# Last Known Book — Canonical Demo Film Contract Status: **WINNING INTELLIGENCE TARGETED REWORK — VIDEO SCRIPT FROZEN FOR FIRST ENCODE** Target runtime: **155 seconds (2:35)** Allowed final range: **120–180 seconds** Product: **Last Known Book** Tagline: **Reconstruct the market your agent actually traded.** This is the canonical first encoded demo contract. It optimizes judge comprehension, not feature count. ## Truth rules for the film - Captured Shannon incident cases are **replay evidence**, not Last Known Book-originated writes. - Packet 003 is a separate **Last Known Book-originated real Shannon testnet behavior proof**. - Packet 003 proves technical behavior and operational containment, not production reliability, profitability, ROI, prevalence, fraud detection or MTTR improvement. - Do not show the disposable wallet, private key, seed phrase or full private receipt bundle. - No live blockchain write is required in the film. - If an explorer or network surface is slow, use committed evidence and label it as captured proof. Never fabricate a live success. ## 0:00–0:10 — Hook ### Picture Open immediately on the casefile, not architecture. Show **AGENT INTENT ≠ VENUE REALITY**. ### Voice > An autonomous trading agent gets an execution that looks wrong. Was it bad trading — or did we misunderstand what the venue actually did? ### Judge objective Understand the tension in one view. --- ## 0:10–0:25 — Pain / baseline ### Picture Quick controlled focus across intent, venue reality, evidence and authority gate. ### Voice > Today, an operator can jump between the order request, receipts, balances, the order book and an indexer — and still choose the wrong corrective action. Last Known Book reconstructs one incident as a single evidence contract. ### Judge objective Specific user: autonomous DreamDEX agent operator. Specific moment: unexpected execution. Baseline: manual cross-surface reconciliation. --- ## 0:25–0:55 — Hero causal reveal ### Picture Use `LKB-001`. Keep the full causal slice visible: `AGENT INTENT ≠ VENUE REALITY → MINT_A_PAIR + CHAIN_INDEXER_DIVERGENCE → MIXED → RETRY_READ → WRITE REFUSED` ### Voice > Here the apparent execution is easy to misread. DreamDEX matched the complementary sides through `MINT_A_PAIR`, while indexed verification lagged behind chain truth. Last Known Book corrects the economics, separates what is observed from what is inferred, and refuses a blind resubmission. > > **This wasn't a whale. It was a mint.** ### Judge objective Land the 24-hour memory hook before minute one. --- ## 0:55–1:20 — Generality without a feature dump ### Picture Click through two cases quickly: 1. resting SELL escrow; 2. expected-vs-actual fill. ### Voice > The same incident contract handles very different causes. A resting sell can escrow outcome tokens without losing them. A fill can differ from the request and still be correctly reconciled. In both cases, a compensating trade would create a second incident, so the correct response can be `NO_ACTION`. ### Judge objective Show this is a mechanism, not a one-case trick. --- ## 1:20–1:55 — Real Last Known Book-originated Shannon proof ### Picture Show the redacted **REAL SHANNON PROOF** card. Use simple event animation or progressive highlights; do not expose private identifiers. Required visible sequence: `Shannon 50312` `fresh exact market / pool / Trading gate` `PostOnly placement: receipt success` `0 fills` `OrderPlaced → OrderRested` `exact returned order cancelled` `OrderCancelled` `tUSDC 1 raw → 1 raw` ### Voice > Replay evidence is useful, but we also wanted a product-originated proof. On Shannon chain 50312, Last Known Book revalidated the exact market, pool, lifecycle and order-book parameters before execution. A bounded PostOnly order was placed, rested with zero fills, and returned one order ID. We cancelled exactly that order. DreamDEX emitted `OrderPlaced`, `OrderRested` and `OrderCancelled`, and collateral reconciled exactly from one raw tUSDC back to one raw. ### Judge objective Make `TECHNICAL_PROOF + BEHAVIOR_PROOF` undeniable without overstating outcome proof. --- ## 1:55–2:15 — Why DreamDEX is load-bearing ### Picture Overlay only the primitives that matter: `marketId` · rolling pool binding · lifecycle · `MINT_A_PAIR` · resting escrow · native order events. ### Voice > DreamDEX is not a logo in the stack. `marketId`, rolling pool binding, Event Contract lifecycle, `MINT_A_PAIR`, escrow semantics and native order events are the vocabulary Last Known Book uses to explain what actually happened. ### Judge objective Close sponsor-native-necessity Q&A before it is asked. --- ## 2:15–2:30 — Ecosystem value + truth boundary ### Picture Show a clean excerpt/title from `docs/SDK-FEEDBACK.md`, then return to the product. ### Voice > The build also produced concrete SDK and documentation feedback around market identity, bounded approvals, raw-unit arithmetic and PostOnly event reconciliation. This is real testnet behavior evidence — not a claim of production reliability, ROI or incident prevalence. ### Judge objective Business/ecosystem value without fake metrics. --- ## 2:30–2:35 — Close ### Picture Wordmark + hero product frame. ### Voice > **Last Known Book. Reconstruct the market your agent actually traded.** ## Required screen text The encoded film must make these strings legible at least once: - `LAST KNOWN BOOK` - `AGENT INTENT ≠ VENUE REALITY` - `MINT_A_PAIR + CHAIN_INDEXER_DIVERGENCE` - `WRITE REFUSED` - `REAL SHANNON PROOF` - `0 FILLS` - `OrderPlaced → OrderRested → OrderCancelled` - `tUSDC 1 → 1 RAW` - `TESTNET BEHAVIOR PROOF ≠ PRODUCTION EVIDENCE` ## Editing constraints - Do not open with a technology-stack montage. - Do not spend more time on architecture than on behavior. - Do not introduce a second narrative or product name. - Keep captured replay and product-originated Packet 003 proof visually distinct. - Avoid generic crypto-trading B-roll, dark dashboard montage or decorative blockchain imagery. - Every technical detail shown must answer a likely judge objection. - Prefer screen capture + purposeful annotations over slides. ## TRACE final encoded-video gate Before the video is considered complete, TRACE must review the **encoded artifact**, not only this script, for: - first-view clarity; - first-30-second problem/mechanism comprehension; - first-minute aha; - text readability at normal playback speed; - replay-vs-real-proof truth separation; - claim fidelity; - audio intelligibility; - visual continuity with the judge-facing product; - final duration within 2–3 minutes. `SCRIPT_READY != VIDEO_READY`.