Matt Shumer (HyperWrite) has a documented streak of viral AI demos: supplied snippets show him claiming Claude Opus 5 one-shotted a fully custom-code game with no external assets, and separately asking Opus 5.5 for an animated short film entirely in code. The game claim was partially audited — an independent panel scored it 3.5–5/10 against the original, and reporting revealed the 'one-shot' actually used 'fan out sub-agents' with iterative visual-QA loops against reference footage rather than a literal single pass. The specific 277k-gate computer-in-code artifact at the center of this case is not directly covered in the supplied snippets — it arrives only via Reddit echoes — while surrounding coverage is dominated by backlash to Shumer's viral 'Something Big Is Happening' essay, with Gary Marcus and others citing an earlier model claim of his that didn't replicate. Net: the snippets establish a claimant with contested credibility and an escalating pattern of one-shot demos, but not the content, provenance, or standing of this particular computer demo.
Converges with his model-plus-harness benchmark unit: the audit of Shumer's earlier 'one-shot' already reattributed a model claim to fan-out sub-agents with iterative visual-QA loops, so this same-claimant claim — so far arriving only via Reddit echoes — is a live attribution test with an inspectable artifact whose gate-level machine is a hard external oracle (the gates either run the game or they don't). Either resolution feeds load-bearing positions: a literal single pass would be a dated capability marker for one-shot large-system generation, while a repeat of the harness pattern is a receipt for his discussed-is-not-deployed discipline and verification-loop anatomy applied to viral capability claims.
ip:concept.model-plus-harness-benchmark-unitip:framework.discussed-is-not-deployedip:source.subagents-speed-accuracy-ebookip:concept.verification-loopsradar:opus55-hwe-riscv-beats-vexriscvradar:opus55-video-as-coderadar:schema-arc-agi-3-claim
queries asked of Scott's wikis
- one-shot generation as frontier capability marker
- verifying model capability claims through artifact replication
- sub-agent fan-out and QA-loop orchestration in coding harnesses
- model vs harness credit attribution
- LLM-generated hardware simulators logic gates CPU
- frontier coding benchmark escalation and hype filtering
now 0 pts/hpeak 414 pts/hcomments 0/hpeers p33momentum: steady3 platformsage 326h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
2026-10-09T18:14:34Z
New HN evidence (50014384) is a genuine Opus 5.5 one-shot demo (13-second cartoon from single frame) but it is vastly simpler than the 277k-gate computer claim and drew near-zero engagement; it corroborates the one-shot substrate while leaving Shumer's specific artifact and provenance question entirely untouched. Measured heat confirms quiescence (0.0 pts/h, 30th percentile, 279h age). The case remains a cold, unverified single-source claim whose hard gate-level oracle still awaits its first independent inspection ~11 days in.
2026-10-09T04:52:45Z
evidence attached: hn.story.50014384 — Independent Opus 5.5 one-shot coding demo (cartoon from single frame) corroborates the case's hypothesis about one-shot generation of complex code artifacts
2026-10-09T02:37:29Z
The new 93k-line Godot game is genre periphery, not claim evidence: it is explicitly iterative (3 weeks, Blender MCP, 73 test suites) and drew near-zero engagement, so it corroborates the Opus 5.5 large-artifact substrate while leaving the one-shot provenance question and the gate-level artifact exactly where they were. The case's meaning is unchanged — a cold, unverified single-source claim whose hard external oracle still awaits its first inspection ~11 days in; the accumulating pattern is now several independent iterative builds with zero follow-up on the one claim that actually matters.
2026-10-08T23:06:43Z
evidence attached: reddit.post.1x0x713 — Substantial artifact — 93k lines, 417 Blender-generated models, full game in 3 weeks with Opus 5.5 — corroborating frontier-model coding-agent capability for large multi-layer systems.
2026-10-02T09:35:03Z
The sensor re-fired on the cooling tail of the sibling Elden Ring port; the wave is now fully quiescent (aggregate 0.0 pts/h, 17.6th peer percentile, focal thread up only 8 points in ~2 days), confirming last cycle's read that the 'loud spread' was one sibling object on one platform. The case's meaning is unchanged — a cold, unverified single-source claim whose hard gate-level oracle still awaits its first independent inspection — and per the prior pre-commitment these sibling velocity spikes force no full repricing and no material change.
2026-09-30T20:11:51Z
The wave doubled again — Tobyn Jacobs' Elden Ring port to 3016 pts/310 comments — but it is the same sibling object, now past peak and cooling, and ~30h of elevated genre attention has failed to pull a single inspection onto the gate-level artifact; the focal thread remains frozen. The case's meaning reverts to a cold, unverified single-source claim whose realistic resolution path is a late independent audit à la Shumer's Opus 5 game. Heat cools to low: the magnitude valve's loud-spread reading decomposes to one cooling Reddit object on one platform, and the periphery added no new objects, communities, or outlets this cycle — further velocity spikes on the sibling should not force full reprices.
2026-09-29T22:33:31Z
The case's evidence set now rides a live wave — the attached Elden Ring port doubled to 1088 pts/132 comments at ~94th percentile within the hour — but the loudness is genre heat, not claim movement: the focal thread is frozen at 171/35, and still nothing inspects the gate-level artifact or touches one-shot provenance. Heat rises to medium because the periphery is genuinely expanding and a viral same-model moment is the environment most likely to pull an inspection onto the artifact; it stops short of high because the exploding object is a sibling claimant's iterative build, not spread of this demo.
2026-09-29T20:53:11Z
evidence attached: reddit.post.1wtiubh — A widely echoed second large whole-system Opus 5.5 artifact (full game port) corroborates the frontier coding-capability-marker genre, though not one-shot.
2026-09-29T11:04:30Z
magnitude valve eligible (multi-platform, top-decile engagement) and never alerted; deterministic escalation to deliver
2026-09-28T14:17:53Z
Second-party evidence arrived: an independent, unverified Reddit report of a large multi-layer CAD kernel built iteratively on Opus 5.5 in 72h — the general large-system-generation capability now has a claimed second instance, raising the prior that Shumer's artifact is real while sharpening the live question to provenance (literal one-shot vs his documented fan-out/QA-loop pattern) and to verification of the gate-level oracle; the case moves seed→watching on that second line. Heat cools to low because the demo's own thread is in past-peak decay (~9 pts/h vs ~57 peak) with no inspections, derivatives, or spread of the demo itself, and the magnitude valve is not eligible.
2026-09-28T13:35:16Z
evidence attached: reddit.post.1wsc1nv — Another claimed large multi-layer system build on Opus 5.5 (CAD kernel in 72h), additional evidence for the frontier large-system generation capability marker that case tracks.
2026-09-28T09:49:15Z
No material development: the velocity flag is a 4.5x multiple on a low peer baseline, while the thread itself added one point and zero comments — the measured line (peak ~49 pts/h, now ~10, 77th percentile) reads as past-peak plateau, so I discount the 'accelerating' label and hold medium. The case's meaning is unchanged: still one claimant-pattern test awaiting the first independent inspection of the published gate-level artifact; new comments are jokes and an unsupported fraud assertion plus one commenter's parallel (non-replicating) machine project.
2026-09-28T02:33:08Z
grounded: converges/high — Converges with his model-plus-harness benchmark unit: the audit of Shumer's earlier 'one-shot' already reattributed a model claim to fan-out sub-agents with ite
2026-09-28T02:24:48Z
case created — A distinct first-party capability claim with a published inspectable artifact — not covered by the open HWE-benchmark or video-as-code Opus 5.5 cases — whose confirmation or debunking will resolve through inspection and replication attempts.