Configuration reference

Everything lives in .dedrift/config.toml, created by dedrift init. Every key below is optional; omitted keys use the documented default.

[project]
name = "my-agent"
canary_repetitions = 7        # N runs per canary per cycle
rolling_window_cycles = 5     # K cycles in the rolling reference

[detection]
fdr_q = 0.05                  # Benjamini–Hochberg level over primary tests
permutations = 500            # resamples for permutation tests (seeded)
seed = 1729                   # global seed, recorded in every report
ph_lambda = 12.0              # Page–Hinkley alarm threshold (robust-scale units)
ph_delta = 0.3                # Page–Hinkley dead-zone

[materiality]
refusal_rate_pp = 2.0         # min refusal-rate shift, percentage points
format_validity_pp = 1.0      # min format-validity shift, pp
rate_default_pp = 2.0         # other rate signatures, pp
ks_distance = 0.15            # min KS statistic D for KS alerts (see note)
scalar_cohen_d = 0.5          # location gate for non-KS scalar channels
variance_ratio = 1.5          # dispersion gate: var must grow/shrink 50%
p95_relative = 0.10           # tail gate: min relative P95 shift
embedding_mmd2_floor = -1.0   # -1 = auto-calibrate; 0 = off; >0 = explicit

[embeddings]
model = ""                    # set via `dedrift embedder pin`, not by hand

Notes that matter

canary_repetitions (N)

The single biggest power lever. The power table shows what each N buys; cost grows linearly.

fdr_q

Applied once per check across all primary tests (both baselines together). Corroboration tests (Anderson–Darling, Welch) never enter the pool and never alert.

ks_distance

The KS channel gates — and reports — the KS statistic D (sup-norm CDF distance), because a shape change with equal means has Cohen's d ≈ 0 and is still real drift. Binding scale, stated plainly: the raw-α=0.05 critical value is ≈ 1.36·√((n+m)/nm), which for equal arms stays above 0.15 until n ≳ 165 per arm — so at typical per-family scale, significance is the stricter filter and this gate exists to stop trivially significant tiny D from alerting at large n.

embedding_mmd2_floor

Left at -1, the MMD² materiality floor is auto-calibrated per (baseline, family) as the 95th percentile of MMD² between pairs of the baseline's own cycles — an empirical null from known-same-distribution data, computed under the same kernel bandwidth as the observed statistic. Needs ≥ 3 reference cycles; below that the floor is 0 and the report says "uncalibratable".

ph_lambda, ph_delta

Page–Hinkley is a diagnostic, not an alert: it localizes drift onsets for attribution. Defaults are calibrated to ≈1.5% per-stream false-flag rate over 30-cycle horizons (measured, enforced < 3% in CI). Flags compound across streams — a stable agent shows ≥1 flag on 56% of checks — which is why flags never page anyone.

Embeddings

dedrift embedder pin hash                    # zero-dependency, always available
dedrift embedder pin st:all-MiniLM-L6-v2    # requires dedrift[embeddings]

Pinning is forever per project: changing the embedder invalidates all semantic history, so dedrift refuses to compare across embedder versions rather than produce quietly meaningless numbers. Pick deliberately, then leave it.

[anytime] — anytime-valid mode

[detection]
inference = "fixed"        # or "anytime"

[anytime]
alpha = 0.05               # lifetime, battery-wide false-alert budget
gamma_total = 0.02         # coverage budget; alpha_prime = alpha - gamma_total
tilts = [1.5, 2.0, 3.0]    # symmetrised to {psi, 1/psi}
epoch_allocation = "per_epoch"   # or "geometric"

Switches the inference layer from per-check FDR to e-processes with a lifetime guarantee. The trade — validity over an unbounded horizon, paid for in detection power — is quantified in anytime-valid mode. Default stays fixed.

Exit codes

Command	Exit	Meaning
dedrift check	0	OK (or NO REFERENCE)
dedrift check	2	DRIFT DETECTED — wire this to alerting