2026-08-11 — verex history

Source: jay's request (portfolio-readiness for prospective employers) + live investigation of residual balances on prod. Related: 2026-08-10 history — the question-key nonce fix (PR #23) whose absence in old data explains today's finding.

Prod ghost balances: root cause is pre-nonce markets on a persistent chain, not the DB

• Cause: jay noticed re-seeded/removed markets leaving "remaining data" and suspected reused market ids.
• Reasoning: every DB id is @default(cuid()) (random per row), so DB ids can never survive a re-seed — the DB is clean. What survives is Sepolia (prod chainId 11155111, persistent). Prod's 10 markets predate PR #23's question-key nonce, so their questionId → conditionId → tokenIds are pure functions of the slug: identical across every seed generation ever run against the prod backbone. Old sessions' ERC-1155 balances therefore reattach to today's markets.
• Change: none yet — diagnosis only. Verified live: prod wallet 1 holds 4,217 USDC and positions like 270.3 tokens against a $100 cost basis; wallet 3 holds exactly 2,000 USDC (two seed-era 1,000 mints — the seed's top-up guard normalizes up, never down). Local (fresh nonce-era seed) shows all wallets clean at 1,000/0.
• Result: fix plan proposed to jay: (1) seed dates relative to run time, (2) seed normalizes USDC down to the 1,000 target, (3) full prod re-seed (drop SKIP_SEED) so nonce-era question keys mint fresh token ids and evict the ghosts. jay deferred all of it ("never mind the previous things except the one") — parked alongside the other open decisions; the diagnosis above stands whenever it's picked back up.

Docs site: new "Architecture: web, API & Sepolia" page (/docs/chain)

• Cause: jay wants to show Verex to prospective employers and asked for a docs web page (not a markdown file) explaining how the product talks to Sepolia and the contracts — specifically naming the contracts and how the API and web relate to them.
• Reasoning: the docs system stores documents as en/ko data blocks rendered by DocBody, so the page is a new content file, not new markup. Two supporting changes were needed: (a) DocBody had no link syntax, and Etherscan-verifiable addresses are the point of the page — added a tiny [text](url) inline form; (b) the web Docker build (--source packages/web) can't see packages/contracts/deployments.json, so the addresses are mirrored as data with a source-of-truth comment instead of imported.
• Change: content/docs/chain.ts (5 sections × en/ko: four-layer architecture, the five contracts with origin/role, Sepolia address tables with Etherscan links for staging+prod, server-held-keys custody model, which-layer-calls-what flows), registered in lib/docs.ts after settlement; link support in DocBody.tsx. Also next.config.js now honors NEXT_DIST_DIR so a verification build can run beside a live server on .next — yesterday's build-corrupts-dev-server gotcha, fixed structurally (.next-verify/ gitignored).
• Result: verified via an isolated NEXT_DIST_DIR=.next-verify build served on :3187 — all 5 sections render in both locales, 8 unique Etherscan links present, sidebar/pager pick the doc up. (jay's local :3000 next start still serves yesterday's corrupted .next — every docs page 500s there, unrelated to this change; a rebuild/restart on his side clears it.)