Research Intelligence / Audit-Grade

Research that defends itself.

Reproducible regulatory and financial research with citation provenance and replay determinism.

Example QuerySynthetic Data For Illustration
run_8f2c1a4e
Basel III Tier 1 capital requirements under CRR III amendments
Deep Diligence

CRR III revises the Tier 1 capital baseline by tightening the output floor to 72.5% of standardised RWAs once fully phased in (2030). Common Equity Tier 1 (CET1) remains the dominant stack and must comprise at least 4.5% of risk-weighted assets, with Additional Tier 1 instruments capped at 1.5%.

Institutions classified as G-SIIs carry a CET1 buffer surcharge ranging from 1.0% to 3.5% on top of the conservation and counter-cyclical buffers; failure to meet the combined buffer triggers automatic distribution restrictions on dividends, AT1 coupons, and discretionary bonus payments.

[1]Reg (EU) 575/2013 art. 92p.87[2]EBA/RTS/2024/03p.14
Hash matchReplay

Trusted by compliance and research teams at

Apex ComplianceMeridian AdvisoryNorthpoint CapitalStella RegTechVeritas Audits

Built for the full research loop

Ask, run, and ingest — every step leaves a verifiable trace.

What are the latest ESMA guidelines on…
[1] ESMA 2025 p.12

Chat

Natural-language queries with cited, replayable answers.

ESMA inquiry
a3f2b1c9
MiFID II review
7d1ef4a2
AML assessment
c8b5e3d0

Runs

Compare runs, inspect hashes, and export audit reports.

PDF
ESMA_2025_Q1.pdfuploaded

Ingest

Upload documents and track lineage from source to citation.

How replay works

Determinism is structural, not promised. The example run below is bound to a frozen prompt and an append-only chain.

Frozen Prompt
sha256:c4f29b76e3a5d8c1
Chain Hash
blk3:7a4f2c8e9…6c5d3f
Replayed
2026-05-08T14:32:11Z
  1. Frozen prompt at run time

    Each run captures the system prompt, model id, tool surface, and retrieval parameters as a SHA-pinned bundle. The bundle is the contract for what the agent saw.

  2. Append-only BLAKE3 chain

    Every tool call, citation, and final answer hashes into a chain bound to the run id. Tampering with any link breaks the chain; verifiers detect it without trusting the operator.

  3. Replay against the chain

    A replay re-executes the run against the frozen prompt and compares hashes link by link. Match means the answer is reproducible. Divergence is surfaced with the first link that drifted.

Every run is auditable.

From prompt to citation, Deep Diligence records the inputs, outputs, and chain of custody required to defend a research conclusion.

Frozen prompts

Every run pins the system prompt, model, and tools so the agent's context is preserved exactly.

Page-level citations

Answers carry citations anchored to source pages, not opaque links.

BLAKE3 chain

An append-only hash chain binds each step; tampering breaks the chain visibly.

One-click replay

Re-run any question months later and compare link-by-link to detect drift.

What teams are saying

Compliance and research teams rely on Deep Diligence for verifiable, reproducible outputs.

Deep Diligence eliminated the audit-trail anxiety we had with every regulatory filing. Now we just share the run ID.

Sarah Chen

Head of Compliance, Meridian Advisory

The replay determinism is the killer feature. Our auditors can re-run any analysis from six months ago and get the same result, hash-verified.

Marcus Webb

Director of Research, Northpoint Capital

We evaluated six research platforms. DeepD was the only one that could prove its outputs were reproducible. That closed the deal.

Dr. Amara Okafor

Chief Risk Officer, Apex Compliance

Start defending your research today.

Join teams that replace “trust me” with “here is the run id.”