RCIRODERICK CONSULTINGTalk to an engineer
← Babel

Research Notes · Babel

Working research notes.

Self-contained pieces drafted as the work develops. Each note stands on its own and is marked with its dates and current status. Later, when the underlying questions have settled, individual notes will be compiled into longer papers.

· Published May 16, 2026

The Two Questions

Two questions structure this research thread. The design question — what does the most efficient programming language look like, given that it needs to serve humans, machines, and LLMs — reduces to which corner of a multi-stakeholder Pareto frontier is worth occupying. The gating question — does building a new language add value over the existing 51-artifact field — has no answer yet, by design; the null result is real and acceptable. The methodology developed across these notes is the apparatus by which both questions become reviewable.

· Published May 16, 2026

Stakeholders, Tokens, Time

Two complementary views of efficiency for LLM-oriented PL design. The structural view inventories nine stakeholders (human author, reviewer, maintainer, compiler, IDE, LLM-as-consumer, LLM-as-generator, code-as-data systems, security auditor). The economic projection reduces those nine onto two scarce resources — Tokens (priced per million by LLM providers) and Time (priced in dollars-per-hour for everyone else). Quality is a constraint, not a resource. The first move of methodology: name the stakeholder you're optimising for.

· Published May 16, 2026

The Empirical Cascade, Sequenced by Measurement Cost

How to settle LLM-oriented PL design questions empirically without burning the budget on expensive measurements that cheaper ones could have ruled out. Five stages, ordered by cost: free tokenizer measurements → cheap toolchain instrumentation → ~$100s of LLM API calls → ~100× more expensive human studies → synthesis. Stages 1–3 are also the gating-question apparatus: if an existing combination dominates the Pareto frontier on the cheap stages, the right outcome is to ship the methodology and not build the new language.

· Published May 16, 2026

The LLM-Oriented Programming-Language Design Field, mid-2026

Fresh literature survey of the field. 51 distinct artifacts across seven structural categories: whole executable languages, markup/data formats, transpilers and rewriters, host-embedded DSLs, decoder-boundary constraint systems, position papers, empirical studies. Doubles the original May 2026 count, mostly through the LMPL 2025 long tail and a thickening practitioner ecosystem (NanoLang, B-IR, TOON). Includes seed-list attribution and date corrections, and what we'll be watching for LMPL 2026.

· Published May 16, 2026

The Nine Design Axes — Verified, and a Tenth (Enforcement Locus)

Audits the nine design axes the methodology paper proposed (formatting compactness, AST minimalism, vocabulary redesign, morphological density, BPE-tokenizer alignment, grammar regularity, ambiguity load, human-readability priority, ecosystem maturity) against the mid-2026 field. The nine hold up for executable languages. A tenth axis is forced by five artifacts: enforcement_locus, set-valued over {author, decode, parse, runtime}. A weaker eleventh candidate (artifact-kind) belongs in the schema layer rather than the design-axis cluster.

· Published May 16, 2026

Per-Artifact Audit Table — 51 Artifacts Across the (9+1) Axes

The raw audit data: per-artifact scoring across the nine original design axes plus the tenth (enforcement_locus). 22 artifacts that propose a language of their own get real scores; the other 29 (position papers, empirical studies, decoder-boundary constraints) appear as n/a rows on the original nine, though the decoder-boundary cluster gets real values on the tenth. Includes per-artifact justification paragraphs for the non-obvious scoring decisions.

· Published May 15, 2026

The Four-Way Verbosity Stratification

The LLM-oriented PL field has been treating verbosity as a single number. The four cost axes (bytes / characters / morphemes / tokens per op) correlate tightly for analytic-language interventions like SimPy but decorrelate sharply for morphologically-rich substrates like Inflexión. Empirical measurement across five programs and six tokenizers shows the decorrelation directly; the methodological contribution is making the design trade explicit.

· Published May 16, 2026

The Youvan / Inflexión Tension, Resolved Empirically

The field's central-seeming argument — Youvan's "natural-language morphology is inefficient for AI tokenization" versus Inflexión's "morphology packs semantic distinctions tighter than English-keyworded syntax" — turns out not to be an argument. Both positions describe different points on the same two-dimensional trade-off between local semantic density and tokenizer-tail fragmentation cost. The open question is which point wins on a given workload, not which position is right in the abstract. Same pattern applies to SimPy, Pel, Quasar, LLMON: workload-local wins, no shared cross-workload methodology.

· Published May 16, 2026

Disposition: What This Is, What It Is Not, and the Conditional Installment 08

What this research thread is, what it is not, and what gets contributed regardless of which empirical outcome lands. Three claims this is NOT (a benchmark paper, a manifesto against the existing field, a commitment to build a new language); three conditional outcomes for a future Installment 08 (Python+SimPy dominates → stop; specific gap exists → build for the gap; frontier is incoherent → build a deliberate Pareto-corner artifact or stop); five named methodological contributions that stand even if the gating question returns the null result.

· Published May 16, 2026

Babel Parameter-Sheet Candidates (BF-Family, mid-2026)

Fresh survey of esolangs.org BF-family entries that Babel could cover today via parameter sheets (no runtime work). 32 candidates across three sub-populations (unimplemented, abandoned-interpreter, culturally-significant). Top three Babel-fits today: Ook!, Chespirito, Mierda. Four schema gaps surfaced — small InstructionOp / Encoding enum additions plus a completeness-validator tweak — that would unblock 5–6 more candidates without runtime work.

· Published May 16, 2026

Beyond Brainfuck-Tape: Runtime-Extension Feasibility for Babel

Scoping the six non-tape base machines Babel's schema already validates (stack, OISC, fungeoid-2D, register, queue, string-rewriting). Per-family canonical examples, effort estimate, architectural blockers, strategic value. Top recommendation: ship the stack family first (long-weekend runtime delta, 1,003-entry population, queue family rides along, Befunge reachable after). Runner-up: Subleq (~50 lines). End-to-end 4–6-week roadmap to BF-tape + stack + queue + OISC coverage.