The factory, end to end
A small specialist earns its place one artifact at a time. This is the shortest path through the system—from choosing a capability to deciding whether the result should ship, redirect, retry, or stop.
00 / admission rule
If a strong frontier model cannot nearly ace a benchmark, that benchmark cannot grade the Mac model. Fix or drop the ruler first. The objective is frontier capability retained per unit of model size, RAM, latency, and cost.
01–06 / the active loop
The narrative is the index. The linked docs and code remain the source of truth.
Can a narrow capability be specified tightly enough to measure?
Start with the task, the incumbent, and a frozen success bar—not a model. A frontier model first proves the ruler can represent the capability at all.
emitstarget.json + frozen baseline
Which decisions are missing, and where did each training row come from?
Trace review turns misses into slices. Synthetic teachers are useful only when provenance, deduplication, contamination checks, and a held-out gate stay attached to the rows.
emitsdataset + manifest + provenance
Does SFT, preference training, or rejection sampling improve the frozen target?
The readable Swift + MLX path remains the local reference. Every intervention is a recipe with exact data, adapter, optimizer, steps, and composition—not just a technique name.
emitsadapter + recipe + training trace
Did capability rise on the frozen target without unacceptable regressions?
The same evaluator scores baseline and candidate. Slice metrics, legality, raw outputs, latency, RAM, throughput, and uncertainty stay visible; a benchmark whose frontier ceiling fails is rejected.
emitsmetrics + traces + regression gate
Where is this specialist allowed to answer, and when must it redirect?
A specialist artifact includes identity, base model, adapter, prompt contract, eval evidence, resource measurements, and routing constraints. Narrow depth can ship only with an honest boundary.
emitsvalidated specialist package
Should this candidate ship, retry, redirect, or be rejected?
The report binds score delta to regressions, cost, latency, RAM, throughput, missing measurements, and source hashes. Failed attempts remain part of the system because they change what gets tried next.
emitsreport card + ship / retry / reject
case files / decisions, not demos
An improvement, a regression, and a failed ruler are all useful when they force a clear next decision.
The 1.7B arc moved from stock to SFT to GRPO. SFT did most of the work; GRPO added two points, then the approach plateaued below the stronger 4B.
Read the full arc →56 → 76 → 78
The distilled 4B aced the narrow hard gate while losing out-of-domain breadth. That is a routed specialist result, not a general model upgrade.
Inspect the package →58 → 100 / 59.6 → 42.3
The strongest completed frontier attempt did not clear the frozen 1.10× bar over paired random legal play. Training was stopped and the replay was retained.
Watch the failed attempt →1.058× best frontier ratio
The first suite showed a real gradient, but the frontier did not approach the required ceiling. The expanded candidate pool remains for human review—not small-model grading.
Inspect candidate evidence →65% frontier development score
annotated source / where to begin
Qwen Chess made its implementation approachable by naming the modules behind each lesson. These are the equivalent entrypoints here; the repository contains the complete implementation and history.
CLI composition and the top-level run boundary.
State transitions, locks, manifests, and durable run identity.
The readable Mac-local supervised fine-tuning path.
Baseline-versus-candidate comparison under one contract.
Offline compiler from validated evidence to the public before/after report.
The durable memory of shipped, retried, rejected, and incomplete attempts.