[umermjd11]

Summary

task_061026_21 Part E, PR 4: the GI docs, brought in line with the final commands and states. Spec: Developer/tasks/task_061026_21.md. Tracking: Discussion #231.

Draft, and it merges last. Some pages describe behaviour from PRs that are still open:

• #314: deploy --model-id defaults and setDinToken
• #313: per-wallet commit caches (…/commits/<address>/)

I'll mark this ready once both have merged.

Changes, by the spec's table

Doc	What changed
workflows/model-workflow.md	Foundry artifacts + --model-id; create-task-dir / cache-default-artifacts; create-genesis-model / submit-genesis-model; register-request + DIN-Rep approval (with the modelId() check); update-manifest-request; funding before gi start; auditor key prerequisite; train-lms → submit-lm; seed and test-data states; the three reveal windows; T2 set-score; new §16 (claim, release slots) and §17 (both dispute families)
roles/model-owner.md	Deploy options; system get-registry-fee (replacing the nonexistent get-fees); task model-owner … registry commands (replacing the nonexistent model-owner registry …); genesis names; funding; start-reveal for evaluation, T1 and T2; create-testdataset --submit options; set-score; claim; release slots; disputes
roles/auditors.md	register-encryption-key, lms-evaluation reveal, the AuditSeedLocked state, claim, dispute-test-data, dual-role rule
roles/clients.md	train-lms then submit-lm; claim
roles/aggregators.md	lock-seed; claim; reveal-t2 (it has no --batch: there is one T2 batch); dispute / claim-dispute-bond; dual-role rule
getting-started.md	aggregator show-t1/t2-batches (it said model-owner aggregation …); auditor key, lock-seed and reveal; submit-lm; the "rewards not the focus" line replaced; a claim step for each role
manifest.md, guides/keystore-migration.md	Genesis command name; cache-default-artifacts; update-manifest-request; train-lms --submit → submit-lm
workflows/din-workflow.md	The 7 contracts; withdraw → sweep-fees (coordinator and registry); fee values fixed: proprietary registration is 0.00001 ETH, not 0.000001, and open-source requests pay a fee too
setup.md	Option B: pip install "git+https://github.com/InfiniteZeroFoundation/DevNet.git@main" builds from the repo root (main has a root pyproject.toml; dist/ holds only the wheel and sdist). Option A: wheel name dincli-0.2.0, the one on main
DINShared.md §2.2	Funding gate on startGI (TC_GIRewardPoolNotFunded); endGI settles rewards; releaseGIRegistrationSlots is state-neutral. The seed and test-data states were already drawn by #287
CLAUDE.md	7 platform contracts; GI summary with funding, seed locks, keys, reveals, claim and release slots

Also fixed while there:

• The task directory is tasks/<network>/<coordinator_address> (DinContext.get_task_dir). manifest.md and CLAUDE.md had said task_<coordinator_address>.
• guides/fashion-mnist-cnn.md deploys from foundry/out.
• DINTaskAuditor.md §13 No. 9 notes the per-wallet commit caches (test(integration): run a complete GI on the foundry contracts (task_061026_21 Part D) #313).

For Umer

• ROADMAP.md:19: the deploy-blocker sentence ("24,585 B … 9 B over EIP-170") is already gone; the Oct 29 refresh (ROADMAP.md: refresh "What is still open for P3" (several items were fixed in October) #285) rewrote it. Nothing to change here.
• "DevNet 3.0" vs "DevNet 2.0": the README, ROADMAP and public docs all call the next deploy DevNet 2.0. "DevNet 3.0" survives only in BACK_LOG (e.g. BL-27's priority) and the task specs. Which name is right?
• Node README: it still says "local Hardhat devnet". Flagged only, per the spec.

After merge

Follow up on the wiki pages that still say "no dincli command yet": rewards claim/withdraw, release slots, auditor encryption key and disputes all have commands now.

Checks

python3 .github/scripts/check_doc_links.py Documentation   → all inline relative links resolve
python3 .github/scripts/check_doc_links.py Developer       → all inline relative links resolve

Every command and option in these pages was checked against the CLI's command tree on this branch.

🤖 Generated with Claude Code