{ "schema_version": 1, "builder_plan": { "scenario": "Start with one repeated product task that has a measurable failure or cost.", "change": "Read the methods and identify one change you can isolate in that workflow.", "baseline": "Your current implementation on the same held-out tasks.", "steps": [ "Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.", "Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.", "Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.", "Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test." ], "measure": "Successful outcomes, response time, human rework, and total cost per successful task.", "decision": "Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation." }, "builder_review": null, "arxiv_id": "2609.19145", "title": "Objective vs. Search: Decomposing What Makes a Good Tokeniser", "summary": "The analysis replaces the default confounded choice between BPE and UnigramLM with a 2x2 decomposition (objective vs. search procedure) and two new tokenisers, BottomUpLL and TopDownComp, showing that bottom-up search, not the objective, drives better language modelling. The cost is only that of retraining a tokeniser and, if adopted, the model itself; inference cost and complexity are unchanged for the reader using standard off-the-shelf tokenisers. The finding can fail to translate into downstream task gains, since BLiMP results show no consistent effect, so grammar-sensitive or task-specific behaviour may not improve even when bits-per-byte does.", "rationale": "For a small team that trains or fine-tunes with a chosen vocabulary, the actionable guidance is simply to prefer bottom-up tokenisers such as BPE, which is already the widely available default and requires no new infrastructure.", "technique": "Decomposed tokeniser design: bottom-up vs top-down search", "family": "outro", "recommendation": "adotar", "published": "2026-09-16", "as_of": "2026-10-18", "editorial_status": "indexed", "source_url": "https://arxiv.org/abs/2609.19145", "independent_implementations": 0, "total_implementations": 1, "stars": 1, "citations": null, "claims": [ { "claim_id": "claim-01", "statement": "The analysis replaces the default confounded choice between BPE and UnigramLM with a 2x2 decomposition (objective vs. search procedure) and two new tokenisers, BottomUpLL and TopDownComp, showing that bottom-up search, not the objective, drives better language modelling. The cost is only that of retraining a tokeniser and, if adopted, the model itself; inference cost and complexity are unchanged for the reader using standard off-the-shelf tokenisers. The finding can fail to translate into downstream task gains, since BLiMP results show no consistent effect, so grammar-sensitive or task-specific behaviour may not improve even when bits-per-byte does.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.19145", "source_page": null, "source_excerpt": "" }, { "claim_id": "claim-02", "statement": "Bottom-up tokenisers consistently achieve lower bits-per-byte than top-down ones across model sizes, vocabulary sizes, and domains, with no quantified margin reported; BLiMP shows no consistent relationship to tokeniser design.", "basis": "inferred", "result": "", "baseline": "", "conditions": "", "source_url": "https://arxiv.org/abs/2609.19145", "source_page": null, "source_excerpt": "" } ], "exposure_map": [ { "dimension": "quality", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compute", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "latency", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "operations", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "compatibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "security", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "data_and_training", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" }, { "dimension": "reproducibility", "basis": "not_evaluated", "finding": "", "interpretation": "", "limitation": "", "next_check": "", "source_url": "", "source_page": null, "source_excerpt": "" } ], "risks": [], "minimum_test": [], "open_questions": [], "independent_tests": [], "lab_reproductions": [], "report_available": false, "validation_tier": "unknown", "evidence_tier": "unknown", "infrastructure_basis": "unknown", "equations": [], "equations_status": "not_fetched", "core_kind": "", "equations_fetched_at": "", "equations_source_url": "", "equations_source_sha256": "", "exposures_reviewed_at": "", "exposures_source_sha256": "" }