docs(reflow/architecture): trim to consumer-depth system overview
The page had drifted into contributor-depth territory — service
class names, Lua scripts, conditional tool registration, middleware
execution. That content is aimed at someone about to open the
source, not someone evaluating Reflow for their institution or
integrating via the API. The reflow repo's
docs/explanation/architecture.md (1421 lines) already covers all
of that at the right depth for contributors; duplicating a partial
copy here just creates a drift machine.
Rewrite from 176 → ~115 lines focused on the questions a consumer
or evaluator actually asks:
- What is Reflow composed of (the three services)
- System diagram (kept — useful at any depth)
- How a document moves through the system (data flow narrative)
- How real-time progress works (SSE architecture summary)
- Technology choices (as context, not implementation detail)
- Privacy and security posture — new section. Institutions care
about this before piloting; previous version didn't surface it
- Deployment shapes — local / UIC production / other deployments
- Resilience summary — short, links out for thresholds
Dropped sections (content is intact in the contributor repo):
- Service Layer per-class descriptions (ProcessingService, StorageService,
QueueService, JobService, PIIDetectionService — implementation detail)
- AI Agent Architecture internals (conditional tool registration,
token waste optimisation — implementation detail)
- Infrastructure port table (duplicated in AGENTS.md in the reflow
repo; consumers don't need port specifics)
- Resilience specifics (circuit breaker thresholds etc. — now a
short summary with a link to the contributor repo's deep version)
New bit worth calling out: the Privacy and security posture
section. That's what institutional IT, legal, and procurement
want to read before green-lighting a pilot, and the old version
buried it in the data flow bullet.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>