North star
Help defense teams produce mitigation that is more complete, better corroborated, earlier, and cheaper to assemble: the input that persuades the people who actually set sentences (prosecutors, judges, parole boards, governors). Aperio never claims to "reduce sentences" itself: the defensible, evidence-grounded promise is earlier and more complete corroborated mitigation per scarce specialist-hour, with provenance on every assertion and the supervising attorney accountable for all work product. The durable system of record is a verifiable, auditable casefile of source-linked facts, not the chat stream.
Where Aperio fits in the defense lifecycle
Mitigation is not a single step: it threads the whole case, from the first appearance to clemency. The highest-leverage insertion points are early (capturing social facts at intake) and at plea (where ~95% of cases actually resolve), not just the sentencing hearing.
Arrest, intake & first appearance
Bottleneck: Social facts that matter at sentencing (housing, employment, dependents, treatment, disability) are never captured at the moment of first contact; nothing structured exists yet, and the highest-leverage early-mitigation window is missed under caseload pressure.
Aperio: Stand up a casefile at intake and capture a structured first-contact mitigation snapshot (stable housing, dependents, treatment history, employment) so early social facts become provenance-bearing leads rather than lost notes.
Discovery review
Bottleneck: Police reports, bodycam, jail calls, 911, interrogations, and lab results are a major review bottleneck; AV evidence is voluminous and time-expensive to comb for mitigation-relevant statements.
Aperio: Ingest timestamped transcripts from defense-native AV tools (JusticeText/Reduct), importing with provenance preserved instead of rebuilding transcription, so client/family/jail-call statements flow into the casefile as cite-able leads pinned to the exact source timestamp.
Investigation & records collection
Bottleneck: The 'scorched-earth' records pull (prenatal, school/IEP (individualized education program), medical, mental-health, substance-use, juvenile/child-welfare, employment, military, prior legal/CPS) is the biggest calendar and dollar sink; service-based retrieval commonly runs weeks per custodian (a practitioner estimate, not a cited SLA) with state-varying per-page fees, HIPAA/authorization friction, and uncooperative custodians.
Aperio: Auto-generate jurisdiction-correct authorization/subpoena request packets, track each pending request against the record category it closes, and surface a per-custodian status tracker. Aperio can chase exogenous delay but cannot eliminate it; set that expectation.
Social-history development
Bottleneck: Compiling a documented multigenerational bio-psychosocial history, life chronology, and collateral interviews is dozens-to-hundreds of hours of skilled labor: the work most often triaged away under caseloads that run well above the new RAND standards.
Aperio: Auto-construct the life chronology and multigenerational genealogy as first-class structured casefile objects (CaseFleet pattern), every fact linked to its source page, while augmenting, never replacing, the specialist's clinical rapport and sensitive-disclosure elicitation, which cannot be automated.
Expert referral
Bottleneck: Indicators (prenatal alcohol exposure, TBI/head injury, special-ed/low IQ, documented trauma) are missed or surfaced late; FASD (fetal alcohol spectrum disorder) is 17–36% prevalent in justice settings and chronically underdiagnosed; the expert depends entirely on the social-history packet the specialist feeds them.
Aperio: Flag clinical indicators in the casefile and prompt a neuropsych/forensic-psych/FASD referral, auto-assembling the social-history packet the expert needs, while deferring clinical framing to attorney/expert judgment, since neurodevelopmental findings can be used against the client as 'future dangerousness.'
Plea negotiation
Bottleneck: ~95% of state and ~98% of federal convictions are by plea, so most sentencing outcomes are negotiated, not litigated, yet mitigation is usually assembled (if at all) only for the sentencing hearing, after the offer is set.
Aperio: Produce a one-page corroborated 'early mitigation / mitigation-forward plea packet' to hand prosecutors pre-offer: the highest-frequency, highest-volume point where a smarter plan, earlier, actually changes charge exposure.
Presentence report (PSR)
Bottleneck: Probation independently writes a thin social-history section; under Fed. R. Crim. P. 32(f) the defense has only 14 days to file written objections, a tight window against an adverse narrative.
Aperio: PSR-objection assistant: diff the probation officer's social-history section against Aperio's corroborated casefile and draft Rule 32 objections inside the 14-day window, flagging omitted mitigation and factual errors with source cites.
Sentencing memo & hearing
Bottleneck: Memos recite guidelines instead of telling the client's story; the §3553(a) variance argument and any $5k–$25k mitigation video are expensive, late, and disconnected from the underlying record.
Aperio: Structured AI-assisted drafting emits three artifacts from one casefile: (1) the social-history report, (2) the §3553(a) sentencing memo with inline pin-cite provenance, and (3) a mitigation-video brief (themes, beats, interviewee shot-list) to hand established producers, every draft gated by a mandatory cite-check pass and attorney sign-off.
Post-conviction, parole & clemency
Bottleneck: The same social-history record is needed for years (commutation, parole, clemency) but is rebuilt from scratch each time, and a thin original investigation record is a leading ineffective-assistance-of-counsel (IAC) reversal ground in capital/appellate work.
Aperio: Reuse the provenance-stamped casefile to assemble parole/clemency packets (personal statement, support letters, programming documentation); the coverage record doubles as a Wiggins/Rompilla IAC-proofing audit trail (strongest in capital/appellate matters).
Agentic-UX principles
In high-stakes, no-fabrication defense work, retrieval grounding does not eliminate hallucination. The interface, not just the model, is what makes the tool safe and defensible. These principles drive Aperio's design.
Provenance-or-quarantine: an uncited assertion is a blocking state, not a warning.
Retrieval grounding does NOT eliminate hallucination: the Stanford/Magesh study measured 17–33% hallucination even in RAG-based legal tools (Lexis+ AI ~17%, Westlaw AI-Assisted ~33%), and a citation can point to a real record that does not support the proposition. Everything is high-stakes, no-fabrication defense work product.
Fact lifecycle as a first-class status: Lead → Finding → Corroborated → Disputed.
Investigations live on uncorroborated tips; the practitioner distinction between 'someone said X' and 'the school record shows X' is exactly what makes mitigation credible and the agent safe. Corroboration is also the mechanism the holistic-defense evidence rewards.
Coverage/gap meter as the home screen: Aperio's framework, presented honestly.
This operationalizes a strategy duty into a visible checklist and is the product's most differentiated screen. Note: there is no ABA-codified enumeration of exactly ten domains; this is Aperio's synthesis of the bio-psychosocial mitigation literature, and reaching 100% coverage is NOT a constitutional-sufficiency guarantee.
Propose/commit human-in-the-loop with a strict friction budget.
The product's whole reason to exist is LESS time per case against a documented caseload crisis; piling on blocking gates makes overworked reviewers rubber-stamp, nullifying both the safety value and the time-savings thesis.
Reasoning feed is a transient assistant; the casefile is the system of record.
Streamed 'thinking' is a processed summary, not a faithful transcript of model computation; marketing it as an audit of 'how the AI thinks' would collapse the credibility of the genuinely defensible artifacts. An exhausted user must read outcomes, not a firehose.
Cite-check export gate produces a defensibility artifact.
2025–26 saw a wave of fabricated-citation sanctions (post Mata v. Avianca); the defensible record is the source-linked casefile plus a verification artifact, not the chat. Verification must confirm the source SUPPORTS the proposition, not merely that a link exists.
Privilege and security as visible affordances backed by real architecture.
ABA Formal Opinion 512 makes confidentiality, supervision, and vendor data-diligence enforceable duties; training on client input or cross-tenant sharing could destroy privilege. A banner is necessary-but-insufficient theater unless the data architecture is genuinely no-train and segregated, a gating adoption requirement for government PD buyers, not a feature add-on.
Redaction is verified, never fire-and-forget.
Real e-discovery tools (e.g., Everlaw) have documented silent-failure modes that dropped redactions on production; CSA (child sexual abuse)/victim/witness exposure carries sanctions and ethical exposure.
Trauma-informed and accessible by default.
Mitigation evidence is inherently traumatic (abuse, CSA, violence, autopsy) and teams face vicarious trauma at high caseloads. (Design adaptation of trauma-informed principles: present as Aperio's adaptation, not verbatim 'the six SAMHSA principles,' which also include peer support.)
Casefile-as-shared-state, organized by mitigation domain and life timeline.
Chat is ephemeral and unverifiable; the leap past streaming chat is one real-time, multi-user object of structured, provenance-bearing facts the agent and team co-edit, where nothing becomes a Finding without a human-committed source link.
Sell rigor and time-savings in the UI; never sell outcomes.
There is no evidence any AI tool causes lower sentences; outcome language invites false-advertising and ethics exposure and misleads users. Even a competitor (SentencingStats) markets 'data-backed arguments for reduced sentences'; Aperio should be more honest than the market.
Integration architecture
Prioritized by leverage and feasibility. Ship file-first ingestion and the defensible casefile before chasing deep integrations; meet private firms in Clio and import AV evidence from defense-native tools rather than rebuilding them.
Most defense evidence is loose PDFs, and underfunded PD offices won't have (or can't quickly procure) a practice-management integration. This is the single most realistic beachhead and decouples adoption from any PM partnership.
How: Ship secure direct upload plus Drive/SharePoint connectors so teams can drop discovery PDFs and records immediately; parse into the casefile with provenance (file, page) preserved on every extracted fact.
Clio is the center of gravity for private criminal-defense firms (~150k+ professionals, 250+ app directory, free App Partner Program with co-marketing, the most mature REST API V4 + OAuth2 + webhooks): the single highest-ROI integration for the private-firm segment. Strategic caveat: the mission-aligned north-star buyer (public defenders) is NOT on Clio, so this is the easiest engineering beachhead, not the mitigation-volume center.
How: Build a public Clio Manage app: OAuth 2.0 auth-code flow, webhook-driven sync of Matters/Contacts/Custom Fields (NOT polling), map the ~10 mitigation domains to custom fields + notes, push the finished memo back as a Document via the 3-step upload. Architect explicitly around the hard 3-requests/sec PER-APP limit shared across all customers (batching, caching, exponential backoff, per-region US/CA/EU/AU hosts); Manage and Grow are separate APIs; target Manage.
JusticeText is the de-facto AV-evidence layer for PDs (800+ defenders, 50+ agencies, ~50% review-time savings, a self-reported throughput figure, not an outcome). Timestamped transcripts are ideal provenance for client/family/jail-call narratives; don't rebuild transcription.
How: Treat as manual export/import TODAY (Word transcripts, MP4 clips) with provenance preserved on import; JusticeText/Reduct have no documented public third-party API. Validate automated export/API availability via direct business development (BD) before promising a live integration on the roadmap.
Where mitigation volume and mission-aligned buyers concentrate (defenderData: 250+ PD offices, all 50 states, 600+ appointed private firms; DEFENDERbyKarpel is the defender sibling of PROSECUTORbyKarpel). This is the north-star channel, but it sits behind government procurement.
How: Open BD; expect partnership-gated, possibly SFTP/flat-file or vendor-mediated integration, not open REST (API openness unverified). Pilot with one PD office to prove privilege-safe handling and pass security review. CRITICAL data-separation guarantee: Karpel builds both prosecutor and defender products (and is now PE-owned), so defense work product must never cross into prosecution-side tenancy.
Completes private-practice PM coverage at low marginal cost once the Clio sync abstraction exists.
How: Reuse the Clio sync/document-push abstraction. Note neither API is self-serve 'public': MyCase's API is gated behind its paid Advanced tier (~$89/mo), and Smokeball's REST/JSON API is partner-enrollment-gated (OAuth2 client credentials issued only to approved partners). Budget partner enrollment, not a drop-in.
Appellate and capital-defense organizations standardize on these (NetDocuments: 150+ integrations, modern API, M365/DocuSign; iManage: SeeUnity/Anaqua connector ecosystem). Well-documented connector targets.
How: Reuse the document-sync layer built for Clio; both are realistic, well-documented connector surfaces.
The single biggest manual cost in mitigation and a defensible differentiator, and the records Wiggins/Rompilla turned on. But honest scoping: this is a hard, capital-intensive SERVICE play. EvenUp's PI 'Pre-Litigation-as-a-Service' model does NOT transfer cleanly: PI contingency-fee, high-margin economics don't map to grant/government-funded indigent defense, and criminal-mitigation records (juvenile/sealed, child-welfare confidentiality, BOP (Federal Bureau of Prisons), military via NPRC (National Personnel Records Center)) are legally harder than PI medical records.
How: Start software-only: a request tracker plus auto-generated jurisdiction-correct authorization/subpoena packets tied to the record category each request closes (per-page fees verified state-by-state, e.g., CA $0.25/pg, MD $0.76/pg; turnaround weeks-per-custodian is a practitioner estimate, not a cited SLA). Layer fulfillment partnerships only after validating coverage for criminal-specific custodians: per-vendor BD + engineering lift, not a universal API.
Grounds the achievable-outcome framing with data. Two hard limits: it is FEDERAL-only (most PD volume is state court), and it is comparative/disparity advocacy, conceptually distinct from bio-psychosocial mitigation.
How: Present strictly as comparative/disparity evidence to support a §3553(a) argument, NEVER a predicted or guaranteed reduction. Keep separate from the mitigation framework so the two don't blur.
Mitigation is fact/biography, not case law, though strong sentencing memos DO cite §3553(a) and departure/variance authority. Developer portals exist (TR Apr-2024; Lexis Mar-2022) but content is license-encumbered and you generally can't freely store/redistribute results inside work product.
How: Treat as link-out plus optional gated citation-checking; defer or skip unless a specific enterprise customer requires it.
E-filing is commoditized and jurisdiction-specific; building direct court integrations is wasted effort.
How: File the finished memo through existing EFSPs rather than integrating courts directly.
How smarter mitigation persuades the people who set sentences
Aperio cannot reduce sentences: it persuades the people who set them. A sentence is chosen by four decision-makers: prosecutors (the plea), judges (departures/variances and alternatives-to-incarceration), parole boards, and governors (clemency). Aperio's lever on all four is the same: make the defense team's mitigation more complete, better corroborated, delivered earlier, and cheaper to produce. That is where the evidence is real and where overclaiming is legally dangerous for a no-fabrication, not-legal-advice product.
The honest evidence base. The closest thing to causal proof is the Anderson, Buenaventura & Heaton holistic-defense study (Harvard Law Review, 2019): 500,000+ Bronx cases over a decade, exploiting as-if-random assignment between provider organizations. Holistic representation did not change conviction rates but cut the likelihood of a custodial sentence by ~16% and expected sentence length by ~24% (~1.1M fewer jail/prison days, ~$160M saved, no rise in recidivism). Treat this as directional, not dispositive: it is a single jurisdiction, it is a natural experiment (not literally a designed RCT), and, critically, it measures the entire bundled holistic staffing model (social workers, civil-legal-aid, life-circumstance advocacy), NOT mitigation/social-history quality in isolation and NOT software. Attributing the 24% specifically to better mitigation work product, or equating it with an AI copilot, is an inferential leap the study does not support. No RCT links any AI mitigation tool to lower sentences, per-case counterfactual sentences don't exist, and confounders (charge, judge, jurisdiction) are large. So the claim is mechanism-level: Aperio improves the input that the strongest available evidence associates with shorter custodial sentences.
Scope caveat. The constitutional mitigation-investigation mandate (Wiggins, Rompilla) and the 'multigenerational bio-psychosocial history' standard-of-care (ABA Supplementary Guidelines) are strongest in capital and appellate work; run-of-the-mill non-capital felony/misdemeanor cases, most of the addressable market, do not carry the same enforceable duty or a funded mitigation-specialist budget line. That is simultaneously Aperio's wedge and its hardest go-to-market problem.
The levers
The economics are lopsided in mitigation's favor, and the binding constraint is specialist time. One avoided year of custody costs the system $44,090 (federal COIF, FY2023), a ~$61k median state figure, and $100k–$285k in high-cost states (CA ~$132k, MA ~$285k). Against that, mitigation labor is cheap but scarce: mitigation specialists earn roughly $24–$34/hr as a wage (and bill ~$50–$100/hr court-appointed; capital investigators $75–$150/hr for hundreds of hours). The defense market already pays $5k–$25k for a single sentencing-mitigation video: proof teams pay real money for mitigation work product.
Aperio's ROI is as a force-multiplier on the scarcest input, not a sentence-reduction guarantee. Public defenders routinely carry well above the new RAND standards (which imply ~35 hrs for a low-severity felony up to 286 for LWOP (life without parole); sustainable load ~59 felonies/attorney/yr), with worst-case overloaded jurisdictions at 300+ felonies/lawyer, the egregious tail, not the norm. Washington cut felony caseload caps from 150 to 47/year; states report 30–67% attorney shortfalls. The right unit metric is specialist-hours-per-case and cases-per-specialist: more complete, corroborated mitigation per scarce hour, surfaced earlier and at lower cost.
Prove it with Aperio's own data. Because no external RCT links the tool to outcomes, instrument outcome capture natively: initial-offer-vs-final delta, guideline-vs-imposed (months), variance/departure flags, ATI/diversion placement, capital-resolved-short-of-death, parole/clemency grant rates, plus process metrics (domain coverage, corroboration rate, provenance completeness, specialist-hours/case). Design for matched-comparison / difference-in-differences and pre/post-adoption cohorts. The proprietary outcomes dataset is both the moat and the only credible path to substantiating the thesis. Pair this with a jurisdiction-specific ROI calculator (specialist-hours saved × loaded hourly cost) to justify procurement to PD offices and grant funders, and hold outcome metrics to the same no-fabrication, honest-baseline standard the product enforces on casefiles (e.g., don't count a charge reduction that would have happened anyway).
Roadmap
Ship the defensible spine (file-first casefile, provenance engine, coverage meter) and lead with plea-stage mitigation, where ~95%+ of cases are actually decided. Land the easiest engineering beachhead (Clio) without betraying the privilege architecture.
- File-first ingestion: secure direct upload + Google Drive/SharePoint, parsing into a casefile with page-level provenance on every fact.
- Casefile-as-shared-state organized by mitigation domain + life timeline; auto-built chronology and multigenerational genealogy as first-class objects.
- Provenance-or-quarantine: citation cards with pin-cite deep-links; uncited facts blocked from work product.
- Fact lifecycle chips (Lead/Finding/Corroborated/Disputed) with human-gated promotion.
- Domains Coverage Meter (radial + per-domain Gap Queue) as the home screen.
- Early-mitigation / mitigation-forward plea packet as the flagship drafting output, with mandatory cite-check + attorney sign-off.
- Public Clio Manage app + App Partner Program listing, architected around the 3 req/sec per-app limit (webhooks, batching, backoff).
- JusticeText/Reduct manual transcript import with provenance preserved.
- Provable no-train, segregated-tenancy (or on-prem) data architecture + persistent work-product/privilege affordances.
Extend from intake-to-sentencing across the real defense lifecycle, open the public-defender distribution channel behind government procurement, and broaden PM/DMS coverage at low marginal cost.
- PSR-objection assistant: diff probation's social history vs. the corroborated casefile, draft Rule 32 objections inside the 14-day window.
- Expert-referral triggers (FASD/TBI/low-IQ/trauma indicators) that auto-assemble the social-history packet the expert needs.
- Three-artifact Structured AI-assisted drafting from one casefile: social-history report, §3553(a) sentencing memo (inline pin-cites), mitigation-video brief.
- Cite-check export gate emitting a timestamped verification PDF; redaction-verification + burn-in preview.
- BD + security-reviewed pilots with one defenderData and one DEFENDERbyKarpel office (expect SFTP/flat-file or vendor-mediated integration; enforce defense/prosecution data separation).
- MyCase + Smokeball connectors (reuse Clio sync abstraction; budget partner enrollment).
- NetDocuments + iManage DMS connectors for appellate/capital orgs.
- Native outcome instrumentation (offer-vs-final, guideline-vs-imposed, variance/ATI flags, specialist-hours/case) feeding a difference-in-differences evidence base.
- ATI/diversion eligibility matcher mapping documented mitigation to local treatment/drug-court criteria.
- Risk-tiered propose/commit autonomy with a measured friction budget and per-task error tracking.
Extend value across the full post-conviction lifecycle, attack the biggest manual cost (records) with realistic expectations, and substantiate the value thesis with proprietary outcome evidence.
- Records-retrieval workflow: request tracker + auto-generated jurisdiction-correct authorization/subpoena packets tied to record categories (software-first; fulfillment partnerships only after validating criminal-specific custodian coverage, not an EvenUp-style PI services clone).
- Clemency/parole module reusing the casefile (personal statement, support letters, programming documentation).
- Sentencing-video handoff (narrative arc, source list, storyboard/shot-list) to established producers or DIY.
- Comparative/disparity sentencing analytics integration (federal-scope, framed as disparity evidence, never a predicted reduction).
- Jurisdiction-specific ROI calculator (specialist-hours saved × loaded hourly cost) for procurement and grant funders.
- Matched-comparison / pre-post outcome study using Aperio's own dataset: the moat and the credible proof of the mechanism.
- Workload-triage view aligned to RAND hour standards so overloaded offices can see which open cases have zero mitigation coverage.
- Optional Westlaw/Lexis link-out + gated cite-checking and EFSP filing handoff for enterprise customers that require them.
Sources
- Anderson, Buenaventura & Heaton, The Effects of Holistic Defense on Criminal Justice Outcomes, Harvard Law Review (2019), natural experiment: ~16% lower custody odds, ~24% shorter sentences, no recidivism rise (directional, single-jurisdiction, bundled staffing model)
- RAND/NCSP National Public Defense Workload Study (2023): ~35 hrs low-severity felony to 286 hrs LWOP; caseloads run well above standard
- USSC 2024 Sourcebook of Federal Sentencing Statistics: 45.7% within-range; ~19.4% non-government downward variances
- NPR: The vast majority of criminal cases end in plea bargains (~95% state / ~98% federal)
- Pew: Only ~2% of federal criminal defendants go to trial (FY2018; FY2023 trial rate ~2.8%)
- 18 U.S.C. §3553(a): sentencing factors / history and characteristics / parsimony principle
- Fed. R. Crim. P. 32: Presentence investigation report and 14-day objection window
- Wiggins v. Smith, 539 U.S. 510 (2003): failure to investigate mitigation = IAC
- Rompilla v. Beard, 545 U.S. 374 (2005)
- Eddings v. Oklahoma, 455 U.S. 104 (1982): Lockett/Eddings individualized-sentencing doctrine
- ABA 2008 Supplementary Guidelines for the Mitigation Function of Defense Teams (capital): Guideline 5.1, multigenerational bio-psychosocial history
- ABA Formal Opinion 512: Generative AI Tools, confidentiality, supervision, vendor data-diligence (UNC Law Library analysis)
- Magesh et al., Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools (Stanford RegLab; JELS 2025): Lexis+ AI ~17%, Westlaw AI-Assisted ~33% hallucination
- Clio Developer Documentation Hub: API V4, OAuth 2.0, webhooks
- Clio API Rate Limits: 3 req/sec per app (HTTP 429), separate regional hosts/limits
- JusticeText: AI AV-evidence review for public defenders (bodycam/jail-call transcription)
- Justice Works: defenderData indigent-defense case management
- DEFENDERbyKarpel: defense case management (sibling of PROSECUTORbyKarpel)
- NetDocuments: document/email management (150+ integrations, modern API)
- iManage: SeeUnity technology partner / connector ecosystem
- Clearbrief: AI legal writing + Cite Check Report audit trail
- Harvey, Vault & Assistant: hover citation preview, 'View Reference', @-mention grounding
- Relativity aiR: natural-language rationales with source citations, aiR for Privilege, audit trail
- Everlaw: privilege review/log workflows (documented silent redaction-drop failure modes)
- Federal Register: Average Cost of Incarceration Fee (COIF), FY2023 = $44,090
- USAFacts: How much states spend on prisoners (median ~$61k; range to ~$285k)
- Washington State Standard: Supreme Court slashes public-defender felony caseload caps (150→47)
- FASD and the criminal-justice system: a guide for legal professionals, 17–36% prevalence, multidisciplinary assessment
- Tavrn: Medical Record Retrieval Costs by State (per-page fees: CA $0.25/pg, MD $0.76/pg)
- Doug Passon Law: sentencing-mitigation videos ($5,000–$25,000)
- EvenUp launches Pre-Litigation-as-a-Service (PI records-retrieval model, cited as a contrast, not a transferable analog)
- UC Berkeley Law CLJC: Existing AI Tools for Public Defenders (JusticeText, Reduct, NiCE, CoCounsel, SentencingStats)
- Center for Community Alternatives: Sentencing Mitigation program (self-reported 90% reduced sentence/diversion; treat as program data, not causal proof)