tegwick
b8a03a3bee
Assistant: codex Assistant-Model: gpt-6-astra Assistant-Session: 01a0cffe-206f-7323-89d3-b20ca4cf4bf9 |
||
|---|---|---|
| .. | ||
| bench | ||
| candidates | ||
| cases | ||
| catalog | ||
| compendium | ||
| concepts | ||
| evaluations/language-pilot-01 | ||
| evolution | ||
| examples | ||
| experiments | ||
| plan | ||
| reviews | ||
| walkthroughs | ||
| README.md | ||
Research
Research material supports the canon but is kept separate from canonical patterns.
The structure follows the C3 research deliverable model where useful:
catalog/— normalized inventories and indices.compendium/— explanatory synthesis of established knowledge.concepts/— emerging models and hypotheses.
Additional directories:
cases/— focused analysis of articles, systems or architectural schools.plan/— research program, methods and open questions.
Research material may be provisional.
Promotion into the canon should happen only after the concept has been reconciled with competing models and tested against multiple frontend architectures.
Start a research session
- Read ResearchProgram for phases and current work records.
- Follow ResearchProtocol for evidence and promotion.
- Select a question from OpenQuestions and a bounded task.
- Use ValidationCases and the templates in
templates/. - Update catalogs and question outcomes, then commit and sync the workplan status.
Execute bounded source investigations with the research bench. Its checked-in fixture verifies the pipeline and is excluded from research evidence.
Current contrast: VC-01 server-validated editing, VC-02 offline editing, and cross-case assessment.
Latest implementation evidence: frontend recovery and walkthrough correction review.
Executable evidence: Matrix SDK recovery experiment, with a synthetic receiver and explicit limits.
Evaluation: matched presentation pilot completed twelve llm-connect sessions; results inconclusive, human review pending.
Pilot follow-up: submitted review adjudication gives 18/24 obligations in both conditions; model-assisted review, not independent human validation.
Current follow-up: shared guidance corrections and navigation/rendering evidence, with provisional VC-03. Ten developed hypotheses; M1 remains open and the broader M2 evidence gate is not satisfied.
Latest expansion: six broader candidates and decision probes cover presentation logic, derived state, feature organization, client activation, failure containment and modal focus. Current total: sixteen developed hypotheses; previous counts above describe earlier slices. No additional runtime or model runs.
Latest execution: rendering/focus boundary experiment uses pinned React/Radix packages in Chromium. It records seven stable case observations and a timing-sensitive nested-dialog failure; no production or accessibility certification.
Nested-dialog follow-up: ownership isolation captures a stale Escape-listener interval in a 48-trial browser/build comparison. A partial guard loses input; a complete remedy remains open. Original results retained.
Latest bounded mechanism test: ownership routing meets 24/24 routed assertions versus 20/24 unmodified controls, with explicit scope limits. No delay or readiness wait; no production fork or canonical promotion.
Latest breadth evidence: feature versus deployment boundaries compares pinned example sources and adds provisional VC-06. Independent Feature Boundary develops an existing seed; current corpus has seventeen hypotheses and four synthetic walkthroughs. No deployment independence was runtime-verified.
Latest execution: feature release lifecycle records 14/14 expected observations across two browsers. It distinguishes fresh, already-loaded and delayed-load sessions, and includes negative failure controls.
Latest release follow-up: activation conflicts and unfinished-work reload records 21/21 expected observations, including blind-update and unguarded-reload loss controls. Original experiments retained; production atomicity/unload limits remain open.