Contact

The AI Era / 2025-present

The method did not change when the industry did

The company's through-line is that the product has always been the check: proprietary verification of a specific population, monetised from whichever side paid better. Storage Unit Auction List telephoned fifty thousand facilities and re-verified within forty eight hours. Boost graded returns item by item on a weighted scale. Inventory Scouts credit-checked suppliers. Moderra.ai is the purest expression of that thesis in the company's historyC.

Moderra's own homepage does not lead with blocking unsafe content. It leads with the opposite failureV:

"Letting unsafe content through isn't the only way moderation breaks trust. Crude refusals do it too, and most teams never measure the cost."

moderra.ai, August 2026 V

They call it the over-refusal tax. The argument is that a crude binary check destroys value, and that a graded, reasoned, auditable check creates it. That is the identical argument Boost Liquidation made in 2012 when it graded returned goods on a weighted scale instead of sorting them pass or fail, and the identical argument Storage Unit Auction List made by re-verifying rather than scraping once. Fourteen years apart, in industries with nothing in common, the same insight: the value is not in the yes or no, it is in the grade and the reason attached to itC.

The proof is in Moderra's own API response, which the site publishesV:

{
  "id": "mod_789xyz",
  "action": "flag",
  "reason": "likeness_violation",
  "confidence": 0.98,
  "route_to_human": true
}

An action, a reason, a confidence score, and an escalation path. This site carries confidence markers on every claim for exactly the same reason: a check that shows its work is worth more than a check that does not.

Section 01 / The stack

4 layers

How a decision moves through the product

The same diagram the operating stacks carry on the portfolio page. Unlike those, this sequence is Moderra's own published pipeline rather than an arrangement this record inferred, and its confidence marker says so.

Stack 04

The AI Stack C

The print stack and the mail stack were arrangements this record inferred from the company's own division headings. This one is different: Moderra.ai publishes its own pipeline, and the sequence below is its own description of how a decision moves through the product. What is inferred here is only the placement of Person.ai beside it, and the reading that both continue a method the company has used since 2011. Input arrives, policy is enforced in real time, the ambiguous cases route to a human, and every decision leaves an audit record a regulator can read. That is a graded, reasoned check with an escalation path, which is the same instrument Storage Unit Auction List pointed at storage facilities and Boost Liquidation pointed at returned goods.

  1. L1

    Enforcement

    The Guardrails API applies policy to prompts, outputs and multi-turn sessions in real time, with latency optimised for production use. Coverage is multimodal: prompt injection, toxicity, PII redaction and jailbreak attempts in text and code; deepfake detection, CSAM, gore and likeness controls in image and vision; real-time interruption, transcript safety and crisis detection in audio and voice; frame-by-frame analysis, audio-visual alignment and consent registries in video. The response carries an action, a stated reason, a confidence score and a route-to-human flag rather than a bare allow or deny.

  2. L2

    Human review

    Purpose-built queueing and disposition workflows route the cases policy cannot settle to trust and safety teams. This is the layer that distinguishes the product from a filter, and it is the same structural decision the company made in 2011 when Storage Unit Auction List staffed a seven-seat verification office rather than trusting a scraper, and in 2012 when Boost Liquidation graded returned goods by hand on a weighted scale rather than sorting them pass or fail.

  3. L3

    Evidence

    Immutable audit logs map operational decisions to regulatory requirements, the EU AI Act among them, and the product is positioned as aligned with AI compliance and child protection frameworks rather than certified against them. Continuous website audits watch public surfaces for disclosure gaps and policy drift. Adversarial red-team simulation and runtime security analytics test the system rather than describing it. The output of this layer is a defensible record, which is the same product the company has sold in five industries: not the decision, but the documented reason for it.

  4. L4

    Applied products

    Person.ai is placed here as the applied side of the same thesis rather than as a component of Moderra. Announced in September 2025 as part of the pivot away from Epic Print after Hurricane Helene, it read a company's data sources overnight and delivered a verified digest each morning. Its present operation is not confirmed: the domain now serves a different product, and whether that represents a pivot by the same team or a change of hands is not established in the public record. The venture appears in this record for what it demonstrably was in 2025, and its status says so.

Section 02 / The ventures

2 entries

Two records, held to the same standard

Each entry carries the status and the confidence the documents support, and each caveat is part of the record rather than a footnote to it.

Moderra.ai logo

Moderra.ai V

Guardrails, human review and audit-ready evidence for AI products

Active Present

Moderra.ai gives AI product teams one place to enforce guardrails, route risk, review edge cases and produce audit-ready evidence across text, image, video and voice. Maddison Lake is chief executive. Six service families are orchestrated through a single API: a guardrails API for real-time policy enforcement on prompts, outputs and multi-turn sessions; human review queueing and disposition workflows; immutable compliance evidence mapping decisions to regulatory requirements including the EU AI Act; continuous website audits watching public surfaces for disclosure gaps and policy drift; agent and quality controls covering groundedness, task adherence, protected material and copyright risk; and adversarial red-team simulation with runtime security analytics.

Its commercial argument is unusual and worth stating precisely, because it is not the argument a moderation vendor normally makes. Moderra leads not with the harm of letting unsafe content through but with the opposite failure: the crude refusal. It calls this the over-refusal tax, and holds that every false block kills a legitimate request, churns a user, spikes a support ticket and quietly collapses trust, at about the same cost as the failure everyone does measure. The alternative it offers is policy enforcement that distinguishes signal from noise: allow what is safe, flag what is grey, route what is hard, refuse what is clearly out of bounds, each with a reason a team and a regulator can read.

Coverage is genuinely multimodal. Text and code: prompt injection, toxicity, PII redaction, jailbreak attempts. Image and vision: deepfake detection, CSAM, gore, likeness controls. Audio and voice: real-time interruption, transcript safety, crisis detection. Video: frame-by-frame analysis, audio-visual alignment, consent registries. The published API response carries an action, a stated reason, a confidence score and a route-to-human flag rather than a bare allow or deny, which is the whole thesis expressed in a JSON object.

Person.ai (2025) logo

Person.ai (2025) C

Executive briefing tool: 65+ data sources read overnight, digest by 7 AM

Unverified 2025

Person.ai was announced by Maddison Lake in September 2025 with a public waitlist and an alpha, described explicitly as part of the pivot away from Epic Print after Hurricane Helene and built by a team that had been working in stealth for several months. The product read a company's data sources overnight, more than sixty-five of them including Salesforce, QuickBooks, Slack, ERP systems, Shopify, Stripe, Xero, HubSpot and Microsoft Teams, and delivered a condensed summary email and a five-minute private podcast by seven each morning. Its own headline put it as the entire business, summarised before sunrise, and its marketing cited advanced reasoning models and real-time alerts on cash flow changes and sentiment shifts.

That is where the verifiable record ends, and this record stops with it. The domain person.ai now serves an unrelated consumer product. Whether that represents a pivot by the same team, a change of ownership, or two ventures sharing a name is not established anywhere in the public record, and the Internet Archive does not resolve it: the captures run from a pre-launch redirect in early 2025 through the executive product in August 2025, and the next preserved body is from 2026. This entry therefore describes what Person.ai demonstrably was in 2025, which is well documented and genuinely part of the post-Helene story, and makes no claim about what the domain serves today.

Section 03 / Coverage

2 tables

What Moderra publishes about itself

Both tables reproduce Moderra's own published descriptions and are marked V for exactly that reason. Nothing here is paraphrased into a stronger claim.

Six service families, orchestrated through one API V
Service family Function
Guardrails API Real-time policy enforcement for prompts, outputs and multi-turn sessions, latency optimised for production
Human Review Purpose-built queueing and disposition workflows routing edge cases to trust and safety teams
Compliance Evidence Immutable audit logs mapping operational decisions to regulatory requirements such as the EU AI Act
Website Audits Continuous monitoring of public surfaces for disclosure gaps, policy drift and unsafe experiences
Agent & Quality Controls Groundedness checks, task-adherence enforcement, protected-material detection, copyright-risk controls
Security & Testing Adversarial red-team simulation, live-stream moderation, runtime GenAI security analytics
Multimodal coverage V
Modality Detections
Text & Code Prompt injection, toxicity, PII redaction, jailbreak attempts
Image & Vision Deepfake detection, CSAM, gore, likeness controls
Audio & Voice Real-time interruption, transcript safety, crisis detection
Video Frame-by-frame analysis, audio-visual alignment, consent registries

Section 04 / The close

One instrument, fifteen years

For twenty years the product has been the check, and never the yes or no. Storage Unit Auction List telephoned fifty thousand storage facilities and re-verified every listing within forty eight hours of the auction. Boost Liquidation graded returned goods item by item on a weighted scale rather than sorting them into sellable and scrap, because the money was in the gradations. Inventory Scouts credit-checked its suppliers before it listed them. Moderra.ai reviews the output of AI systems and returns an action, a reason, a confidence score and a route to a human, then keeps an audit record a regulator can read. Different industries, different decades, one instrument: a graded check that shows its work, sold to whichever side of the transaction valued it more.

Read Part XI, The AI Era, in the full history