journal-entry JR-VE-TYP-001-P0
EX-VE-TYP-001 Phase 0 Computational Preflight Execution
EX-VE-TYP-001 Phase 0 Computational Preflight Execution
Objective
Execute the computational preflight for EX-VE-TYP-001, Individual Readability Envelope, far enough to validate the experimental mechanics before any human-subject exposure.
This record contains engineering and protocol validation only. It contains no human performance data and does not resolve HY-VE-PE-002, HY-VE-PE-003, or HY-VE-PE-006.
Implemented
The executable experiment package is under experiments/ex-ve-typ-001/.
It now contains:
- a dependency-light F# domain and analysis core;
- 24 deterministic, space-filling typography conditions;
- bounded font size, weight, width, optical size, letter spacing, word spacing, line height, line length, polarity, and contrast variables;
- four task classes: prose, interface labels, identifiers, and dense comparison;
- a four-cohort condition-to-task crossover;
- a four-cohort task-order counterbalance;
- a 24-trial primary schedule;
- an eight-trial repeatability schedule;
- deterministic participant scheduling;
- pilot-only envelope thresholds;
- a synthetic mechanics simulator;
- a JSON trial schema;
- an explicit data-minimization policy;
- original task stimuli;
- automated tests;
- a command-line preflight/schedule/simulation runner.
Validation Result
GitHub Actions validation on the final implementation head before this completion record:
| Gate | Result |
|---|---|
| F# build | PASS, 0 warnings, 0 errors |
| EX-VE-TYP-001 tests | PASS, 11/11 |
| Existing VisualEngineering.Core tests | PASS, 43/43 |
| Existing VisualEngineering.Cli tests | PASS, 37/37 |
| Total .NET tests | PASS, 91/91 |
| Research validation | PASS, workflow run 68 |
| ROS validation | PASS, workflow run 103 |
| Limen strict verification | PASS, workflow run 102 |
| Visual Engineering tool/package workflow | PASS, workflow run 50 |
| Packed artifact matrix | PASS on Ubuntu, macOS, and Windows with Node 20 and 24 |
The CI workflow IDs for this validated head were:
- Build Visual Engineering tool: 35704559229
- Validate research: 35704559277
- ROS validation: 35704559233
- Limen verify: 35704559361
Important Failure Found During Preflight
The first implementation of the repeatability scheduler intended to use a new
stimulus in session two but did not guarantee it. A test exposed a real collision:
one repeated dense-comparison condition received DENSE-S03 in both sessions.
That is a protocol defect, not a cosmetic test failure. Reusing the same condition/task/stimulus combination could confound repeatability with recognition or memory.
The scheduler was changed so each repeatability probe:
- preserves the original typography condition;
- preserves the original task class;
- selects a session-two stimulus;
- explicitly compares it with the session-one stimulus;
- deterministically advances to a different stimulus if a collision occurs.
The regression test was expanded across 16 participant cohort indices. It passes.
Assumptions Challenged
Assumption: a session-number offset was sufficient to prevent stimulus reuse
Rejected. Deterministic modular scheduling can still collide when the condition's original within-task position differs from its repeatability position. Freshness must be an explicit invariant, not an emergent property of an offset.
Assumption: a full typography factorial is required to map a response surface
Not supported for Phase 0. A 24-point space-filling design provides broad coverage while keeping the primary participant burden to 24 trials. Human pilot data is still required to determine whether this is dense enough for useful response-surface estimation.
Assumption: one global visual optimum is sufficient
Not tested. The design deliberately rotates the same condition across four task classes so this assumption can be falsified later. Synthetic behavior is not evidence.
What Phase 0 Proves
Phase 0 supports the following engineering claims:
- the condition generator is deterministic;
- the declared pilot bounds are machine-checkable;
- every condition reaches every task across four adjacent cohorts;
- every task occupies every serial block position across four adjacent cohorts;
- every participant receives six primary conditions per task;
- repeatability probes retain condition/task identity and change stimulus;
- envelope classification keeps accuracy, comprehension, duration, and effort as independent gates;
- preference remains a separate outcome from measured speed;
- the synthetic mechanics path is deterministic and bounded;
- the experiment integrates without breaking the Visual Engineering tool, research publisher, ROS, or Limen gates.
What Phase 0 Does Not Prove
It does not show that:
- individualized typography improves human performance;
- a stable personal readability envelope exists;
- task-specific visual optima differ in real users;
- user preference diverges from performance in the target population;
- the provisional pilot thresholds are scientifically appropriate;
- 24 conditions are sufficient for the final response surface;
- any disability group has one characteristic response.
Those remain empirical questions.
Human-Study Boundary
Before human pilot execution, the next phase must provide an accessible browser presentation/collection runner with:
- monotonic trial timing;
- deterministic schedule import;
- exact typography application and verification;
- response capture without free text;
- pause/withdraw behavior;
- device and viewport metadata;
- no diagnosis requirement;
- local validation against the trial-record schema;
- export that keeps raw participant data out of the public repository;
- consent/review appropriate to the execution setting.
A small instrumented pilot should then estimate within-participant variance and measurement reliability. Final thresholds, smallest practically meaningful differences, exclusion rules, stopping rules, and main-study sample size must be frozen only after that pilot and before the confirmatory run.
Status
Phase 0: complete.
Human execution: not started.
Hypotheses resolved: none.