Public diligence artifact

Proof Fabric shows the chain, not just the claim.

AEGIS is evaluated by whether a supported source path can be inspected end to end: source → evidence → catch → human gate → accepted memory/action → governed improvement proposal → signed apply → verification report.

The inspected chain

One supported source path, end to end.

This is the proof pattern AEGIS uses internally and exposes for evaluator diligence. It proves a bounded path, not universal source coverage.

01 Source

Real or synthetic source event

CRM, HubSpot, IMAP/email, document or task input is read under tenant source grants.

scoped
02 Evidence

Citable record

AEGIS keeps source context separate from accepted memory and attaches evidence/source links where supported.

citable
03 Catch

Detected miss or risk

A supported detector or Radar path proposes a catch only when evidence and measurement state support it.

proposed
04 Gate

Human approves or rejects

No consequential external write and no durable accepted Memoria happens without a human decision.

signed
05 Memory / action

Accepted outcome

Approved catches can become accepted Memoria or prepared internal/source-native drafts on selected paths.

governed
06 Improve

Proposal, not mutation

Weak feedback or repeated patterns can create a governed improvement proposal with a diff and evidence.

reviewed
07 Verify

Before/after report

Signed applies open a verification watch and report improved, no-change, worse or inconclusive.

measured
Latest public proof card

ADE / Proof Fabric v6.

Current evidence level: repeatable proof on a fixed synthetic enterprise taxonomy with real IMAP + HubSpot CRM fusion. This is strong proof of the chain shape, not arbitrary detector coverage.

8/8Strict scenario/surface finds
4Clean repeated runs
0False positives
0Ungated actions
Status matrix

What this makes visible.

The goal is to turn “live on selected paths” into “evidenced and repeatable,” while keeping every boundary explicit.

Live / evidenced

  • Source-to-evidence chain on supported paths.
  • Catch Ledger / Radar proposal flow.
  • Human approval boundary.
  • Memoria as accepted memory, separate from raw retrieval.
  • Governed improvement proposal → signed apply → verification.

Live seed / hardening

  • Canonical Business Timeline / Source Intelligence.
  • Broader detector proof beyond the proven path.
  • Full model-policy metadata surfacing.
  • Prompt-injection/content-boundary labeling.
  • Subject-level erasure proof.

Not claimed

  • Universal cross-source reasoning.
  • Regulated-enterprise production readiness.
  • Claim-grade self-hosted/air-gapped deployment.
  • Autonomous external writes.
  • Automatic trusted Memoria writes.
Evaluator runbook

How to test AEGIS in a pilot.

A pilot should measure the proof chain, not the number of visible screens.

Pilot scope
One business seam, one source pair, proposal-only first. Example: CRM + mailbox follow-up vigilance, or document requirement review.
Evidence required
Source record, evidence card/source link, catch fingerprint, approval/rejection record, resulting memory/action record if signed, and verification outcome if improvement is applied.
Success metric
Accepted catches, false-positive rate, approval latency, cost per accepted outcome, and whether the buyer says “we would have missed this.”
Stop condition
If the source is unavailable, unsupported, denied, stale, or unmeasured, AEGIS should say so instead of manufacturing a catch.

Use Proof Fabric for diligence.

Read the framework, tech stack and evaluator Q&A next; then ask AEGIS to prove one supported path on your own data.

Framework Tech Stack Evaluator Q&A Pilot shape