Chain of custody tracks the object. Chain of reasoning tracks the thinking. Reasoning Canvas captures why each step exists while the work happens, so the case can later explain itself.
When a regulator, court, client or internal reviewer asks why a decision was made, teams should not have to reconstruct the answer from logs and memory. The canvas turns case work into a graph of triggers, transformations, references and actions, each with rationale attached.
Chain of custody tracks the object. Chain of reasoning tracks the thinking. We already do the first. We are building the second. Simply Discover - Reasoning Canvas Vision
The problem
These are familiar to anyone running regulated case work. They are expensive because the reasoning sits outside the system that did the work.
01
When the ICO, a court or a client asks how a particular email was classified, redacted or excluded, teams assemble the answer from audit logs, classifier outputs and people’s recollections. That reconstruction is expensive, late and never quite complete.
02
Experienced analysts know the playbook for DSARs, privilege reviews and investigations. Today that knowledge transfers through training and shadowing, not through tooling. When they leave, the playbook leaves with them.
03
Two cases that follow nearly the same shape still begin with the same set of decisions from scratch. The 200th DSAR of the year is configured with the same care as the first because there is no place to capture the pattern.
How it works
Hover any step in this DSAR investigation to see what the officer recorded about why that step exists. Every step carries this. At the end of the case, we walk the graph and produce a narrative the regulator can read.
Hover or tap a step to read the officer’s rationale
The model
Keeping the model small is what makes it usable: four building blocks, three connection types and an open-ended set of behaviours inside each block.
Trigger
starts a flow · out only
When does work begin? An officer clicks open case, an email arrives, a schedule fires, or a deadline is reached.
On-demand, on-event, on-schedule
Transform
in → out
Changes the shape or count of what flows through: filters, classifiers, entity extraction and synthesis.
Filter, classify, extract, synthesise
Action
in only · reaches outside
Where the system commits to a real-world effect: sending, exporting, creating a case, or flagging for review.
Send, export, create case, notify
Reference
no flow · context only
Context that other steps read but does not move data: intake memory, knowledge, notes or prior decisions.
Memory, document, knowledge, sticky note
What the regulator sees
The graph is not just the report’s source. It is the report. We walk it in dependency order and write each step as a paragraph, using the rationale recorded at the time.
The investigation began on 2026-03-14 when A. Halford opened the case in response to a data subject access request from Jane Doe.
Trigger - On-demand · node #1 · 09:12 UTCCorrespondence was gathered from the subject’s mailbox covering the period 2025-03-01 to 2025-09-30. The rationale recorded at the time: “Subject’s complaint refers to HR events in summer 2025; extended to cover surrounding context.”
Filter by custodian · Filter by date range · nodes #2-3 · 09:14 UTCAboutness classification identified 147 emails as being about the subject. Forty-three were labelled irrelevant; eighteen were deferred for further context. The prompt and model fingerprint are available in the appendix.
Classify · node #4 · 09:16-09:42 UTCWho uses it
Senior analyst · process owner
Every DSAR response should start by pulling mail from the subject’s mailbox, then filter to the date range, then classify for aboutness. I want to record the playbook once and have everyone follow it consistently.
Case-working officer · investigator
The subject mentioned a complaint about their manager in March. Let me search that. This classifier flagged privileged documents. I need to see why. My draft response does not mention the HR thread; let me add a step to pull that in.
Positioning
Reasoning Canvas borrows familiar patterns from workflow builders, notebooks and DAG tools, but the product position is different: evidence-grade reasoning for regulated investigations.
| Tool | What it does well | Why Reasoning Canvas is different |
|---|---|---|
| n8n / Zapier | Automation pipelines, trigger-first | The graph is not just plumbing. It records why the work happened and what evidence supported the decision. |
| Miro / FigJam | Free-form visual thinking | The graph executes and records case reasoning. It is not a sketch detached from the workflow. |
| Jupyter notebooks | Reproducible reasoning | The canvas is branching, case-aware and framed for legal and regulatory audit rather than code-first analysis. |
| Airflow / Nextflow | Execution DAGs | The audience is officers, analysts and reviewers, with human decision points and narrative export built in. |
The path
The canvas can land incrementally: first as a prototype, then as a persistent reasoning layer, then as templates, triggers and narrative export.
Phase A
Front-end prototype
Phase B
Persistence
Phase C
Append-only edits
Phase D
Templates & slots
Phase E
Triggers wired
Phase F
Narrative export
Phase G
Migrate cases
Phase H
Rich content
Why now
UK GDPR · EU AI Act · ICO
A team that can produce a reasoning record on demand is in a different posture from one that produces a log dump.
200x
Repeated DSARs, FOI requests and reviews stop starting from a blank page when the proven graph can be reused.
5 min x n
Spot-policy matches, inbound secure messages and draft rejections can be handled on-canvas, auditably.
We can walk through how Reasoning Canvas would sit alongside Discovery, DSAR, FOI, investigations and compliance workflows.