Every project my agent pipeline works on is written up as a climb: the problem it started from, a measurable summit, the rules it may not break, and every merge on the way, each tied to the roadmap line it served or marked as a detour. This is the notebook, one page per project.
4 climbs over the last 30 days, 958 merged steps: 47% joined to the route by an exact key, 1% by inference, 52% off the written route: 35% owner merges, 1% exhaust and tech-debt rows, 3% canaries, 12% plan-less rows. 116 weather episodes and 0 false summits (reverted merges) in the window. 2 more climbs run on repositories that are not published; they are counted here and never listed.
Agent swarms produce activity, not delivery: work that nobody scoped, verified or merged. loom-core's Mills operator turns roadmap intent into scoped items, proves every change in a merge train, and keeps cost, failures and operator controls visible, so the factory can be trusted to run unattended on the lab's top projects.
Auto-merge rate (30d)
95.1%
target ≥ 95%
+41.9 pts since Oct 22, toward target
Enqueue-to-merge p50 (press entries)
0.2 hours
target ≤ 1 hours
+0 hours since Oct 23, away from target
Cost per merged change
2.3 usd
target ≤ 3 usd
−1.3 usd since Oct 22, toward target
Roadmap gates met
14.3%
target ≥ 100%
2 of 14 checked, 0 verified (2 claimed, unverified)
±0 pts since Nov 21
Altitude toward target, all summit metrics; the dotted line is the target.
How the walk joined the route · 643 steps
exact key
61%
inferred
1%
off route
39%
39% off the written route: 14% owner merges, 2% exhaust and tech-debt rows, 5% canaries, 18% plan-less rows.
Conditions in the window
weather
83 episodes, 662.9 h, holding
false summits
0
Latest leg
24 changes merged on 2026-11-20, serving 19 waypoints and moving metrics toward targets.
Healthcare integration engines are closed, expensive and hard to observe. fi-fhir is an open, Kubernetes-native engine with profile-driven parsing, canonical events, durable ingestion and delivery, and a Mapping Studio. It is pre-1.0: its capabilities and tests do not yet certify every supported deployment profile or vendor feed, and the climb is toward a 1.0 whose supported matrix they do.
Roadmap gates met (Now list)
59.4%
target ≥ 100%
19 of 32 checked, 10 verified (9 claimed, unverified)
±0 pts since Nov 21
Main pipeline success ratio (30d)
100%
target ≥ 95%
±0 pts since Oct 28
Enqueue-to-merge p50 (press entries)
0.6 hours
target ≤ 2 hours
+0.6 hours since Oct 30, away from target
Altitude toward target, all summit metrics; the dotted line is the target.
Invisible EDI defects (homoglyphs, short counts, mixed line endings, trailer miscounts) pass through healthcare clearinghouses and fail downstream. edilint is a public Go linter and fixer for X12, HL7v2 and EDIFACT that makes those defects visible in CI, climbing from v0.6.0 toward the v1.0 "legitimately useful" exit criteria.
Roadmap gates met (Now list)
58.3%
target ≥ 100%
7 of 12 checked, 4 verified (3 claimed, unverified)
±0 pts since Nov 21
Merges per week
0 merges
target ≥ 3 merges
±0 merges since Oct 23
GitLab pipeline success ratio
100%
target ≥ 95%
±0 pts since Oct 28
Altitude toward target, all summit metrics; the dotted line is the target.
Self-hosted inference on a mixed homelab fleet (two gfx1100 cards, a Radeon VII and a GTX 980 Ti) breaks in ways nobody measures: lanes that hog a card, crash on load or time out behind a cold start. FlexInfer is the Kubernetes operator and OpenAI-compatible proxy that serves the lab's LLM, embedding, rerank, speech, image, audio and video lanes with scale-to-zero, breakers, reroute and per-GPU certification, climbing toward a fleet where every request the proxy routes is answered and every backend change is certified before it serves.
Proxy request success ratio (rolling 1d)
98.2%
target ≥ 99.5%
−1.7 pts since Nov 12, away from target
Roadmap gates met (Now list)
50%
target ≥ 100%
2 of 4 checked, 1 verified (1 claimed, unverified)
±0 pts since Nov 21
Master pipeline success ratio (7d)
77.4%
target ≥ 90%
+77.4 pts since Oct 23, toward target
Merges per week
37 merges
target ≥ 40 merges
+27 merges since Oct 23, toward target
Altitude toward target, all summit metrics; the dotted line is the target.
How the walk joined the route · 259 steps
exact key
11%
inferred
2%
off route
87%
87% off the written route: 87% owner merges.
Conditions in the window
weather
28 episodes, 266.3 h, holding
false summits
0
Latest leg
Eight unplanned operator merges co-moved with mixed metric shifts on 2026-11-20.
Sanitized from the Mills chronicle read model: run, backlog and roadmap ids are hashed; titles, headlines and narrated sentences are scrubbed of paths, URLs, credentials, addresses and unpublished repository names; only climbs on published repositories are listed. Merge request numbers and short commit hashes are kept because they are public; so is the name of the model that narrated each leg, while its backend, prompts and traces are not.