LAB / WMSI-EXP-002
Long-First vs. Short-First Generation.
Test whether generating a long answer before compression yields structure absent from a direct short answer.
01 / QUESTION
Representation, Output, and Internal State
Test whether generating a long answer before compression yields structure absent from a direct short answer.
This is a protocol scaffold, not a completed experiment or a frozen execution plan. No measured effect or research conclusion is claimed here.
Hypothesis to test
The manipulation described below produces a reproducible, condition-specific change in the measured system behavior or internal state that is not explained by the ordinary controls.
Ordinary competing explanation
The observed pattern is explained by ordinary sampling variation, prompt/representation artifacts, surface lexical cues, evaluator bias, or the known product/runtime mechanics captured by the controls.
02 / PLANNED DESIGN
How we intend to test it
Compare direct-short, long-then-compress, and long-with-hidden-summary conditions on the same tasks.
Controls
Matched final length; same model and seed families; blind judges.
What we would measure
Novel concepts retained, causal depth, correctness, uncertainty retention, compression artifacts.
Access and instruments
Ordinary software or API access. This describes access requirements, not verified readiness on our Spark.
Inference harness; summarization/compression pipeline; graph scorer.
Before a full run
Feasibility=NOW; priority=P0. Before confirmatory execution, run a small pilot to estimate variance/cost, freeze the scoring rules, and assign a run-count target.
Model revisions, final prompts, sample sizes, cost limits, and scoring rules must be settled through the pilot and protocol-freeze process.
03 / INTERPRETATION
What could fool us
Length leakage, judge bias, truncation artifacts, prompt wording.
What would weaken the idea
If direct-short and long-then-compress are equivalent across tasks and seeds, long generation is not adding recoverable structure.
Observation, inference, speculation, unknowns, and apparatus errors will be labeled separately. One striking transcript cannot establish a general effect.
04 / RECORD
Evidence we plan to preserve
Prompt/input, token budget, raw text, token counts, extracted claims/concepts, judge outputs.
Planned visual evidence
Paired difference plots; concept overlap matrices.
No research runs or raw datasets are published for this experiment. The September 26, 2026 registry and protocol compendium (version 1.0) are the sources for this planning record.
Back to all 90 experiments