BLACKICE — REPRODUCTION AND SOURCE SCOPE
Checked: 15 October 2026

The browser World and memory engines were copied unchanged from the supplied
PhantomBridge / BlackIce site/src directory. The portfolio adds a new interface,
controlled replay experiment, and discovery-only memory demonstration.

Run from a checkout of the website with Node.js installed:
  node scripts/check-blackice-parity.mjs

Alternatively, download all files listed in validation.json into one directory,
including package.json, then run there:
  node parity.test.js
  node parity2.test.js
  node mem_parity.test.js

Observed results:
  World suite 1: 73 passed, 0 failed.
  World suite 2: 64 passed, 0 failed.
  Memory/session suite: 67 passed, 0 failed.

These compare against saved project reference captures. They do not regenerate
the Python reference results. The memory suite includes numeric tolerances;
passing is not a claim of universal bitwise equality across languages.

THE LIVE REPLAY EXPERIMENT
Two fresh worlds have the same code, identity, response templates and service
observations. Both execute whoami, read a note, write a result and read it back.
The control changes only the initial note in World B.

The companion records the initial state, command results and final state. Its
serializer sorts object keys and represents regular expressions by source and
flags. Own enumerable world data is included; methods/prototypes are fixed code.
The complete strings are compared directly. SHA-256 hashes are references, not
the equality test. This is not the Python snapshot interchange format.

The original World.digest() is a 32-bit checksum over selected fields. It is not
used to establish equality in the new companion. Repeating a finite script is
an experimental check, not a formal proof for all possible executions.

THE MEMORY EXPERIMENT
The three-command sequence uses a synthetic address (192.0.2.10), a simulated
service listing and authored share/account responses. Nothing is sent to that
address. Commands are recorded in a fresh Session. The browser knowledge slice
contains eleven techniques, not the full external technique catalogs.

Combined score = 0.45 * cosine + 0.25 * path + 0.30 * random-walk score.
These weights and graph rules are implementation choices. Neither a rank score
nor a seeded observation confidence is a calibrated success probability.

SCOPE
The simulated command engine does not execute a host shell or contact network
targets. This companion does not offer the Python service, an arbitrary transcript
ingester, or the Python optional model fallback. No live-system fidelity,
training-outcome or held-out suggestion-quality result is claimed.

PROVENANCE
The recovery record locates the original project under IBM's PhantomBridge
work. It distinguishes explicit historical facts (H1), historical behavior (H2),
reconstruction (R1), and unresolved details (U). The current engine is a
reconstruction. Validation of current code is not recovery of original code.

The original source landing page remains unchanged in the author's source folder.
The portfolio companion replaces its marketing claims with this scoped account.