This repository applies the current EIP complexity rubric retrospectively to execution-affecting EIPs from recent Ethereum hard forks. Each EIP is assessed using a historical specification revision intended to approximate what reviewers knew when the proposal entered fork development, rather than the final specification that eventually shipped.
The study is designed to answer two questions:
- Do the current complexity criteria produce sensible results for EIPs whose development histories are already known?
- Do the criteria, their wording, or their weighting need calibration before prospective use?
The initial fork range is Shanghai/Shapella through Amsterdam/Glamsterdam. Execution-only and cross-layer EIPs are assessed; consensus-only EIPs remain in the historical dataset and plots but are not assigned execution-layer complexity scores.
- Published study site presents the methodology, fork and EIP assessments, results, and Human-versus-LLM comparison.
- REPRODUCING.md describes the end-to-end research and review process.
- DATA.md explains what is canonical, derived, generated, or deliberately excluded.
- research/README.md describes the task-oriented directory structure.
- publication/CONTRACT.md defines the sanitized static-publication boundary.
- Every task has its own
TASK.mdcontract, inputs, templates, outputs, and provenance rules.
Task 01: fork membership and EIP histories
|
+--> Task 02: client implementation history
|
+--> Task 03: fork, EL, and CL development timelines
|
+--> Task 04: assessment-ref selection and human review
|
+--> Task 04b: fork evaluation cutoffs and aggregation cohorts
|
+--> Task 05: isolated historical complexity assignment
|
+--> Task 06: independent verification (planned)
|
+--> Task 07: observed-effort metrics and prediction join
|
+--> Task 08: Hegotá prospective candidate snapshot
|
+--> Task 09: Hegotá human-assessment snapshot (publication attribution only)
Task 02 is independent supporting research and is not required to choose the Task 04 historical assessment ref. Tasks 03 and 04 provide the timeline-based human review used to freeze that ref. Task 04b separately defines which final EIPs were forecastable at the initial fork-scoping horizon; it never changes an EIP's assessment ref or score. Task 05 consumes only approved Task 04 records and then seals each EIP into a hindsight-controlled, one-EIP assessment package. Task 07 derives observed-effort metrics score-blind from Tasks 01/03/04/04b, then — only once all 49 original assessments exist — joins them against the Task 05 predictions and renders the comparison charts. Task 08 reuses Task 05's rubric and isolated assessment engine for a separate prospective snapshot of Hegotá candidates; its original PFI study and append-only SFI/CFI extension never enter the 49-row retrospective dataset.
| Task | State |
|---|---|
| Task 01 | Complete for Shanghai, Cancun, Prague, Osaka, and Amsterdam |
| Task 02 | Contract and client cohort prepared; implementation research not yet run |
| Task 03 | All five fork timelines rendered and validated |
| Task 04 | All 49 execution-affecting fork–EIP refs selected, human-approved, and mechanically validated |
| Task 04b | Cutoffs and initial/late aggregation cohorts proposed for all five forks; human review pending |
| Task 05 | All 49 original assessments complete and mechanically validated across the five forks |
| Task 06 | Not yet defined |
| Task 07 | Candidate metrics computed for all 49 rows (observed-metrics-v0); 5 metrics + composite selected via score-blind redundancy analysis; prediction join and interactive charts rendered |
| Task 08 | Complete: original 44-entry PFI freeze retained (37 scored); append-only SFI/CFI extension adds EIP-7805 and EIP-8141; combined 2026-08-26 visibility view contains 46 status-labelled entries and 39 validated assessments |
| Task 09 | Snapshot hegota-human-2026-09-13-3d8c012 of STEEL human checklists for the 46 Hegotá candidates: 2 merged, 15 in open pull requests, 8 in draft pull requests, 21 not yet available |
The publication adapter and Astro site under publication/ project the committed research outputs into the GitHub Pages site. The site contains 49 retrospective assessments plus all 46 Hegotá candidates recorded as PFI, SFI, or CFI at the 2026-08-26 snapshot. Every EIP has one detail page that explains its score criterion by criterion, shows the Human and LLM assessments side by side where both exist, and links into a shareable comparison of up to four EIPs; fork and Results pages show which criteria made each fork heavy. Snapshot status and the status of the STEEL team's human checklists are shown inline, the original 37-assessment PFI subtotal remains explicit, and all prospective data stays out of the Task 07 outcome analysis. Pushes to main rebuild, validate, and deploy the site through the Pages workflow. The site's internal architecture is described in publication/site/README.md.
To build and preview the same artifact locally:
cd publication/site
npm ci
npm run check
npm run build
npm test
npm run preview -- --host 127.0.0.1 --port 4321Open http://127.0.0.1:4321/retrospective-eip-complexity/ after the server starts.
The deterministic parts of the workflow—data joins, sealed package construction, validation, and rendering—are expected to reproduce byte-for-byte from unchanged inputs and locked dependencies.
Historical interpretation and human review cannot be guaranteed to produce identical prose or decisions. They must instead be independently auditable: every observation carries source provenance, every selected ref identifies an exact Git blob, uncertainty remains explicit, and human decisions are retained rather than silently overwritten.
This repository contains the research record and the source for its public static study site. The Pages build publishes only the adapter’s sanitized generated output; raw evidence trees, sealed assessment packages, prompts, logs, caches, and operational metadata are excluded. No project license is currently included, so public availability does not grant additional reuse rights.