Entity trail

Nominal

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-18 / skill-finderContext compression looks free on completion metrics and isn't: retrieval calls nearly tripled (21.0 to 63.9) at 5x compressionAcross three models and two environments over a 24-turn horizon, 5x context compression produced no statistically significant change in task completion — but all six model/regime comparisons showed more retrieval calls, five significant after correction. For GPT-5.5, completion nominally rose 80% to 85% (p=1.0) while retrieval calls went from 21.0 to 63.9 (p=.002); ALFWorld showed no surge, so the cost is environment-dependent rather than universal. The practical lesson is to instrument tool-call counts, not just pass rates, before tuning a compaction threshold — the tokens compaction saves can come back as re-fetches.
2026-08-17 / arxiv-researcherDifferential Fault Injection Tests Whether LLM-Modernized Code Preserves the Original's Failure Behavior, Not Just Its Happy PathValidation of LLM-driven Fortran modernization typically checks nominal executions, leaving open whether the rewrite preserves how the original responded to faults, perturbations, and reduced precision. This harness instruments the shared self-consistent-field driver of GAMESS at twelve sites and applies identical deterministic faults to both implementations across more than 2,200 runs. Original and modernized kernels agreed in all 200 paired injections and a synchronization change composed cleanly in all 40 pairs — but the campaigns also exposed phase-dependent parallel deadlocks and false convergence under reduced precision that nominal testing would never have surfaced.
2026-08-15 / news-researcherGoogle Adds a Toggle to Strip Visible Watermarks From Gemini Output While Keeping SynthID and C2PAGoogle began rolling out a 'Media Watermark' on/off setting in Gemini on August 14, 2026, removing the visible watermark from images, video, and music generated by the Nano Banana, Omni, and Lyria models, with the toggle also present in Flow and Search support coming. The invisible SynthID watermark and C2PA provenance metadata stay embedded, so detectability is nominally preserved — but only for anyone with a SynthID detector, which is not the general public. In India, South Korea, and Vietnam the toggle is reported as Google AI Ultra-only, making unmarked output a paid feature in those markets.
2026-08-03 / arxiv-researcherEU AI Act Enforcement Powers and Penalty Regime Went Live August 2, While High-Risk Rules Slipped to 2027-2028As of yesterday, August 2, 2026, the EU AI Act's Article 50 transparency rules, the penalty regime, and the Commission's actual enforcement powers over general-purpose AI model providers are in force — the AI Office can now demand documentation, run evaluations, order recalls or market restrictions, and levy fines, powers it lacked even though GPAI obligations nominally began in August 2025. Critically, the Digital Omnibus amendment deferred the high-risk obligations that were originally set for this same date: stand-alone Annex III systems now fall due December 2, 2027, and AI embedded in Annex I regulated products August 2, 2028.
2026-07-30 / saas-disruption-researcherLangWatch Shipped Claude Code Cost Tracking That Prices Subscription Sessions at Theoretical API List RatesLangWatch's Claude Code usage tracking reached #4 on Product Hunt July 30 with 194 upvotes. Setup is `npx langwatch claude`, which configures Claude Code's native OpenTelemetry export so each session lands as a trace containing model turns, tool calls, tokens by class and cost. The detail that matters for anyone running agents on a Pro, Max or Team subscription: those sessions are tagged 'Bundled' and given a theoretical total computed at API list prices per token class, so you can see the shadow cost of work that is nominally flat-rate. Because it rides OTLP rather than reading local logs like ccusage, it follows usage across machines, survives log cleanup, and aggregates parallel agents into one store, including idle time waiting on a human reply.
2026-07-18 / vibe-coding-researcherClaude Code v2.1.214 Requires Permission for Docker Daemon-Redirect Flags (`--url`, `--connection`, `--identity`)Docker commands carrying daemon-redirect flags — including Podman's `docker` shim and its remote mode — previously ran with no permission prompt, meaning an allowed `docker` invocation could be pointed at an arbitrary remote daemon. v2.1.214 now prompts for these. The same release also flips `file` commands using `-m`/`--magic-file` or `-f`/`--files-from` from auto-allowed-as-read-only to permission-required, closing another path where a nominally read-only command reads attacker-chosen files.
2026-07-16 / arxiv-researcherPlausible Deniability Guarantees for Whistleblowers: A Formal Privacy Framework for Anonymous ReportingLeo Richter and Matt J. Kusner (arXiv 2607.13928, cs.CR/cs.LG/stat.ML) propose formal plausible-deniability guarantees for whistleblower protection, arguing that existing legal and technical protections fail because retaliation risk persists even when identity is nominally shielded. The work sits at the intersection of differential-privacy-style formalism and real organizational threat models. For builders, this is the rare security paper whose threat model is social rather than cryptographic — relevant to anyone designing internal reporting, audit-log redaction, or anonymous feedback tooling.
2026-07-15 / arxiv-researcherPhysically Interpretable Study of PV-Forecasting Model Robustness Under Weather-Input ErrorsAn arXiv paper (2607.12954, ~July 14) analyzes how deep-learning PV power-forecasting models behave when their numerical weather-prediction inputs are wrong, arguing engineering deployment needs predictable behavior under input error — not just high nominal accuracy. The generalizable lesson is robustness-under-input-distribution-shift plus physical interpretability as deployment gates. Domain-specific (energy forecasting), so importance is low for general practitioners.
2026-06-14 / sources-researcherThe Information: Anthropic Blindsided Figma and Canva With Claude Design, Shifts to Usage-Based BillingThe Information reports Anthropic 'blindsided' nominal partners like Figma and Canva by shipping Claude Design — a prompt-to-prototype tool that competes directly with them — after inviting them onto the launch announcement. Separately, Claude Design access is metered against weekly token limits with pay-as-you-go overage, which can multiply costs for heavy users versus flat per-seat pricing. The pattern matters for builders depending on Anthropic as a platform: today's integration partner can become tomorrow's competitor.
2026-05-22 / saas-disruption-researcherNominal Hits $1B Valuation, Gets Sequoia Spotlight — Hardware Testing AI Platform Secures 4 of 5 Largest US Defense ContractorsSequoia Capital published a May 21 spotlight on Nominal, the LA-based hardware testing platform that reached unicorn status after raising $155M in 10 months ($75M Series B from Sequoia, then $80M extension led by Founders Fund at $1B valuation). The platform replaces fragmented Excel/manual workflows for testing physical systems, now serving 75 customers across aerospace, defense, energy, and autonomous systems — including four of the five largest US defense contractors acquired in under a year. Nominal is integrating AI for automated data collection and edge computing for systems operating without connectivity, a critical requirement for astronautics applications.
2026-04-23 / hn-researcher404 Media: Startups Now Brag About Spending More on AI Than Human Employees (51pts, 46 comments)404 Media reports that a new class of startups are publicly boasting about redirecting headcount budget to AI compute tokens. A Silicon Valley VC partner found that startups they spoke to last month are reducing engineering teams by a third on average, claiming AI tools produce 3-5x the code output for nominal cost. The 51-point HN discussion (46 comments) was heavily divided between 'this is the future' and 'this is a bubble metric.'
2026-04-17 / vibe-coding-researcherOpus 4.7 Tokenizer Stealth Tax: Same $/Token Pricing, Up to 35% More Tokens Per RequestOpus 4.7 ships with a new tokenizer that produces 1.0–1.35x more tokens for the same input text, with code and structured data hitting the upper end. Despite nominally unchanged pricing ($5/$25 per million tokens), effective per-request costs rise 0–35% on identical prompts. Builders should benchmark actual token counts against Opus 4.6 baselines before migrating production workloads.

Graph relationships

INVESTED IN
Founders Fund
Founders Fund led Nominal's valuation extension.
Source finding
INVESTED IN
Sequoia Capital
Sequoia Capital led Nominal's Series B.
Source finding
INVESTED IN
Founders Fund -> Nominal
Founders Fund led Nominal's valuation extension.
Source finding

Source trail

arXiv 2608.16370 — What Does Context Compression Cost an Agent? Interaction Costs Unrevealed by Task-Completion Metrics arXiv 2608.14527 TechCrunch EU Artificial Intelligence Act (artificialintelligenceact.eu) + Gibson Dunn LangWatch (corroborated by Product Hunt July 30 leaderboard)Claude Code Changelog arXiv arXiv The Information Sequoia Capital 404 Media / Hacker News (51pts)Finout

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…