MO§ES™ — Honest Comparison

MO§ES™

#1 in All 5 · Honest · Raw JSONL

Five Measured Kernels · vs. Field Average

seven days · 1,465 tasks · 35,242 LOC

	Input	Output	Cache Create	Cache Read	Total	Sessions / Tasks	LOC
MO§ES™	123K	3.9M	34.8M	1.08B	1.12B	98 / 1,465	35,242
Field avg per model	162.9M	17.2M	—	1.49B	1.67B	1 / 358	7,160
Δ field ÷ MO§ES	1,322×	4.4×	—	1.4×	1.5×	0.24× (MO§ES did 4× more)	0.20× (MO§ES shipped 4.9× more)

Field avg per model = mean of 13 published AA models · raw AA telemetry · 1 benchmark run × 358 fixed tasks per model · LOC = AA's 20 LOC/task convention (not measured shipped code) · Cache Create not reported by AA · Time: 44.9 hrs (MO§ES total) vs ~71 hrs (per field model run)

MO§ES™ (raw JSONL)

Field average (AA Index)

I · Cache

Cache Hit Rate

94.66%

90.68%

MO§ES

Field avg

+3.98pp · #1

Less fresh compute per session

Every MO§ES session starts with 5,162 tokens of governance context already cached: constitutional framework, coordination state, project history. The field starts cold every task. That 3.98pp gap understates the real advantage: MO§ES reads 1.08B from cache across 1,465 tasks. The field averages 455K fresh input per task, 3,700× more cold ingestion than MO§ES, and discards most of it between runs. Cache reuse, not hit rate, is the structural edge.

II · Compression

Output : Input Ratio

17.9×

0.162×

MO§ES

Field avg

110× more output per input token

More output per token of fresh input

To produce the same 3.9M output tokens, the field average needs 24.1M input tokens. MO§ES needed 218K. The conservation cascade: cached context carries the semantic load so fresh input drives maximum output. The field average uses 110× more input to produce the same output.

III · Efficiency

Tokens per Task

810K

4.67M

MO§ES

Field avg

5.8× fewer tokens · same output

Extra tokens the field would burn

Across 1,465 tasks: MO§ES uses 1.12B tokens. The field needs 6.84B. That 5.72B gap isn't smaller tasks. MO§ES shipped 35,242 LOC. It's reuse: governance context injected once, cached across every session, never recomputed.

IV · Speed

Time per Task

1.84m

11.92m

MO§ES

Field avg

6.5× faster · 246 hrs saved

Hours saved across the window

44.9 hrs of compute vs ~291 hrs for the same 1,465 tasks. The speed advantage isn't hardware. It's session bootstrapping. MO§ES sessions start with full system state already loaded. No ramp-up. No re-establishing context. Direct task engagement from message one.

V · Cost

Cost per LOC

$0.0007

$0.067

MO§ES

Field avg

96× cheaper per line shipped

On 35,242 lines of production code

$23.33 vs ~$1,963 for the same LOC output. MO§ES runs on a $23.33/week subscription. Flat rate regardless of volume. The field pays per token at market rates. Token efficiency compounds the gap: even at pay-per-token pricing, the 6.1× token advantage alone would cut the field's bill by 84%.