GAL-2 72h Three-Node Live-Operation Baseline with Local Continuity Daemon on Commodity Macs

This record contains the evidence package for a 72-hour, three-node, live-operation baseline of the GAL-2 local daemon layer running on commodity Macs and backed by the live GAL-2 production API as authority. The bundle documents a local daemonized continuity layer exposing application-visible temporal state through `/health` and `/now`, including fields such as `daemon_mode`, `holdover_active`, `cache_age_sec`, `slew_active`, `slew_remaining_sec`, `source`, `gal2_time`, and `utc_time`. This run was conducted under real network conditions, against the live production GAL-2 API, and on real commodity endpoint machines rather than in a closed simulator or synthetic timing harness. The local daemon was exercised as an application-facing continuity layer in a fully live API-backed configuration, with normal upstream refresh, natural network variability, and one naturally observed short upstream interruption handled through holdover and clean rejoin semantics. Architecturally, GAL-2 is intended to function as a plug-and-play local continuity layer above an authority source, exposing application-consumable temporal state without requiring applications to interact directly with upstream timing complexity. Node roles:- mac2025 = controller- mac2015 = node A- mac2016 = node B Run characteristics:- target duration: 72 hours- logger interval: 5 seconds- local daemon bind: 127.0.0.1:9097- final CSV structure: 1 header row + 51,840 data rows per node Primary purpose of this record:- document sustained three-node live operation over 72 hours- document monotonic served time per node- document long-duration local daemon continuity- document application-visible categorical temporal state- document a naturally observed short upstream interruption handled through holdover and clean rejoin semantics Main demonstrated results in this bundle:- 0 monotonicity violations in `gal2_time` across all three final node CSVs- continuous logging over 72 hours on all three nodes- stable daemon operation with mode/state exposure- one natural short upstream interruption observed and handled without backward step- evidence of low-millisecond inter-node harmonization when analyzed on the appropriate time axis Important scope note:This record is not presented as an intentionally injected outage experiment, nor as a fine-grained precision-metrology result. It is a continuity and live-operation baseline. This record does not, by itself, claim:- elite precision superiority over PTP or GNSS-class systems- formal fine-grained inter-node coherence bounds- a finalized quantitative uncertainty contract- heterogeneous-source validation across PTP, NTP, and GNSS direct inputs Methodological note:Host-local clocks on the three Macs were not mutually disciplined. Therefore, apparent large inter-node divergence can arise if rows are aligned by host-local `local_utc`. When analyzed on the common `utc_time` axis returned by the daemon/API path, inter-node `gal2_time` behavior is in the low-millisecond range rather than seconds. This record is therefore strongest as evidence of per-node continuity and multi-node live operation, while formal inter-node coherence remains a separate future experiment. Contents include:- final 72h CSVs for all three nodes- run-start note- initial checkpoints- final snapshots- analysis summary- continuity contract draft- failure model draft- daemon and logger source snapshot- minimal example consumer- supplementary pre-run smoke artifacts- integrity files (`MANIFEST.txt`, `SHA256SUMS.txt`) This is a new standalone record focused on multi-node continuity and live-operation semantics, rather than a revision of earlier single-run stability framing. Final bundle SHA-256:ffd7947da57d1c385d3bdf4b83ab329758af8079a652b7190c18543f466c226a