State over time · what fasten records that observability doesn't

Metrics, logs, and traces answer "what is my system doing right now, and what did it do recently?" fasten answers a different question: "what did my system commit to at timestamp T, and can I reconstruct that state exactly?" This is the temporal anchor of the positioning triangle.

Three questions observability cannot answer

Every observability stack answers point-in-time or windowed signal. None reconstruct state.

1. "What did entity acct_042 believe about field risk_tier at 09:14 UTC?" Metrics are aggregate counts; they don't carry the field value at an entity. Logs are lines; they capture writes but not the derived belief. Traces are spans; a trace shows a request, not a stored fact.
2. "Between 09:00 and 10:00, what was the ordered sequence of state commits to entity X?" A trace shows spans within one request. A log query returns matching lines but no guaranteed order across services or across languages. fasten's monotonic_seq is the total order per chain.
3. "Has any row committed before now been altered since it was written?" The single question tamper-evident audit exists to answer. Metrics and logs offer no answer. Traces have no answer. fasten's hash chain answers deterministically.

The four primitives that make state replayable

Every fasten audit row carries four fields whose only job is making state queryable across time.

Together they give the reader a way to ask what state existed at any prior timestamp, and the auditor a way to prove that state was not rewritten.

A worked question

A regulator asks: "On the morning of 2026-08-14, what was your credit-scoring service's belief about applicant acct_042's risk tier when it decided to reject the loan?"

Against a metrics-and-logs stack, the answer is a reconstruction: pull logs for that request_id, hope the belief was logged verbatim, correlate with a metric time series, tell a story. Against fasten, the answer is a query:

The reader returns the belief row that was current at that timestamp, plus its prev_hash so the auditor can verify no row between then and now has been rewritten. The reconstruction is a query, not a narrative.

When you need this

• Regulated environments (BFSI, insurance, healthcare) where a supervisor asks "prove what your system committed to."
• Systems where AI agents write to state that has to be defended later against a claim of manipulation.
• Post-incident audits where "what did we know at time T" must be answered without reliance on log retention or human notes.
• Compliance frameworks requiring provable, non-repudiable audit trails.

When you don't

• Pure real-time monitoring (metrics + alerts are fine · fasten is not an APM).
• Debugging a single request in isolation (a trace is faster).
• Systems with no external accountability requirement · the primitives are free of charge but the discipline costs writing time; skip if the ROI isn't there.