ORACLE BOOTSTRAP — the hardware question, answered honestly
============================================================
Question: can two PHYSICAL nodes share the oracle without transmitting it?
This is the one empirical piece under mutual reconstruction. Measured, not assumed.

MEASUREMENT (tsc_corr.c, real RDTSC):
  GetTSC HIGH bits = SHARED (common clock): 64/64 samples match on >>20.
  GetTSC LOW bits  = PRIVATE (sampling jitter): differ by ~500-3500 ticks.
  => what's shared is the CLOCK; what's private is the JITTER. (Inverts the
     naive assumption that co-sampled GetTSC is wholesale shared.)

BOOTSTRAP: drop the private jitter bits, keep the shared clock epoch, derive the
same oracle seed on both nodes independently.

CROSS-PROCESS REALITY (bootstrap_proc2.c, separate scheduled processes):
  drop=18 (~87µs epoch):  0/30 agree   <- in-process success was misleading
  drop=20 (~350µs):       9/30
  drop=22 (~1.4ms):      25/30
  drop=24 (~5.6ms):      28/30
  Scheduling gap between separate processes >> in-process jitter. Raw seed-
  guessing across real processes FAILS at fine granularity.

HONEST VERDICT — TWO REGIMES:
  REGIME 1 (co-located, shared clock domain): nothing-transmitted mutual comms
    is REAL. Same board / clock line / hardware trigger → epochs align <1 bin.
    Verified exact in-process (mutual comms err 1.8e-15). This is the literal
    "shared world": two nodes in one entropy field.
  REGIME 2 (remote nodes): needs a PUBLIC epoch-frame sync (a nonce — reveals
    the clock frame, not the fine oracle). Shared secret = band between epoch
    granularity and jitter (~2 bits/sample at measured jitter). LOW-LEAKAGE,
    not zero-transmission. Stated as such — not called "nothing transmitted."

WHAT THIS MEANS:
  The exact-math results upstream (law inversion, mutual cancellation) are
  unconditional. The NOTHING-TRANSMITTED property is conditional on a shared
  clock domain — real for co-located nodes, requiring a public frame for remote.
  The project's honesty rule applied to hardware: don't claim capability the
  silicon doesn't give.

NEXT REAL STEP: measure GetTSC epoch alignment between two nodes sharing a
physical reference (GPIO trigger, or disciplined to the same GPS/PTP clock).
That is the bench test that would move Regime 1 from same-machine to same-lab.

Files: oracle_bootstrap.c (primitive), tsc_corr.c (measurement).
hdgl.hdgl: 6 glyphs, 0 comments, bootstrap + both regimes in STEP.
