# Arkiv Ideathon — August 2026

[![verify](https://github.com/PugarHuda/arkiv-ideathon/actions/workflows/verify.yml/badge.svg)](https://github.com/PugarHuda/arkiv-ideathon/actions/workflows/verify.yml) — tsc against the published SDK · 17 invariants · Playwright page QA on Linux, every push

Two entries for *What can YOU [ ARKIV ] ?*, the Arkiv Network ideathon (ideas + Arkiv data-model design, no deployment). One idea per track, as the rules allow.

| Entry | Track | Closes | Write-up | Form answers |
| :--- | :--- | :--- | :--- | :--- |
| **SELISIH** — a multi-witness flight recorder for DeFi risk state | Challenge 3 · DeFi (host: Marcos Miranda) | 31 Aug 2026, 23:59 UTC | [`submissions/01-defi-selisih.md`](submissions/01-defi-selisih.md) · [artifact](https://claude.ai/code/artifact/c4bf64e1-1f53-4a42-b6a3-de9e0aa91ee2) | [`submissions/FORM-selisih.md`](submissions/FORM-selisih.md) |
| **LAYAK** — safety certification that cannot be out of date | Open lane · Other (host: Santiago Zuluaga) | 31 Aug 2026, 23:59 UTC | [`submissions/02-other-layak.md`](submissions/02-other-layak.md) · [artifact](https://claude.ai/code/artifact/60440808-a338-45c2-9c30-a390939e2226) | [`submissions/FORM-layak.md`](submissions/FORM-layak.md) |

**Public site (no login):** <https://pugarhuda.github.io/arkiv-ideathon/> — [SELISIH](https://pugarhuda.github.io/arkiv-ideathon/submissions/selisih.html) · [LAYAK](https://pugarhuda.github.io/arkiv-ideathon/submissions/layak.html) · [evidence](https://pugarhuda.github.io/arkiv-ideathon/submissions/evidence/)

Submission form: <https://tally.so/r/OD9eeY>. The form answers **are** the submission; the artifacts are the optional supporting link. Artifacts are private until shared from the page's share menu — share before pasting the link.

## The two ideas in one line each

- **SELISIH** (Indonesian: the *difference* between two numbers; *berselisih*, to be in dispute). Independent witnesses each publish their own signed snapshot of the same lending market, and the product is the disagreement between them — kept apart, never aggregated, funded by whoever wants it kept. Kill-tested twice on real Ethereum mainnet state, including Chainlink's own 31 node observations decoded from `NewTransmission`.
- **LAYAK** (Indonesian: *fit* — fit to operate; a crane's certificate is a *Surat Keterangan Layak Operasi*). Statutory inspection certificates written as entities whose lifetime **is** the validity period, so an expired certificate is not a row with a stale date — it is a row that no longer exists to be returned. Grounded in Permenaker 8/2020 and positioned against Kemnaker's TemanK3.

## Repository layout

```
README.md                         this file
audit.py                          coverage check: 25 Arkiv primitives × both write-ups
review.py                         calls the Ideathon MCP's review_my_idea tool on a draft
submissions/
  01-defi-selisih.md              full SELISIH write-up (source of truth)
  02-other-layak.md               full LAYAK write-up (source of truth)
  FORM-selisih.md                 SELISIH answers mapped to the Tally form's fields (long versions)
  FORM-layak.md                   LAYAK answers mapped to the Tally form's fields (long versions)
  FORM-700.py                     what actually gets pasted: every long-text field trimmed to Tally's
                                  real 700-char cap (the form rejects more), plus its two tooling
                                  questions. `python submissions/FORM-700.py` prints each field with its
                                  count and exits 1 if any field is over.
  selisih.html / layak.html       the published artifact pages (diagrams, charts, embedded code)
  selisih.sketch.ts               entity-model sketch — type-checks against @arkiv-network/sdk@0.7.0
  layak.sketch.ts                 entity-model sketch — type-checks against @arkiv-network/sdk@0.7.0
  evidence/                       everything that is evidence rather than description — see its README
  video/demo/                     the two demo pages — each runs the real sketch code against MemArkiv with
                                  blocks ticking at 2s off the wall clock. index.html + demo.ts (LAYAK gate
                                  check) · selisih.html + selisih.ts (divergence board, reading the decoded
                                  Chainlink round from evidence/) · narrate.py (neural TTS + subtitles) ·
                                  record.mjs (the 1080p capture) — both driven by a spec name
  video/LAYAK-expiry.mp4          one real-time take: a 40-second certificate goes green, then red
  video/SELISIH-divergence.mp4    one real-time take: 31 decoded node readings, 868 bps apart, and the outlier
                                  lapsing while the pin a disputant paid for survives
  video/layak.html                one page per entry — a judge reviewing one track never lands on the other's
  video/selisih.html              videos. index.html is a thin hub linking the two.
  video/                          LAYAK.mp4 · SELISIH.mp4 (+ .srt/.vtt) — Playwright recordings of the live pages with an
                                  injected pointer and burned-in captions, neural voiceover (Edge TTS), a toggleable
                                  English subtitle track, and Remotion intro/outro cards. 1080p, ~2:17 each. Nothing staged.
```

## What is verified, and how

Nothing here is deployed — the event forbids it and there is no open Arkiv network during August. Everything below runs offline or against public Ethereum mainnet.

| Claim | How it is checked | Where |
| :--- | :--- | :--- |
| The schemas are real Arkiv code, not pseudocode | `tsc --strict` against the published `@arkiv-network/sdk@0.7.0` package, installed from npm | `submissions/*.sketch.ts`, `evidence/tsconfig.sketch.json` |
| The designs' logic holds under Arkiv's documented rules | 17 invariants execute the sketch code against `MemArkiv`, an executable spec with every rule cited | `evidence/memarkiv.ts`, `evidence/invariants.test.ts`, `evidence/invariants.result.txt` |
| Independent observers of a DeFi market really do disagree, hardest under stress | Chainlink vs Uniswap ETH/USD at the same block, 62 blocks across the 5 Aug 2024 crash and a calm window, via archive `eth_call` | `evidence/killtest.py`, `killtest.txt`, `killtest.json` |
| The incumbent oracle's own nodes disagree — and the disagreement is discarded | 179 `NewTransmission` rounds decoded; per-node observations attributed by index | `evidence/chainlink_nodes.py`, `chainlink_nodes.json`, `chainlink_round_20458998.json` |
| Both write-ups use (or explicitly decline) every documented Arkiv primitive | `python audit.py submissions/01-defi-selisih.md submissions/02-other-layak.md` | `audit.py` |
| The walkthrough videos show the real pages | Playwright `recordVideo` against the public GitHub Pages URLs; pointer and captions injected into the DOM; per-step timing derived from the measured length of each narration line; Remotion renders the title/outro cards; ffmpeg muxes voiceover and subtitles | `submissions/video/` — the two walkthroughs were built in a scratch workspace and only the finished MP4s were kept; **the expiry demo below is the one whose pipeline is committed and rerunnable** |
| The SELISIH demo runs on the real decoded round, not on invented numbers | The page fetches `evidence/chainlink_round_20458998.json` and writes one entity per observation through the sketch's own `reveal()`; `qa/demo.mjs` asserts the board reproduces the evidence — 31 reporting, 868 bps spread, the +454 bps outlier named by `$creator` — then that it drops to 30 when the outlier lapses, with the disputant's `extendEntity` refused and its pin surviving | `submissions/video/demo/selisih.ts`, `qa/demo.mjs`, CI |
| The expiry video is one unedited real-time take | `record.mjs` writes the certificate, records 1080p while the wall clock runs, and prints when the gate flipped: `certificate lapsed at t=40.06s; the "Red" line starts at 39.924s`. Nothing is sped up or cut — captions and the entity read the same clock, and the lifetime is derived from the narration so they cannot drift. SELISIH the same way: `ttl 62s; entity lapsed at t=62.57s; the "h" line starts at 61.422s` | `submissions/video/demo/record.mjs`, `narrate.py` |
| An expired certificate really does leave the query surface | `qa/demo.mjs` drives the demo page: GREEN at the start, `RED_NO_CERT` once the lifetime lapses, blocks advanced on the wall clock, `ArkivEntityExpired` emitted, no console errors — on Linux, every push | `submissions/video/demo/`, `qa/demo.mjs`, CI |
| The pages render correctly on phones and desktops | `qa/site.mjs` — Playwright against a local server and against the live site: standards mode + viewport meta, no overflow, fonts, SVG label collisions, dark-mode contrast, embedded code, link check | `qa/site.mjs`, CI |

## One command

```sh
npm install && npm test        # tsc --strict against @arkiv-network/sdk@0.7.0, then the 17 invariants
npm run qa:site                # Playwright: every page × iPhone 13 / 929 / 1440 — standards mode, overflow, fonts,
                               # SVG label collisions, dark-mode contrast, embedded code, every internal link
npm run qa:site:live           # the same against the published GitHub Pages site
npm run build:demo && npm run qa:demo   # the expiry demo: green, then red, with nothing run in between
npm run video:demo             # re-record LAYAK — neural voiceover, subtitles, one real-time 1080p take
npm run video:selisih          # the same for SELISIH
```

The same three steps run in GitHub Actions on every push (`.github/workflows/verify.yml`).

Attribute naming follows Arkiv's published `arkiv-best-practices` skill: every entity carries a `project` attribute and a `type` classifier; numerics are integers; expiry via `ExpirationTime` helpers; one wallet never writes in parallel.

## Reproducing the evidence

```sh
# 1. type-check the sketches against the real SDK
mkdir sdk && cd sdk && npm init -y && npm i @arkiv-network/sdk@0.7.0 typescript @types/node
cp ../submissions/*.sketch.ts . && cp ../submissions/evidence/tsconfig.sketch.json tsconfig.json
npx tsc -p tsconfig.json            # exit 0

# 2. run the invariants (Node ≥ 20)
mkdir -p run/evidence && cp ../submissions/*.sketch.ts run/ && cp ../submissions/evidence/{memarkiv,invariants.test}.ts run/evidence/
# tsconfig: module es2022, moduleResolution bundler, rootDir run, outDir dist — then:
npx tsc -p tsconfig.run.json && echo '{"type":"module"}' > dist/package.json && node --test dist/evidence/invariants.test.js

# 3. the kill tests (needs an archive RPC; eth.drpc.org serves historical state without a key)
python submissions/evidence/killtest.py          # ~3 min
python submissions/evidence/chainlink_nodes.py   # ~1 min
```

## Sources the designs stand on

- Ideathon MCP: `https://ideathon-mcp.arkiv.network/api/mcp` — rules, rubric, `arkiv-fundamentals`, `ideation-guide`, `network-and-roadmap`.
- `@arkiv-network/sdk` 0.7.0 (published) and 0.8.0-dev.3 (the September testnet's architecture) — read from source.
- Permenaker No. 8 Tahun 2020 (K3 Pesawat Angkat dan Angkut); LOLER 1998 (UK) for the retention duty; Kemnaker TemanK3 (digital K3 documents since 3 Dec 2025).
- Aave: the 5 Aug 2024 liquidations (~$231M) and the 10 Mar 2026 CAPO incident (~$27M, Chaos Labs post-mortem).

## Status

Both entries are complete and paste-ready, and both demos are filmed — one real-time take each, nothing cut: [`LAYAK-expiry.mp4`](submissions/video/LAYAK-expiry.mp4) (a 40-second certificate going green, then red, with nothing run in between) and [`SELISIH-divergence.mp4`](submissions/video/SELISIH-divergence.mp4) (31 decoded Chainlink node readings becoming 31 attributable entities, and the outlier lapsing). Not yet done, and only the author can do them: share the two artifacts, and submit both forms (the first submission's timestamp is what settles precedence against a similar idea; later revisions keep it).