promotion criteria met at the time of the last promotion review.`,source:`Signed promotion memo`,cadence:`Per promotion event`},{metric:`Certified-practitioner coverage`,definition:`Share of agents in production with at least two certified practitioners as named operators.`,source:`Practitioner roster`,cadence:`Continuous; re-verified quarterly`}],faq:[{q:`What is ELEVATE?`,a:`ELEVATE is ASION's progressive autonomy and capability-building framework. It moves agents through three tiers — assistive, supervised, autonomous — with signed promotion criteria at each transition, while certifying the client practitioners who operate them. Autonomy is earned on evidence, not granted on deadline.`},{q:`When does the client own ELEVATE?`,a:`Ownership transfers progressively. Tier decisions are client-owned from day one — ASION provides the readiness evidence but the named client operator signs the promotion. By the time the agent reaches autonomous, the certified-practitioner roster is already in place and ASION's role is advisory, not operational.`},{q:`How does ELEVATE integrate with PROVE?`,a:`PROVE's five gates define what quality looks like at a single decision point; ELEVATE defines how much autonomy the agent has earned to act on its own PROVE-approved decisions. An agent can pass PROVE in assistive tier with zero autonomy; it must pass PROVE continuously to stay in autonomous tier.`},{q:`How long does ELEVATE take?`,a:`In deployed engagements, assistive-to-supervised has run four to eight weeks depending on workflow complexity. Supervised-to-autonomous typically takes longer because the criteria are stricter — often eight to sixteen weeks. There is no time-based default; the criteria are the gate.`},{q:`Can we skip ELEVATE and deploy directly to autonomous?`,a:`Technically yes; operationally expensive. Direct autonomous deployment produces a 62 percent higher incident rate in our engagement data and removes the supervised tier where operating discipline is built. Organizations that skip ELEVATE almost always reconstruct it after their first production incident.`},{q:`What are ELEVATE's common failure modes?`,a:`The recurring pattern: (1) supervised tier is compressed to hit a launch date, so operating discipline is never defined; (2) promotion criteria are softened under pressure instead of the launch being delayed; (3) demotion is treated as a failure rather than a normal operating event, so teams delay it until the incident is already material.`}],relatedBlogSlugs:[`progressive-ai-autonomy-elevate`,`agentic-ai-enterprise-guide`,`ai-change-management-enterprise`]},{slug:`verify`,paperSlug:`verification-gap`,name:`VERIFY`,tagline:`Real-time hallucination detection with multi-model cross-reference and grounding scores.`,fullTitle:`VERIFY — Hallucination Detection`,category:`Verification`,accent:`#0B1222`,summary:`VERIFY cross-references every AI output against source data and secondary models, attaches a per-claim grounding score, and blocks or escalates unsupported claims before they reach the system boundary. In deployed engagements it achieves above 95 percent detection accuracy on unsupported claims in high-stakes domains.`,whenToUse:[`On any workflow where unsupported claims reaching an external audience carry regulatory, reputational, or safety risk.`,`Before an agent is allowed to produce customer-facing content, regulatory filings, or clinical summaries.`,`When existing RAG pipelines produce outputs that look cited but are not actually grounded in the retrieved sources.`,`As a continuous gate during ELEVATE's supervised tier to generate the evidence promotion criteria require.`],outputs:[`Per-claim grounding scores attached to every material output.`,`Inline citations traceable to the exact source passage, not just the source document.`,`A verification audit trail showing which claims passed, which were escalated, and which were blocked.`,`Threshold policies your team owns — accept, escalate, or block — tuned per workflow and per regulatory posture.`],phases:[{name:`Source-corpus registration`,detail:`The data sources VERIFY will treat as authoritative are registered with access controls, freshness windows, and an explicit boundary between "authoritative" and "reference". What is not registered cannot ground a claim.`},{name:`Cross-reference configuration`,detail:`Secondary models — typically from a different provider family — are configured for consensus scoring. The secondary check is not a second opinion on style; it is an independent grounding check on factual claims.`}

1,{name:`Threshold policy tuning`,detail:`Accept, escalate, and block thresholds are tuned per workflow against a labeled evaluation set. Thresholds are documented and change-controlled; they are not moved silently to improve throughput.`},{name:`Continuous operation & drift review`,detail:`VERIFY runs on every inference in scope. Detection accuracy and false-escalate rate are reviewed on a scheduled cadence; thresholds are re-tuned when source corpora or workflows materially change.`}],metrics:[{metric:`Detection accuracy`,definition:`Share of unsupported claims correctly flagged by VERIFY against a labeled evaluation set.`,source:`Labeled evaluation harness`,cadence:`Per deployment; re-run quarterly`},{metric:`Grounding score`,definition:`Per-claim confidence that the claim is supported by the registered authoritative sources.`,source:`VERIFY inference log`,cadence:`Continuous; per inference`},{metric:`Escalation rate`,definition:`Share of inferences escalated to a human reviewer for grounding adjudication.`,source:`VERIFY inference log`,cadence:`Continuous; reviewed weekly`},{metric:`Block rate`,definition:`Share of inferences blocked at the boundary because grounding fell below the block threshold.`,source:`VERIFY inference log`,cadence:`Continuous; reviewed weekly`},{metric:`False-escalate rate`,definition:`Share of escalations where the human reviewer confirmed the original output was adequately grounded.`,source:`Reviewer adjudication log`,cadence:`Continuous; reviewed monthly`}],faq:[{q:`What is VERIFY?`,a:`VERIFY is ASION's hallucination-detection framework. It cross-references every AI output against registered authoritative sources and secondary models, attaches a per-claim grounding score, and accepts, escalates, or blocks each output against documented thresholds. It is deployed as a gate at the system boundary, not a post-hoc review.`},{q:`When does the client own VERIFY?`,a:`The client owns VERIFY's configuration and threshold policies from day one — ASION tunes them collaboratively, but every threshold change is signed by a named client owner. The framework, the tuning harness, and the evaluation set transfer at engagement close.`},{q:`How does VERIFY integrate with PROVE?`,a:`VERIFY is the Validation gate within PROVE's five-gate architecture. A decision cannot pass PROVE if it fails VERIFY's grounding threshold. In practice, VERIFY runs upstream and PROVE records the outcome; the two frameworks share an audit trail so every production decision has a continuous evidentiary chain.`},{q:`How long does VERIFY deployment take?`,a:`Source-corpus registration and initial threshold tuning typically run three to six weeks. Detection accuracy reaches deployment readiness — above 95 percent in high-stakes domains — after the first full evaluation cycle, which depends on the size of the labeled set.`},{q:`Can we skip VERIFY if we already have RAG with citations?`,a:`RAG with citations is not the same as grounded with evidence. Many RAG systems attach a citation to an output that the cited passage does not actually support. VERIFY's grounding score is a different artifact — it measures whether the claim is supported, not whether a citation was produced.`},{q:`What are VERIFY's common failure modes?`,a:`Three recurring failures: (1) the source corpus is under-registered, so legitimate claims fail to ground and escalation rate explodes; (2) thresholds are moved under throughput pressure without sign-off, silently eroding detection; (3) the secondary model is from the same provider family, so consensus scoring degenerates into self-agreement.`}],relatedBlogSlugs:[`ai-hallucination-detection-production`,`enterprise-rag-architecture`,`ai-safety-production-enterprise`]},{slug:`prove`,paperSlug:`verification-gap`,name:`PROVE`,tagline:`Five gates — Performance, Reliability, Oversight, Validation, Ethics — on every production decision.`,fullTitle:`PROVE — Production Gates`,category:`Governance`,accent:`#0B2A6E`,summary:`PROVE is a five-gate policy engine applied at the system boundary. Every material decision passes through Performance, Reliability, Oversight, Validation, and Ethics gates and is accepted, escalated, or blocked with a signed audit trail. It is the governance surface an external auditor can actually read.`,whenToUse:[`On any production AI decision whose failure would be material to customers, regulators, or the balance sheet.`,`Before a deployment is considered audit-ready for SOC 2, ISO 42001, HIPAA, or the EU AI Act.`,`When multiple agents and pipelines share a system boundary and a single governance surface is needed.`,`When existing governance is policy-on-paper and lacks an auditable per-decision trail.`],outputs:[`A per-decision audit trail with the outcome of each of the five gates and a signed disposition.`,`Five gate policies — Performance, Reliability, Oversight, Validation, Ethics — tuned per workflow.`,`An escalation-routing map so blocked or escalated decisions reach the right human within the right window.`,`A quarterly attestation pack summarizing gate performance for the governance committee and external auditors.`],phases:[{name:`Gate-policy definition`,detail:`Each of the five gates is defined as a policy with explicit thresholds, data dependencies, and owners. The definitions are change-controlled; silent tuning is not permitted.`},{name:`Threshold tuning per workflow`,detail:`Thresholds are calibrated per workflow against historical data and risk appetite. A medical-summarization workflow and a marketing-copy workflow do not share thresholds, and the difference is documented.`}

1,{name:`Escalation routing`,detail:`Every gate has a routing rule for escalate and block outcomes — named role, time window, and fallback. Escalations that miss the window trigger a secondary route so decisions never stall silently.`},{name:`Continuous operation & attestation`,detail:`PROVE runs on every inference in scope. A quarterly attestation pack is generated automatically from the audit trail and presented to the governance committee with a signed cover memo.`}],metrics:[{metric:`Gate-pass rate`,definition:`Share of decisions that passed all five gates without escalation or block.`,source:`PROVE audit trail`,cadence:`Continuous; reviewed weekly`},{metric:`Per-gate failure distribution`,definition:`Share of failures attributable to each of Performance, Reliability, Oversight, Validation, Ethics.`,source:`PROVE audit trail`,cadence:`Continuous; reviewed monthly`},{metric:`Escalation response latency`,definition:`Median minutes from escalation to a signed human disposition.`,source:`Escalation routing log`,cadence:`Continuous; reviewed weekly`},{metric:`Threshold-change cadence`,definition:`Count of change-controlled threshold updates and the signer of each.`,source:`PROVE change log`,cadence:`Continuous; attested quarterly`},{metric:`Attestation coverage`,definition:`Share of in-scope workflows covered by the latest quarterly attestation pack.`,source:`Attestation workbook`,cadence:`Quarterly`}],faq:[{q:`What is PROVE?`,a:`PROVE is ASION's five-gate production governance framework. Every material AI decision is evaluated against Performance, Reliability, Oversight, Validation, and Ethics gates and accepted, escalated, or blocked with a signed audit trail. It is the governance surface an external auditor can read end to end.`},{q:`When does the client own PROVE?`,a:`The client owns PROVE's gate policies and thresholds from the moment they are signed — ASION tunes collaboratively but the signer is always a named client owner. The framework itself, including the audit-trail schema and attestation templates, transfers at engagement close.`},{q:`How does PROVE integrate with VERIFY?`,a:`VERIFY is operated as the Validation gate within PROVE. VERIFY's grounding score is one of the five gate signals; the other four — Performance, Reliability, Oversight, Ethics — are independent policies. The combined audit trail is what external auditors read.`},{q:`How long does PROVE take to stand up?`,a:`Gate-policy definition and per-workflow threshold tuning typically run six to ten weeks for a single workflow and longer across a portfolio. The first attestation pack is generated at the end of the first full quarter of operation.`},{q:`Can we skip PROVE if we already have an AI policy?`,a:`Policy-on-paper is not a governance surface. An AI policy that cannot produce a per-decision audit trail under auditor request will not survive SOC 2, ISO 42001, or EU AI Act scrutiny. PROVE's deliverable is the audit trail, not the policy document.`},{q:`What are PROVE's common failure modes?`,a:`Three recurring failures: (1) thresholds are tuned silently to improve throughput, eroding the audit trail; (2) escalations have no fallback route and stall when the primary owner is unavailable; (3) the attestation pack is written as a narrative rather than generated from the audit trail, so the evidentiary chain breaks at the committee boundary.`}],relatedBlogSlugs:[`ai-governance-framework-enterprise`,`enterprise-ai-compliance-soc2-hipaa-gdpr`,`ai-safety-production-enterprise`]},{slug:`pulse`,paperSlug:`pulse-continuous-assurance`,name:`PULSE`,tagline:`Continuous telemetry — drift, escalation rate, policy violations — with sub-hour detection.`,fullTitle:`PULSE — Continuous Production Telemetry`,category:`Operations`,accent:`#7B61FF`,summary:`PULSE is the always-on telemetry and drift-detection layer for agents in production. It watches behavioral stability, escalation-rate trends, and policy-violation events; detects material degradation in under an hour; and auto-generates remediation tickets to the owning function.`,whenToUse:[`On every agent operating at ELEVATE's supervised or autonomous tier in production.`,`When drift would materially change business outcome, regulatory exposure, or customer experience.`,`When regulatory posture — EU AI Act post-deployment monitoring, SOC 2 CC7 — requires continuous evidence.`,`After a PROVE gate-failure pattern emerges, to attach continuous monitoring to the underlying signal.`],outputs:[`Live drift, escalation-rate, and policy-violation telemetry per agent.`,`Auto-generated remediation tickets routed to the owning function with suggested disposition.`,`Demotion triggers wired into ELEVATE so unhealthy autonomous-tier agents step back without manual action.`,`A quarterly assurance dossier summarizing telemetry for the governance committee and external auditors.`],phases:[{name:`Baseline establishment`,detail

1:`Behavioral baselines are established during ELEVATE's supervised tier — volume, output distributions, escalation mix. The baselines are per-workflow and per-agent; they are not imported from a reference deployment.`},{name:`Drift, anomaly, and policy signals`,detail:`PULSE monitors three signal families continuously: behavioral drift against baseline, anomalous usage or output patterns, and live policy-violation events surfaced from PROVE.`},{name:`Remediation routing`,detail:`Triggered signals generate remediation tickets routed to the owning function with suggested disposition — demote, re-tune, investigate. Unacknowledged tickets escalate on the BEACON conflict-resolution path.`},{name:`Assurance dossier`,detail:`A quarterly dossier is generated from telemetry — not from narrative — and presented to the governance committee. The dossier feeds the PROVE attestation pack for the same period.`}],metrics:[{metric:`Detection latency`,definition:`Median time from a material behavioral change to a PULSE alert being raised.`,source:`PULSE event log`,cadence:`Continuous; reviewed weekly`},{metric:`Drift index`,definition:`Composite score of behavioral drift against the registered baseline for each agent.`,source:`PULSE telemetry stream`,cadence:`Continuous`},{metric:`Escalation-rate trend`,definition:`Rolling escalation rate per workflow compared to the baseline escalation rate.`,source:`ELEVATE + PROVE logs`,cadence:`Continuous; reviewed weekly`},{metric:`Policy-violation event count`,definition:`Count of PROVE-surfaced policy violations per agent over a rolling window.`,source:`PROVE audit trail`,cadence:`Continuous`},{metric:`Ticket-to-disposition latency`,definition:`Median hours from a PULSE-generated ticket to a signed disposition by the owning function.`,source:`Remediation ticket log`,cadence:`Continuous; reviewed weekly`}],faq:[{q:`What is PULSE?`,a:`PULSE is ASION's continuous production telemetry framework. It monitors drift, escalation-rate trends, and policy-violation events on every agent in production; detects material degradation in under an hour; and auto-generates remediation tickets routed to the owning function. Demotion triggers are wired directly into ELEVATE.`},{q:`When does the client own PULSE?`,a:`The client owns PULSE telemetry and remediation routing from day one — the data and the tickets are client assets. The framework itself, including baseline definitions and signal configurations, transfers at engagement close so internal operations teams run PULSE without ASION.`},{q:`How does PULSE integrate with ELEVATE?`,a:`PULSE is the sensor layer that ELEVATE's demotion triggers attach to. When PULSE detects drift, sustained escalation-rate spikes, or repeated policy violations, the configured demotion trigger steps the agent back to a lower autonomy tier automatically. Demotion is logged and investigated but does not require manual intervention.`},{q:`How long does PULSE take to stand up?`,a:`Baseline establishment requires ELEVATE's supervised tier to have run long enough to produce a stable baseline — typically four to eight weeks. Signal configuration and routing setup add two to three weeks in parallel. The first assurance dossier covers the first full quarter of operation.`},{q:`Can we skip PULSE if we have generic application monitoring?`,a:`Generic APM catches infrastructure and latency failures; it does not catch behavioral drift, escalation-mix shifts, or policy-violation patterns that are specific to AI workloads. PULSE complements APM — it does not replace it — and both are typically needed for a defensible post-deployment posture.`},{q:`What are PULSE's common failure modes?`,a:`Three recurring failures: (1) baselines are imported from a reference deployment rather than built from the live agent, so drift detection is miscalibrated; (2) remediation tickets have no owning function, so they age unacknowledged; (3) demotion triggers are configured but never tested, so the first live trigger is also the first rehearsal.`}],relatedBlogSlugs:[`llmops-enterprise-guide`,`ai-safety-production-enterprise`,`ai-governance-framework-enterprise`]}];function t(t){return e.find(e=>e.slug===t)}export{t as n,e as t};