If you have 5 minutes:

1. Read SUBMISSION.md.
2. Run python -m pytest.
3. Run python src/generate_evidence.py --reviewer-demo.
4. Open evidence/manifest.json and evidence/scenarios/stream_outage_with_fallback_replay.md.
5. Read docs/evidence_log.md for claim-to-proof mapping.

Accepted-Event Durability Proof Packet

This repo is a small failure-semantics lab for a real-time analytics migration. It does not benchmark AWS locally. It proves the reliability boundary the architecture depends on, then backs the plan with source-labeled capacity math and executable scenarios.

Setup

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt

Python 3.11 or newer is expected. The artifact runs without AWS credentials.

Reviewer Commands

python -m pytest
python src/generate_evidence.py --reviewer-demo

The evidence command regenerates only known files under evidence/ and leaves unknown files alone.

Evidence Files

• evidence/manifest.json: command, timestamp, Python version, generated files, proven and unproven claims.
• evidence/capacity_model.json: source-labeled deterministic capacity and directional cost model.
• evidence/scenarios/*.md: scenario summaries with counters, invariant result, limitations, and operator actions.
• docs/evidence_log.md: claim-to-proof ledger.
• docs/sources.md: official/public sources and access date.

Reviewer Checklist

• The reliability boundary is accepted events, not browser-generated events.
• The simulator asserts accepted_count == processed_unique_count + duplicate_deduped_count + erased_or_tombstoned_count + pending_replay_count.
• Fallback write failures are not counted as accepted loss because they are not acknowledged.
• Cost and scale numbers are editable assumptions or calculations, not quotes.
• Compliance language separates hot-state deletion from legal deletion policy.