1.2.50

• Push alerts are sealed with post-quantum encryption wherever the daemon
  can. Every alert is end-to-end encrypted, so the relay that forwards it
  never sees plaintext, but the x25519 sealing behind that is the kind a
  future quantum computer could break, and traffic recorded today could be
  decrypted then. A device paired under the xwing suite closes that
  window: its alerts are sealed with X-Wing (ML-KEM-768 + X25519), a
  hybrid of post-quantum and classical key exchange, through HPKE in base
  mode (HKDF-SHA256, AES-256-GCM). The push extra carries the
  cryptography library that seals it on every platform with a wheel, so
  post-quantum sealing needs no further install step. GET /whoami names
  the suites a daemon can seal to as sealableSuites; the app pairs under
  xwing when it is listed, moves an existing x25519 pairing up on its
  own once a daemon starts advertising xwing, and moves an xwing
  pairing back down if its daemon stops sealing it. The post-quantum key
  material costs 1136 bytes of the fixed notification budget, so an
  xwing alert carries a shorter log tail. A daemon without the library
  refuses xwing pairings and says so at startup, x25519 pairings are
  unchanged, and the wire construction is normative in
  docs/relay-protocol.md.
• Post-quantum sealing ships in every release binary and container image
  whose platform can compile it. Where PyPI publishes no cryptography
  wheel, CI builds the library from source against an OpenSSL 3.5 it
  installs or builds: for the Linux builds without a wheel (including the
  glibc linux-armv6 for the Raspberry Pi 1 and Zero), for
  windows-arm64, for the FreeBSD, OpenBSD, NetBSD, and illumos builds,
  and for every image platform. A build that fails to compile the library
  ships without the suite, and each one that succeeds proves itself with a
  real X-Wing seal before it is frozen in. Intel
  macOS and 32-bit Windows carry cryptography 48.x, the last line with a
  wheel for them. The linux-ppc64le binary needs glibc 2.28 rather than
  2.17, the floor of the only ppc64le wheel. The illumos-amd64 binary
  also carries the push reporter itself, with libsodium from the OmniOS
  extra publisher. linux-mips64le and linux-armel seal x25519 only,
  because no Rust toolchain exists for their architectures.
• A daemon event alert whose subject or message is empty leaves that field
  out of the sealed plaintext instead of sending it as null, which is what
  the relay protocol's field contract specifies.
• The reaper learns of a finished job through a done callback on that job's
  wait task, and filing the completion is one list append however many jobs
  are running. Draining 500 completions one at a time with 500 jobs running
  measures 6 ms locally against 50 ms.
• A bare cronstable --version prints the version before it builds the
  argument parser and its subparsers. The command measures 54 ms locally
  against 64 ms.
• The SLA pass walks a memoized list of the jobs that carry an sla block
  and reads the latch map to find a latch left on a job whose block a
  reload removed. One pass over 2,000 jobs with 20 SLA blocks measures
  0.02 ms locally against 0.15 ms; over 10,000 jobs with 100 blocks,
  0.09 ms against 0.79 ms.
• The Prometheus scrape carries a rendered block on each shared per-job
  label set, renders small integral values from a table filled on the
  first scrape, builds each sample line as one string, and keeps its
  sorted job order between scrapes. Two renders of a 500-job exposition
  with run counters measure 8.7 ms locally against 13 ms.
• The calendar feed folds each line once as it builds it, describes each
  distinct schedule without an H item once, and stops a capped entry
  before converting the fire it drops. A 500-entry, seven-day feed
  measures 25 ms locally against 43 ms.
• The configuration parser forks a strictyaml position pointer with one
  list copy, passes a value holding no $ through the interpolation walk
  without a call, and hands every job with a clean schedule the same empty
  findings list. A 300-job parse measures 96 ms locally against 108 ms, an
  interpolation walk over 2,000 references 41 ms against 49 ms, and a full
  garbage collection with 100,000 jobs resident 34 ms against 99 ms.
• The filesystem state store writes each atomic file through its raw
  descriptor, opened in binary mode on every platform, lists a records or
  documents root with scandir so the directory test reads each entry's
  own type, sorts each stream listing once in the order the caller reads it,
  and takes its try-lock through the one lock routine's non-blocking lane.
  Twenty repeat listings of an unchanged 2,000-record stream measure 83 ms
  locally against 103 ms, and a garbage-collection sweep over 2,000 unkept
  streams 103 ms against 169 ms.
• The dashboard's schedule engine jumps a restricted hour, minute, or
  second field straight to its next listed value, keeps one collator for
  every name sort, memoizes each schedule's prose, counts the running,
  failing, ok, and paused jobs in one pass, and checks a fleet poll against
  the response text it parsed. The log drawer keeps its search pattern
  compiled across streamed lines, caches each line's stripped and lowercased
  text, writes the match count only when it changes, and shows bad regex
  in its place for a pattern that doesn't compile. Counting the matches in
  a 5,000-line buffer measures 0.04 ms locally against 0.28 ms, and a cold
  week walk over 500 schedules 1.75 ms against 4.2 ms.
• The terminal dashboard caches short SGR sequences and their bounded
  rewrites per theme. Discarded control sequences and oversized rewrites
  stay out of the cache. The log drawer, the DAG log tab, and the tail view
  bind their style helpers and stream markers once per frame. Restyling a
  5,000-line drawer measures 17 ms locally against 25 ms, and a drawer
  scroll walk with steady repaints 67 ms against 101 ms.
• Config load rejects a job whose name matches a DAG's schedule job
  (dag:<dag name>), including collisions across files and includes.
  Other dag: job names and classic crontabs named dag are valid.
• A failed cluster peer poll removes the peer from agreement, including
  failures while parsing telemetry. Invalid telemetry fields are discarded
  while the peer's membership observation is processed.

Performance vs 1.2.49

Gate: passed. Every metric stayed inside its regression limit.

Both versions ran interleaved on one runner. Time metrics compare the best round of each and memory metrics the median. A negative change means faster or smaller.

A regression gates only when it exceeds both its declared limit and 2 noise bands. The +- column is that band: the round-to-round scatter of the two sides, combined in quadrature.

All benchmark results (97 metrics)

Benchmark	1.2.49	1.2.50	Change	Noise +-	Gate (eff.)
loop.stall_completions_500	110.26 ms	14.92 ms	-86.5%	1.0%	25%
webui.log_count_5k	660.0 us	90.0 us	-86.4%	4.4%	303% (declared 25%)
webui.week_walk_500	7.65 ms	3.50 ms	-54.2%	1.2%	25%
ical.render_500x7d	93.49 ms	51.46 ms	-45.0%	2.5%	15%
prometheus.render_500	30.04 ms	17.74 ms	-40.9%	2.1%	33.29% (declared 15%)
tui.drawer_paint_5k	223.63 ms	137.50 ms	-38.5%	0.4%	15%
tui.log_restyle_5k	53.01 ms	32.64 ms	-38.4%	1.3%	18.87% (declared 15%)
mem.gc_pause_100k	177.71 ms	126.70 ms	-28.7%	4.0%	25%
startup.version	68.91 ms	54.48 ms	-28.0%	0.2%	25%
state.gc_sweep_2k_streams	59.98 ms	44.64 ms	-25.6%	1.7%	25%
config.parse_yaml_300	269.36 ms	202.56 ms	-24.8%	0.9%	15%
config.parse_yaml_3k	2.677 s	2.026 s	-24.3%	1.0%	15%
config.reload_warm_50	61.15 ms	47.44 ms	-22.4%	0.8%	25%
config.reload_warm_include_50	129.67 ms	101.46 ms	-21.8%	0.5%	25%
cluster.parse_summaries_6k	110.28 ms	86.99 ms	-21.1%	0.6%	15%
schedule.reseed_local_20k	118.99 ms	94.86 ms	-20.3%	1.1%	15%
cronexpr.test_match_200k	33.60 ms	26.94 ms	-19.8%	2.2%	29.76% (declared 15%)
config.interp_2k	97.18 ms	78.35 ms	-19.4%	0.4%	15%
webui.radar_walk_500	470.0 us	390.0 us	-17.0%	0.0%	425.5% (declared 25%)
schedule.due_pass_100k	18.23 ms	15.44 ms	-15.3%	1.8%	54.85% (declared 25%)
config.reload_gc_100k	2.849 s	2.417 s	-15.2%	1.2%	25%
startup.validate_config_100	256.12 ms	223.60 ms	-13.6%	0.2%	25%
mem.jobconfig_2k	1.70 MB	1.49 MB	-12.6%	0.1%	29.42% (declared 15%)
startup.job_set_id_100	259.99 ms	229.53 ms	-12.6%	0.4%	25%
cronexpr.occurrences_1k	16.18 ms	14.19 ms	-12.3%	2.1%	61.8% (declared 15%)
schedule.suggest_slot_5k	4.90 ms	4.31 ms	-12.2%	6.2%	203.9% (declared 15%)
state.list_records_warm	144.04 ms	126.70 ms	-12.0%	0.5%	15%
cronexpr.next_simple	43.81 ms	38.72 ms	-11.6%	1.9%	22.83% (declared 15%)
schedule.reseed_100k	232.53 ms	205.71 ms	-11.5%	1.9%	15%
json.roundtrip_3k	658.32 ms	588.55 ms	-10.6%	0.9%	15%
mem.rss_version	16.22 MB	14.62 MB	-9.9%	0.5%	25%
schedule.next_fires_2k	29.82 ms	27.00 ms	-9.5%	1.2%	33.53% (declared 15%)
json.roundtrip_orjson_3k	235.90 ms	213.96 ms	-9.3%	0.5%	15%
startup.import_config	104.61 ms	97.11 ms	-8.6%	0.4%	25%
state.fanout_gather_100	27.15 ms	24.81 ms	-8.6%	4.9%	36.83% (declared 25%)
mcp.handle_200	144.32 ms	132.40 ms	-8.3%	0.5%	15%
schedule.duplicates_20k	19.32 ms	17.74 ms	-8.2%	5.2%	51.75% (declared 15%)
cluster.job_owner_2k	65.18 ms	60.21 ms	-7.6%	3.1%	15%
schedule.pressure_20k_48h	18.21 ms	16.88 ms	-7.3%	5.9%	54.93% (declared 15%)
cronexpr.next_complex	17.28 ms	16.05 ms	-7.1%	1.1%	57.87% (declared 15%)
schedule.cold_build_100k	543.51 ms	505.88 ms	-6.9%	0.8%	15%
dag.plan_claim_10k	17.92 ms	16.70 ms	-6.8%	3.6%	55.81% (declared 15%)
loop.stall_metrics_2000	29.36 ms	27.55 ms	-6.2%	1.8%	info
state.mutate_document_1k	227.91 ms	215.20 ms	-5.6%	0.5%	25%
cronexpr.next_dst_2k	36.70 ms	34.71 ms	-5.4%	1.9%	27.24% (declared 15%)
state.artifact_list_churn	1.70 ms	1.61 ms	-5.3%	4.6%	293.7% (declared 25%)
schedule.lint_250_zoned	7.94 ms	7.54 ms	-5.0%	1.0%	126% (declared 15%)
push.seal_500	99.47 ms	94.65 ms	-4.8%	0.4%	15%
webui.render_fleet_15x400	25.21 ms	23.99 ms	-4.8%	0.8%	25%
state.artifact_get_newest	24.34 ms	23.26 ms	-4.5%	1.0%	25%
state.derive_max_warm	285.97 ms	273.33 ms	-4.4%	0.4%	15%
config.jobconfig_3k	70.34 ms	67.28 ms	-4.4%	2.3%	15%
dag.finish_fanin_1k	4.25 ms	4.07 ms	-4.2%	1.2%	117.6% (declared 25%)
webui.render_rows_500	17.63 ms	16.93 ms	-4.0%	0.5%	25%
state.list_records_2k	30.31 ms	29.18 ms	-3.7%	1.6%	32.99% (declared 15%)
dag.mapped_drain_256	32.30 ms	31.12 ms	-3.6%	2.3%	30.96% (declared 25%)
state.derive_max_cold	27.58 ms	26.71 ms	-3.1%	2.1%	36.26% (declared 15%)
redact.clean_20k	21.16 ms	20.53 ms	-3.0%	1.8%	47.25% (declared 15%)
startup.import_cronexpr	28.42 ms	28.77 ms	+2.8%	0.5%	25%
config.parse_crontab_1k	12.21 ms	11.89 ms	-2.6%	1.4%	81.9% (declared 15%)
state.list_documents_600	71.06 ms	69.22 ms	-2.6%	0.8%	16.89% (declared 15%)
loop.idle_wake_rate	21.90 ms	21.35 ms	-2.5%	1.5%	25%
state.depends_on_past_gate	3.60 ms	3.51 ms	-2.5%	2.9%	55.59% (declared 25%)
state.boot_rehydrate_populated	49.63 ms	48.45 ms	-2.4%	1.0%	20.15% (declared 15%)
dag.adopt_scan_500	108.77 ms	106.24 ms	-2.3%	1.3%	25%
dag.list_dags_warm	1.09 ms	1.11 ms	+2.3%	2.3%	183.7% (declared 25%)
loop.stall_jobs_500	21.54 ms	21.11 ms	-2.0%	0.6%	info
webui.render_term_5k	130.35 ms	127.80 ms	-2.0%	1.1%	25%
state.lease_renew_200	134.15 ms	131.54 ms	-1.9%	1.8%	37.27% (declared 25%)
state.kv_roundtrip_200	184.99 ms	181.63 ms	-1.8%	1.5%	15%
dag.build_chain_10k	29.40 ms	28.95 ms	-1.5%	2.6%	34.01% (declared 15%)
job.stream_capture_120k	45.70 ms	45.09 ms	-1.3%	1.7%	21.88% (declared 15%)
startup.import_daemon	176.76 ms	178.65 ms	+1.2%	0.8%	25%
dag.build_layered_10k	49.79 ms	50.35 ms	+1.1%	3.1%	20.09% (declared 15%)
cronexpr.parse_complex	24.25 ms	24.51 ms	+1.1%	1.3%	41.23% (declared 15%)
webapi.sse_burst_20k	58.14 ms	58.74 ms	+1.0%	1.2%	15%
redact.adversarial_10k	37.63 ms	37.32 ms	-0.8%	0.8%	26.57% (declared 15%)
redact.secrets_5k	21.02 ms	21.14 ms	+0.6%	1.0%	47.57% (declared 15%)
job.report_noop_100k	48.81 ms	49.07 ms	+0.5%	1.9%	15%
dag.list_runs_warm	12.91 ms	12.84 ms	-0.5%	1.6%	25%
cronexpr.parse_simple	44.93 ms	45.15 ms	+0.5%	1.3%	22.26% (declared 15%)
state.append_1k	724.81 ms	721.67 ms	-0.4%	2.3%	15%
webapi.auth_scope_20k	93.66 ms	93.29 ms	-0.4%	0.5%	15%
cluster.fleet_view_15x400	90.85 ms	90.50 ms	-0.4%	0.5%	25%
dag.advance_quiescent_1k	73.63 ms	73.35 ms	-0.4%	0.7%	25%
startup.python_baseline	17.32 ms	17.36 ms	+0.2%	0.4%	info
webapi.jobs_payload_500	59.04 ms	58.92 ms	-0.2%	0.9%	16.94% (declared 15%)
dag.advance_quiescent_chain	72.65 ms	72.51 ms	-0.2%	1.0%	25%
mem.rss_daemon_import	36.12 MB	36.06 MB	-0.2%	0.2%	25%
resources.monitor_stop_100	288.04 ms	288.44 ms	+0.1%	1.9%	25%
fingerprint.job_set_id_10k	297.42 ms	297.05 ms	-0.1%	0.5%	15%
tui.log_search_20k	16.34 ms	16.33 ms	-0.1%	0.9%	61.18% (declared 15%)
statsd.emit_2k	14.74 ms	14.74 ms	+0.0%	0.9%	67.84% (declared 25%)
webui.append_line_5k	15.0 us	15.0 us	+0.0%	4.4%	25%
webapi.jobs_bytes_500	181.42 KB	181.42 KB	+0.0%	0.0%	15%
webapi.jobs_gzip_500	5.19 KB	5.19 KB	+0.0%	0.0%	15%
mem.crontab_10k	2.80 MB	2.80 MB	+0.0%	0.0%	17.85% (declared 15%)

What's Changed

• Bump the github-actions group with 6 updates by @dependabot[bot] in #14

Full Changelog: 1.2.49...1.2.50