# Binary live study: interruption reconciliation and one recovery allocation

Registered before recovery outcomes on 9 September 2026. The unchanged original
24-program live protocol remains the method. Its first ledger has starts at
slots 0, 1 and 2 and qualified captures at slots 0 and 1. Slot 2 has persisted
repair construction but no trajectory, success marker, completed transaction or
execution-ledger entry. No matching process remains. The interruption's cause
is unknown; it is not evidence of a physical task success or failure.

Preserve every original file and bind its hash. Re-score the two qualified
captures and independently recheck their saved references, own histories,
transactions and physical arrays before reusing them. This invokes saved-output
verification only, with no optimizer, decoder, forecast or physics replay.

Allocate one recovery attempt to original slot 2, followed by the 21 unstarted
slots 3 through 23 in original index order. Use a separate output directory and
receipt. Do not append to, complete or repair the original ledgers. There is no
outcome-based retry and no second recovery in this allocation. A physical task
failure continues; an integrity failure halts subsequent launches and retains
all remaining planned slots explicitly. If all captures qualify, coverage is
24 programs with 2 reused and 22 new captures; there are 25 attempt starts
including the preserved interrupted attempt.

Report two views: (1) original first attempts, with slot 2 unavailable and all
24 allocated programs retained, and (2) the coverage-completed panel using its
single declared recovery. Never label the second view 24 successful original
first attempts or untouched confirmation. Count successes, available scores,
physical terminations, STOP coverage, source failures and missing records
separately. Keep baseline, replay and live comparisons paired by program.

Method, checkpoint, source files, runtime control hash map, exact causal input
references, generation seed 1703, simulation seed 1701, decision ticks, 99/101
forecast horizons, two nominal model calls and original task thresholds are
unchanged. Pin the controller package to the registered execution-model-history
runtime before importing legacy orchestration helpers, which alter sys.path.
This is an orchestration import qualification, not a runtime-source change.
The physical backend remains paused during preparation. Run one own simulator
at a time and leave other users' processes untouched. Final36 remains sealed.

A new source receipt binds this protocol, the new orchestration/audit/tests,
the passing preflight, the reconciliation and original partial files, in
addition to the unchanged original source closure. Frozen worker requests use
only the original control-source map; external comparator and recovery audit
files never become worker inputs. The root reviews the narrow runner diff and
fake allocation tests before freezing. No new peer-agent work is required for
this orchestration-only recovery; the original worker review is unchanged.