EpochSketch · tag:16 | time:24 | count:24 — v2 spec

Decay is interpreted, never performed. Eviction follows the selected schedule — M3 fixed-point (primary, Doc 03.2), M1 backup, deprecated M2, or Option C's RNG reference — with the Doc 03.3 finalized draw. Watch every attempt in the Chip Inspector.
Global epoch
100
Last touched slot — raw 64-bit word —
tag (16)timestamp (24)count (24)

Incoming traffic

The table — 4 buckets × 8 slots (bucket = one 64-byte cache line)

increment fold / claim (first sighting) chip attempt 🌱 newborn on audition

Sampling decision (last record)

— send something —
f̂ = effective count
p_keep = min(1, C / f̂), C = 8
first sighting ⇒ forced keep

Operation log

    Chip inspector — every eviction attempt, dissected

    No lockouts yet. Fill a bucket (spam churn keys) to see chipping.
    Reading a row: victim = weakest slot by effective count (deterministic — never a lottery) · k = zero-bits demanded by the selected schedule (M3: (eff×227)»11 · M1: eff»3 · M2: bitlen — deprecated per Doc 03.2 §6, kept for teaching) · draw = finalize(spare ⊕ victimWord ⊕ salt×ODD) per Doc 03.3 — low 8 bits shown, the underlined k bits must ALL be 0 · Option C rows instead show p = 1.08^−eff and the raw RNG draw · BOUNCE = nothing changed · CHIP = victim −1, keeps slot · TAKEOVER = challenger claims slot at count 1, 🌱 audition begins (root at eff ≥ 8).
    Suggested experiment: ① Spam /healthz ×50 — watch p_keep collapse. ② Tick epoch twice — slots don't change, effective counts halve (lazy decay). ③ Send /healthz ×1 — amber fold + forced keep (diversity, free). ④ Now the new part: spam churn keys until a bucket fills, then keep going and read the Chip Inspector: victims nominated deterministically, k widening as victims strengthen, draws bouncing off strong slots and grinding down weak ones. ⑤ When a TAKEOVER lands, watch the 🌱 newborn: keep feeding the same burst and it takes root; abandon it and challengers chip it back out. The traffic performs the audition. ⑥ Now flip the eviction schedule — rerun the same spam under M3, M1, deprecated M2, and Option C's RNG, and compare the Chip Inspector's k/p values and outcomes: you are hand-running a miniature of the D2 experiment.