Program menu: what we build to put Mo through its paces
Robert (session 2, on Q14): one program is not enough; build a variety of kinds to find Mo’s strengths and weaknesses and collect real data to improve the language. This is the measurement engine behind direction 28.
The menu, in rough order
| # | Program | Stresses | Measures |
|---|---|---|---|
| 1 | Durable job queue + HTTP API | processes, supervision, deadlines, never | the founding premise: agent builds it, Robert reads at spec altitude |
| 2 | CLI tool (log analyzer) | pure code, stdlib, single binary | stdlib gaps; is Mo pleasant with no concurrency? |
| 3 | KV store with a wire protocol | bytes, sized ints, hot loops | runtime speed, overflow-check cost ([[q04-integer-types-and-overflow |
| 4 | Web backend (“with Postgres” as proposed; built as notes over Fs, no database brick yet, program-4) | package story, capability-scoped deps | [[q17-package-management-and-supply-chain |
| 5 | Agent harness (tools, budgets, retries) | capabilities as permissions, deadlines | is Mo good at the thing it is for? |
| 6 | Payments ledger with invariants | never, contracts, SMT tier, fault injection | proof rate, tier-3 timings ([[q08-verification-tiers |
| 7 | The Mo toolchain itself (late; program 8 on the roadmap, where program 7 is a Redis subset against Redis’s own tests) | everything, at scale | compile speed on 100K lines ([[d23-compile-speed-first-class |
Collected on every program
- build time per KLOC; incremental rebuild latency
- agent loops-to-green per task
- contract and proof density;
verified:line over time - lines-per-function distribution (tests the 70-line law, Q12)
- crash reports per KLOC and time-to-autonomous-fix (direction 21)