dedrift 0.1.0

Statistically rigorous behavioral drift detection for AI agents. Agents don't throw errors when they degrade — dedrift catches it.

dedrift

Agents don't throw errors when they degrade. They keep confidently producing worse outputs. dedrift catches it.

dedrift is an open-source Python package that detects silent behavioral drift in AI agents. It logs agent interactions, runs a frozen canary suite repeatedly, extracts behavioral signatures, and applies statistically rigorous drift detection with config-change attribution ("behavior shifted within 6h of the model version change").

Why dedrift

Model updates, prompt edits, tool-schema changes, RAG index refreshes, and provider-side silent updates all shift agent behavior without throwing a single error. Existing monitoring catches crashes, not character changes.

dedrift's differentiation is statistical correctness:

• Every detector controls its documented false-alarm rate — validated by simulation tests in CI.
• All alerting passes through FDR control (Benjamini–Hochberg). No raw per-test p-values dressed up as alerts.
• Every alert requires both statistical significance and a configurable effect-size (materiality) threshold. Fewer, higher-confidence alerts.
• LLM outputs are stochastic: canaries run N times per cycle and we compare distributions, never single outputs.
• Dual baselines: every check runs against a rolling recent window (sudden breaks) and a frozen golden baseline (boiling-frog drift).
• Honest about power: small N means low detection power, and the docs show you the math instead of hiding it.

Status

Pre-alpha, under active development. Working today: logging schema + store, canary runner (N repetitions per cycle), Tier-1 structural signatures, Tier-2 semantic signatures (pinned embedder, semantic displacement, MMD-RBF with a seeded permutation null and an auto-calibrated materiality floor), the full detector battery (KS/AD/Welch/Levene/bootstrap-P95/two-proportion z, PSI, Page–Hinkley) with BH-FDR and materiality gating, dual baselines, config-change attribution, and deterministic markdown reports — all with calibration and power tests enforced in CI.

Install

pip install dedrift              # core: zero ML dependencies
pip install "dedrift[embeddings]"  # + semantic signatures (sentence-transformers)
pip install "dedrift[judge]"       # + LLM-judge tier

For development: pip install -e ".[dev]".

Quickstart — a full simulated drift incident in five commands

No API keys needed: dedrift sim ships a seeded synthetic agent whose "model version" is swapped mid-history, shifting output length, refusal rate, and format validity — the classic silent degradation.

pip install dedrift
mkdir drift-demo && cd drift-demo

dedrift init                                   # create the project
dedrift embedder pin hash                      # optional: enable Tier-2 semantic signatures
dedrift sim --cycles 8 --change-cycle 7        # 8 canary cycles; model swap at cycle 7
dedrift baseline set cycle-0000 cycle-0001 cycle-0002   # freeze known-good cycles
dedrift check                                  # exits 2: DRIFT DETECTED (both baselines)
dedrift report --out report.md                 # deterministic markdown report

The report shows what shifted in plain units (e.g. refusal +21 pp, output variance ratio ~9x), BH-adjusted p-values, and attribution: "nearest config event: model fingerprint change, 0.0 h before onset." With your own agent, replace sim with dedrift canary run --suite canaries.yaml --agent yourmodule:agent_fn --model 'provider/model@version' on a schedule.

Detection power: the honest table

Statistical power depends on sample size, and canary suites are small. For a rate signature (e.g. refusal) at a 5% baseline in a family of 30 canaries, two-sided α=0.05 per test (before FDR, which reduces power further), simulated power to detect a shift of the given size:

Repetitions N	n per window	+2 pp	+5 pp	+10 pp	+15 pp
5	150	0.07	0.28	0.78	0.97
7 (default)	210	0.11	0.42	0.91	1.00
10	300	0.13	0.60	0.99	1.00

Read the first column honestly: a 2 pp refusal shift is essentially undetectable at this scale. dedrift's default materiality gate (2 pp) is a floor on what may alert, not a promise of what will be detected. If small rate shifts matter to you, grow the refusal-boundary family or raise N — power comes from samples, not from wishful thresholds.

dedrift Pro

Anytime-valid sequential inference (e-processes), conditional production-traffic drift, and importance weighting are part of a separate commercial tier and are not in this repository.

Contact

Questions, bug reports, or interest in being a design partner: open an issue or email [e-mail].

License

AGPL-3.0-only. See LICENSE.