coulomb/frontend-patterns
5
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Audit open work and define a bounded language 0.1 preview

Browse source 
Assistant: codex
Assistant-Model: gpt-6-astra
Assistant-Session: 01a0cffe-206f-7323-89d3-b20ca4cf4bf9
This commit is contained in:
tegwick 2026-09-27 02:19:31 +02:00
parent 5db2d0fdb0
commit 39fea9046e
6 changed files with 217 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

108
history/2026-09-27T021659+0200-workplans-v01-readiness.md Normal file
View file

@ -0,0 +1,108 @@
# Open workplans and a useful language 0.1
Reviewed: 2026-09-27T02:16:59+02:00, Europe/Berlin. Baseline: `5db2d0f`.
Reviewer: Codex, author-model repository audit, not independent validation.
## Conclusion
A useful **0.1 research preview** is feasible from this corpus, after a bounded
consolidation and packaging pass. Finishing every research task is neither necessary
nor currently supported by the evidence. Publication version, editorial completeness
and evidence maturity must remain separate. This audit does not declare a release,
close M1/M2/M3, change task status or confer canonical admission.
The bottleneck is now making the accumulated knowledge navigable and usable: a reader
should be able to start with a problem, follow a short decision sequence, compare
alternatives and reach evidence-qualified candidate guidance. More source collection
alone will not provide that reference surface.
## Inventory and source of truth
Six file-owned workplans: two finished (0001, 0005), three active (0002, 0003, 0006),
one proposed (0004). Eighteen task blocks: nine done, four progress, five todo.
The nine open tasks are assessed below. Generated WORK-RECORDS.md and the offline
brief are stale; they omit recent work/status changes. Use source workplans and the
current State Hub reconciliation receipt, not those historical generated summaries.
Inventory correction: the initial **compact candidate table has 16 seeds**, whereas
[patterns/README.md](../patterns/README.md) contains **59 broader seed names**, already
present at initial commit `2a5dabd`. The earlier 7 developed / 9 undeveloped comparison
applies only to the compact table. It is not coverage of all initial ideas. Neither
list is a release quota. There are still 17 developed candidate records, all hypotheses.
## Task-by-task disposition
| Workplan / task | What exists | What prevents honest closure | Recommended handling for 0.1 |
| --- | --- | --- | --- |
| 0002 T02 — source evidence (progress) | Carniato/local-first reviews, source locators, later conflict/recovery and implementation studies | Initial source records still say extracted; later task notes accumulated storage/receiver/recurrence obligations | Near closure of its original source-review deliverable: write a final source assessment, reconcile statuses and explicitly hand implementation residuals to live tasks. Do not simply mark all evidence work done. |
| 0002 T03 — connected corpus (progress) | Eight connected recovery drafts, VC-01/02 synthetic traces, frozen pilot and submitted model review | M1-SeedReview.md absent; independent human validation/fresh replication still open in task history; usefulness unproven | Produce an honest pass/revise/defer M1 review. Package its supported subset while preserving open validation; no forced M1 pass. |
| 0003 T01 — six streams (progress) | Coverage matrix; documentary material in A–F; extensive B/D/E/F slices and C/D comparison | Independent known uses and complete contracts uneven; bounded Zulip and authority/projection follow-ups remain | Consolidate evidence per stream and name residuals. Further source work only if a release claim depends on it; do not make every interesting follow-up a 0.1 dependency. |
| 0003 T02 — contrasting walkthroughs (todo) | VC-03 and VC-06 provisional; coverage inventory for ten cases | M1 prerequisite open; VC-04/05 absent, VC-03/06 not completed | Keep full M2 coverage post-preview. For 0.1, expose two existing core traces and clearly label provisional/unsupported cases. |
| 0003 T03 — ontology/M2 review (todo) | Seed dispositions, classification mappings and authority/projection decision aid | No M2-CoverageReview.md; case gate open; relation/overlap questions remain | Record a bounded ontology review of the published subset, including Conflict Copy / Explicit Reconciliation. Full M2 stays open. |
| 0004 T01 — metadata/graph (todo) | Schema examples, candidate IDs, classification/evolution checks | Full PatternSchema validation/admitted graph absent; M1 prerequisite open | A candidate-only index/export can support the preview under WP-0007. It must label proposed relations and does not complete this gated canonical task. |
| 0004 T02 — held-out retrieval (todo) | Consumed six-pair development pilot, equal adjudicated totals | M2 prerequisite and fresh held-out comparison unmet; human usefulness unresolved | Defer formal M3 evaluation; perform disclosed development navigation checks for preview usability only. Use llm-connect if live inference is needed. |
| 0004 T03 — real repositories (todo) | Pinned Zulip/Element and feature-example slices | At least two contrasting whole-task applicability studies and M3-EmpiricalReview.md absent | Reuse partial source findings with explicit limits. Do not count file reads or example repos as the completed M3 gate. |
| 0006 T03 — comparable cycles (progress) | Five prospective cycles, protocols, effort, snapshots and report | Five different groups; no comparable trend window; 14 unknown substantive-review dates | Reconcile dates where supported and publish unknowns. Keep longitudinal work live beyond 0.1; do not manufacture cycles to finish a task. |
WP-0001 and WP-0005 remain finished. No existing open workplan can be declared fully
finished by this audit. In particular, tests and many research receipts do not substitute
for its specified review artifact or independent validation.
## Proposed 0.1 release contract
The proposed [WP-0007](../workplans/FEP-WP-0007-language-v01-preview.md) owns packaging
and release review, avoiding duplicate ownership of empirical tasks.
A reader should be able to:
1. Start from a transactional or offline problem and identify assumptions and forces.
2. Follow one of two short decision sequences to a connected core subset of the
existing candidates, with alternatives, resulting contexts and failure boundaries.
3. Find all seventeen drafts through family/scale/responsibility/force indexes; tell
developed candidates from the 16 compact and 59 broader seed inventories.
4. Reach evidence-qualified implementation examples for both core cases without
mistaking a mechanism example for a certified complete implementation.
5. Export stable research identities and proposed relations with provenance and
limitations; inspect current evidence and the history of revisions.
6. See what is unsupported, how to report a counterexample and what changes in later
versions. Hypothesis status remains visible at point of use.
Release acceptance: references resolve, identities are stable, proposed edges have
rationale and sources/unknowns, source/draft ownership is clear, and development
navigation checks expose no unresolved misleading claims. Existing appropriate tests
pass. The release review identifies reviewer role and evidence limits, pins the
corpus revision, and lists open validation tasks. This is a practical reference
preview—not an empirical claim that the language improves decisions.
Excluded from the preview gate: all 59 seeds developed, every case complete, canonical
maturity, general framework coverage, proven production guarantees, independent human
usefulness, and diminishing-returns measurement. These exclusions change the preview
boundary, not the acceptance criteria of the existing research milestones.
## Order of work
1. Reconcile source/seed/ontology review artifacts against actual evidence; classify
acceptance criteria as pass, revise or defer. Transfer residuals into explicit live
task records before finishing any workplan. No historical evidence rewriting.
2. Curate the existing eight recovery candidates as the initial connected core, with
the remaining nine available as extension hypotheses. Publish decision entry points,
two sequences and scoped technology/example mappings; avoid a new application UI.
3. Add the smallest candidate-reference export and validation needed by those paths.
Inspect selected links and decisions through fixed development queries. Preserve
the frozen pilot; do not call these fresh held-out validation.
4. Review the package as a versioned 0.1 research preview, record remaining limits and
produce release notes. Tag/publish only when that release record supports it.
The currently pending pinned-application comparison is not automatically the next
priority if consolidation is chosen. It remains live under WP-0003 T01, alongside
the Zulip residual. This audit recommends concentrating on the finite release contract
rather than allowing every new source question to extend it indefinitely.
## Evidence and limitations
Based on all six source workplans, [ResearchProgram](../research/plan/ResearchProgram.md),
[ValidationCases](../research/plan/ValidationCases.md), [Admission](../research/bench/Admission.md),
[the evolution report](../research/evolution/EvolutionReport.md), source reviews,
walkthroughs and the reference stubs. State Hub reported no FEP tasks flagged
needs_human; that administrative flag does not mean human validation is complete.
No new external research, inference or application experiment was performed.