Co-authored-by: cce <51567+cce@users.noreply.github.com>

verify_deployment.py:276 printed a U+2194 inside the DRIFT branch. On a
cp1252 console that raises UnicodeEncodeError *after* the comparison has
run and correctly found the mismatch, so the process dies inside its own
failure report and exits 2 ("could not complete the check") instead of 1
("DRIFT").

The exit codes are a contract, stated in that file's own docstring: 0
match, 1 drift, 2 could not check. A reviewer who sees 2 concludes the
tooling failed, not that the deployment diverged. So this downgraded a
real finding to an inconclusive one -- silently, and only on the platform
the maintainer actually works on.

It survived this long because the line only executes when something is
wrong. No green run could ever reach it. That is exactly the shape of
defect this repo keeps finding: a path that is never exercised until it
matters.

Adds sdk/tests/test_reports_print_on_a_windows_console.py, which scans the
scripts a reviewer is told to run for characters cp1252 cannot encode on
print lines. Em dash and ellipsis are deliberately allowed -- they ARE in
cp1252 and are used throughout these reports; flagging them would make the
guard unusable and it would be deleted.

The guard was mutation-tested rather than trusted: reintroducing the U+2194
makes it fail, and it does not fire on plain ASCII or on em dash/ellipsis.
A scan that cannot fail would have been worth nothing here.

Suite: 145 passed, 5 skipped. verify_deployment.py still prints MATCH.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Close the TestNet contract drift: deploy app 770964251 from the committed artifact

See https://github.com/algorandfoundation/algokit-utils-ts/pull/612

README tells reviewers to run the hermetic verification, and running it
unpacks the pinned Falcon tarball to falcon-src/ and builds in place,
leaving codec.lib and libfalcondet1024.pdb behind. .gitignore covered
build/ but not these, so following the documented path left three
untracked files and a dirty `git status`.

Found by actually running the advertised command rather than assuming it
was clean. Full run: 4 axes PASS, exit 0 -- including Axis C re-signing
the committed goldens byte-identically from a ziglang build, which is a
different compiler from the MinGW gcc used elsewhere in this session.
Cross-compiler byte-identity is what FALCON_FPEMU=1 exists to guarantee,
now demonstrated rather than asserted.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

Co-authored-by: cce <51567+cce@users.noreply.github.com>

This branch diverged 6 ahead / 5 behind. main had built an apparatus around
the drift being OPEN -- a keyed redeploy workflow, a fail-closed secret gate,
a drift banner test, BLOCKERS.md -- while this branch closed the drift by
deploying app 770964251. Pushing without merging would have reverted all of
it.

Kept both approaches; they are complementary, not competing:
  * main's contracts/require_testnet_deploy_secret.py + testnet-redeploy.yml
    for a keyed CI deploy;
  * this branch's scripts/deploy_testnet_secure.py and _fresh.py for a human
    at a terminal, needing no repository secret.

Conflicts (5) resolved:
  * .github/workflows/ci.yml -> took main. It deliberately split the live
    bytecode match out of the required gates because that check needs live
    algod. That decision stands; the check now passes, but a network-dependent
    job still should not block merges.
  * README / REVIEWER / ROADMAP / AUDIT_READINESS -> took this branch. main's
    side described the drift as open, which is no longer true.

The partition guard did its job: the merge introduced six files mentioning an
app id that nobody had classified, and it failed until each was decided.

  FROZEN   sdk/tests/test_testnet_drift_banner.py  (synthetic fixture; kept as
           a regression guard so a future mismatch cannot degrade to a skip)
           BLOCKERS.md  (must keep quoting the measured 660 B / d24d9071 and
           709 B / 6fa5cee1 -- the banner test asserts they are still there)
           LOCALNET_VALIDATION_2026-06-01.md  (a dated record)
  LIVE     CONTRIBUTING.md, testnet-followup.yml, testnet-redeploy.yml

Prose again, not just constants: ci.yml, CONTRIBUTING.md and both workflows
still said "until app 763809096 is replaced" and "predates the committed
contract". It has been replaced. Corrected to say so, with the three-month
red period kept on the record rather than tidied away.

BLOCKERS.md is now CLOSED with the measured drift preserved verbatim below a
historical-record heading.

Verified: MATCH on app 770964251; verify_trelyan.py 18 passed, 0 failed;
suite 142 passed, 5 skipped; partition guard 30 passed.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

See https://github.com/algorandfoundation/algokit-utils-ts/pull/612