Synchronized starts
ReadyEdge-hosted bootstrap and per-attempt authority absorb synchronized launch patterns. Load testing confirms the architecture handles thousands of learners starting in the same window.
Delivery
Built for final exams, certifications, and other high-stakes assessments.
QFlowLearn runs assessment delivery on Cloudflare Workers and Durable Objects: fast serverless primitives that spin up on demand, isolate each attempt, absorb launch and submit spikes, and avoid the old model of spinning up servers that get overwhelmed as usage spikes.
Delivery operations
What the runtime handles on assessment day.
The hard parts of high-stakes assessment delivery are operational: concurrency, connectivity, revisions, and audit gaps. QFlowLearn handles those assessment-day patterns directly, with load-tested confidence for thousands of concurrent learners.
Deadline submissions
ReadySubmit is built around durable receipts and idempotency. The finish spike moves through a Cloudflare-managed queue instead of landing on an overwhelmed server.
Connection drops
ReadyAn IndexedDB-backed recovery store records work locally before send, restores queued work after reload or tab reopen, and replays on reconnect with durable receipt coverage.
Stale tabs and duplicate submits
In build for launchPer-attempt authority and idempotent submit keys prevent an old browser session from overwriting authoritative state.
Post-publication edits
ReadyPublished assessment versions are immutable. A correction becomes a new version with auditable provenance. There are no silent rewrites of work in progress.
Bad questions and wrong keys
ReadyRe-scoring runs as an overlay on raw attempt evidence. The original responses are never destroyed. Re-scoring is an additional, auditable record.
Late submits and reconnects
ReadyReconnect replay and receipt coverage preserve the learner action first, then reconcile timing and policy against the attempt ledger.
Scoring corrections and result challenges
ReadyFinal scores are projections from raw evidence, recomputable from source. The attempt ledger supports audit, review, and rescoring.
Operations packet
Buyers need operational evidence, not just architecture.
The Cloudflare architecture matters because it changes how launch, submit, reconnect, scoring, export, and recovery behave under stress. The packet should turn that into concrete evidence rows an evaluator can score.
| Operational scenario | Evidence to include | Status |
|---|---|---|
| Synchronized start window | Load-test summary, launch plan, edge bootstrap path, package version proof, monitoring signals | Ready |
| Deadline submission window | Load-test summary, durable receipt sample, idempotency behavior, queue and scoring handoff | Ready |
| Bad connection or replayed submit | IndexedDB recovery store, offline replay E2E, reload and tab reopen tests, degraded submit recovery test | Ready |
| Stale tab or duplicate submit | Per-attempt authority test and stale-session reconciliation evidence | Ready |
| Bad item or wrong key | Immutable content version, scoring overlay, correction provenance sample | Ready |
| Incident and recovery response | Escalation roles, support timeline, audit export, remediation queue view | In build for launch |
| Load-test evidence | Load-tested pattern for thousands of concurrent learners on Cloudflare serverless primitives | Ready |
Why Cloudflare Workers
Serverless primitives fit assessment spikes.
Synchronized launches, deadline submissions, and reconnects are burst-load problems. Cloudflare Workers, Durable Objects, and Queues let delivery work scale with demand: many small units of work, one clear authority per attempt, and a durable path for submissions during traffic spikes.
Workers at the edge
Bootstrap and item delivery run near the learner, start quickly, and avoid a single regional VM bottleneck.
Durable Objects per attempt
Each learner attempt has a single authoritative actor. Stale tabs and duplicate submits route to the same authority and lose to it.
Queues for submit
Submission is durable, idempotent, and absorbs the deadline spike. Scoring and exports run from the queue without backpressure on learners.
Attempt evidence
Final scores are projections, not the source of truth.
The attempt event ledger is the authoritative artifact. Final scores are computed from it and can be recomputed at any time. When a result is challenged, we do not ask the institution to trust a single number. We provide the attempt trail.
Internally we treat the runner runtime as an internal codename. Externally, the
public delivery hostname is take.qflowlearn.com.
Read the runtime contract.
Technical evaluators can review the runtime contract, the network recovery proof, the architecture summary, and the load-test plan.