Research

Transparent Continuation
Checkpointing.

Why TCC exists: durable recovery by preserving resumable execution state rather than reconstructing program position from accumulated history.

TCC eliminates prefix replay when recovering from a committed continuation checkpoint.

Read the technical report →

Current engine evidence

Portable TCC Engine vs matched history replay

Same suite, host, workload, and durable effect journal — only the recovery architecture differs. Journaled effects prevent the provider from firing on replay. Host: Apple M1 Pro, Node 22, tcc-engine 26.10.0. Not production Cloud latency.

Matched recovery latency vs completed history depth at ~4 KB live continuation state 0.05 0.1 1 10 100 10 100 500 1k 5k ms (log) completed history Replay TCC
Portable TCC Engine vs matched history replay · ~4 KB live continuation · log Y-axis

0.042–0.066 ms

TCC median recovery across depths 10 → 5,000 at ~4 KB live continuation.

2.7 → 336 ms

Matched history-replay median over the same depths (~41× → ~5,186× vs TCC).

0.065 → 0.285 ms

At depth 5,000, TCC scales with live state: ~4 KB → ~16 KB → ~64 KB.

Suite: tcc-engine/bench. Portable WebAssembly engine on the Node SQLite reference host. Healthy-path cost is a separate research-prototype measurement in the technical report, not this series.

TCC Recovery Lab →

Historical prototype

Temporal comparison (research prototype — labeled)

Earlier evaluation against Temporal used a research prototype, not the portable TCC Engine shipped today. Numbers below are prototype results and are retained for the technical report narrative.

Prototype recovery work

load continuation → decode state → hydrate runtime → resume

Recovery latency vs prior durable-boundary depth 1ms 10ms 100ms 1s 10 100 1000 durable-boundary depth
TCC recovery Temporal reconstruction Controlled evaluation · ~4 KB live state

At fixed ~4 KB live continuation state, prototype TCC recovery remained approximately 0.6–0.9 ms as tested history depth increased from 10 to 1,000, while Temporal reconstruction grew with accumulated workflow history in that configuration.

0.6–0.9 ms

Prototype TCC recovery · depth 10 → 1,000 at ~4 KB live state.

50,000

Generated semantic fuzz cases with zero observed failures in the tested suite.

Controlled prototype evaluation. Not a claim about the portable engine or production Cloud.

Prototype benchmarks → Detailed Temporal comparison charts

Technical report

Transparent Continuation Checkpointing

Execution model, recovery semantics, evaluation methodology, healthy-path cost, limitations, and related work.

Read the technical report →

Limitations Scope and non-claims

Start building →