The Signal Record · Issue 02 · August 2026 ▾
The SignalRecord
Technology news tells you what was reported. The Signal Record measures the coverage itself: where attention went, what stage the reported developments had reached, how the coverage was framed, and what departed from its recent pattern.
This is not a news roundup, a technology ranking, or a verdict on which claims are true. It is a measurement of the public record created by the technology press.
Each month, Compass classifies coverage from a standing panel of approximately 140 technology publications. It turns that coverage into a structured, source-linked record for people who need to interpret technology narratives without mistaking the volume of attention for the stage of what was reported.
Which Subjects and Fields drew the coverage?
Was it an Announcement, a Launch, Speculation, or a Reported Result?
Was the coverage Neutral, Promotional, Critical or Skeptical?
Which subjects and fields Moved or Emerged against their own recent record?
Four sections, one per question, each answered for the whole month with one principal chart. The charts carry the numbers; the text says what each one shows and what follows from it. Open the Research Record for the supporting exhibits, full tables, source trail, exclusions and method.
Most of the month’s coverage reported an ‘announcement’; about one item in 13 reported a ‘completed result’.
Attention and completed results were not strongly related this month — the largest fields were neither the richest nor the thinnest in completed results. What the month does show clearly is its composition, and how far apart the fields sit inside it. Figure 1 shows the whole pattern in a single view.
share taken by the largest field cluster
items reporting a completed result
promotional to critical-plus-skeptical
eligible fields breaking their own pattern
Attention · Where August 2026's coverage went
Coverage concentrates in a few areas and spreads very thinly across everything else.
Counted in field attributions: one coverage item counts once for each subject it names in a field.
Figure 1 places every one of Compass’s ten field clusters on two scales side by side: on the left, the share of the month’s coverage it drew; on the right, how much of that coverage described a completed result. Clusters are ordered by share, largest first, so the two panels can be read across.
The two panels do not line up in either direction this month. How much of a cluster’s coverage described a completed result was largely independent of how much coverage it drew.
Beneath that structure the month is long-tailed: a few hundred subjects carry about half the coverage, and most of the rest were named only a handful of times. The cumulative concentration curve, the subject-type mix and the most-covered table are in the Research Record. Open the Research Record →
Stage · What status the coverage reported
Compass records each classified item as one of four reporting statuses.
Counted in coverage items: 47,462 classified developments from 29,256 articles.
Figure 2 lays the month’s whole classified corpus out by reporting status, in proportion. The four statuses are mutually exclusive rather than a sequence — speculation can precede or follow a launch — so the bar is ordered by size, not by maturity. Only the last of them describes something the coverage reported as already completed and measured.
That single proportion conceals most of what matters, because it varies between fields by more than an order of magnitude. The distribution across every eligible field, the full field ladder, the story-type mix and the matched-panel trend are in the Research Record. Open the Research Record →
Framing · The register the month was written in
Where coverage was not neutral, the balance between promotional and critical framing is the month’s register.
Counted in coverage items: 47,462 classified developments.
Figure 3 sets promotional framing against critical and skeptical framing combined, across all sources. Bars extend right for promotional and left for critical-plus-skeptical from a shared centre. Neutral items, the large majority, are omitted from the comparison.
Source roles are gated. Compass has no measured role for a publication. The split depends on a curated outlet map reviewed by hand, and until each outlet measured in this month carries a reviewed role, this record will not print a role comparison it cannot stand behind.
Register describes how coverage is framed. It says nothing about whether the underlying claim is true.
Change · What broke from the recent pattern
Compass reports a change only when a field or subject departs from its own recent record, never from a leaderboard.
Fields counted in attributions; subjects counted in coverage items.
Figure 4 answers both halves of the question. Panel A is a slope chart: each row is a field whose reported-result share moved against its own twelve-month record, hollow marker to filled. Panel B lists the subjects whose coverage cleared all three emergence tests, ranked by how far each ran above its own baseline. 31 of the 48 eligible fields moved less than the screen’s threshold and are not shown. Every candidate that cleared the volume and multiple tests also cleared source breadth.
| Subject | Coverage items · axis 0–398 | Items | Outlets | Against its own baseline |
|---|---|---|---|---|
| Cyberleek | 67 | 8 | no prior record | |
| GLM-5.3 | 60 | 10 | no prior record | |
| Coldcard | 83 | 7 | 77× | |
| Astra | 66 | 19 | 55× | |
| Pixel 11 Pro | 183 | 16 | 44× | |
| Pixel 11 Pro XL | 92 | 15 | 38× | |
| Pixel 11 | 398 | 20 | 24× | |
| Apple Watch Series 12 | 61 | 9 | 23× | |
| Pixel 11 Pro Fold | 101 | 16 | 23× | |
| Pixel Watch 5 | 145 | 18 | 23× | |
| M5 Ultra | 67 | 15 | 19× | |
| M6 | 115 | 18 | 18× |
In August 2026, coverage of GPT-5.6 Sol accounted for most of this field’s month, and that coverage sat at announcement and launch stage while its reported-result share fell. 23 outlets carried that coverage and none accounted for more than 11.5% of it, so this is not a panel artifact — it is a field-level reading moving because a single subject moved inside it. That is why Compass reports fields and subjects as separate layers. The volume-and-share chart and the full source trail are in the Research Record.
What this record cannot establish
Compass measures the technology press. It does not measure adoption, revenue, product quality or market performance.
A reported result is something coverage says has happened. Compass does not determine whether it is true or well evidenced.
An absence means not observed in this measured panel. It is not proof that no reporting exists elsewhere.
Compass measures one company, product, person or standard the same way — updated daily.
The same four questions, answered for anything you follow: where its attention came from, what stage it had reached, how it was framed, and what changed.
Look up any topic — free Open the Research RecordInspect the complete issue
Every supporting exhibit, table, exclusion, panel definition and method rule behind the four sections above.
Attention · concentration, subject types and the most-covered subjects
10,611 subjects drew measurable coverage, producing 74,539 subject-items. Attributed to fields those become 83,165 field attributions, because a subject can sit in more than one field. This census counts every measured subject, including cryptoasset subjects, which Compass's interactive dashboards screen out.
| Top 1 subject | 4.5% |
| Top 10 | 16.6% |
| Top 25 | 22.7% |
| Top 100 | 35% |
| Top 500 | 54.5% |
| Top 10,611 | 100% |
| Subjects with 100+ items | 75 |
| Subjects with under 10 items | 9,429 |
| Company | 30,494 | 40.9% |
| Product | 21,845 | 29.3% |
| Technology | 7,481 | 10% |
| Organization | 7,367 | 9.9% |
| Person | 2,641 | 3.5% |
| Model | 2,424 | 3.3% |
| Event | 1,336 | 1.8% |
| Regulation | 493 | 0.7% |
| Standard | 433 | 0.6% |
| Topic | 25 | <0.1% |
| Subject | Subject-items | Share |
|---|---|---|
| Apple | 3,339 | 4.48% |
| 1,709 | 2.29% | |
| OpenAI | 1,402 | 1.88% |
| Amazon | 1,273 | 1.71% |
| Microsoft | 925 | 1.24% |
| Anthropic | 923 | 1.24% |
| Nvidia | 845 | 1.13% |
| Meta | 794 | 1.07% |
| Samsung | 710 | 0.95% |
| SpaceX | 472 | 0.63% |
| Bitcoin | 422 | 0.57% |
| NASA | 406 | 0.54% |
| Pixel 11 | 398 | 0.53% |
Stage · how far apart the fields sit
One dot for each field that cleared the sample floor, positioned by its own reported-result share, with the interquartile box and median drawn behind them. Bands are quartiles of the 48 eligible fields, recomputed each issue — not fixed editorial thresholds.
Half the eligible fields fall between 4.7% and 8.6%, but the full range runs from 1.6% to 53.6%. How much of the public record describes something completed depends far more on which field you read than on how busy the month was.
Stage · all 59 measured fields, ranked by reported-result share
Bands are quartiles of the 48 eligible fields, recomputed each issue. 11 fields drew too little coverage to clear the publication floor: shown here, never used in a finding.
| Field | Attributions | Results | Share · axis 0–54% | Share | Band |
|---|---|---|---|---|---|
| AI Evaluation & Benchmarks | 140 | 75 | 53.6% | Top quartile | |
| AI Inference & Serving | 256 | 52 | 20.3% | Top quartile | |
| Foundation Models & LLMs | 4,102 | 562 | 13.7% | Top quartile | |
| AI Chips & Accelerators | 1,729 | 225 | 13% | Top quartile | |
| Blockchain & Crypto | 1,739 | 214 | 12.3% | Top quartile | |
| AI in Healthcare & Life Sciences | 129 | 15 | 11.6% | Top quartile | |
| Quantum Computing | 132 | 14 | 10.6% | Top quartile | |
| AI for Software Development | 1,186 | 124 | 10.5% | Top quartile | |
| Fintech & Payments | 825 | 86 | 10.4% | Top quartile | |
| Semiconductors & Fabrication | 1,516 | 150 | 9.9% | Top quartile | |
| Memory & Storage | 693 | 64 | 9.2% | Top quartile | |
| Data Centers & AI Infrastructure | 496 | 43 | 8.7% | Top quartile | |
| Chip Design & EDA | 151 | 13 | 8.6% | Middle half | |
| AI Agents & Autonomy | 594 | 51 | 8.6% | Middle half | |
| Enterprise AI Adoption | 2,667 | 228 | 8.5% | Middle half | |
| AI Developer Tooling & MLOps | 466 | 39 | 8.4% | Middle half | |
| Identity & Access Management | 168 | 14 | 8.3% | Middle half | |
| Open Source Software | 1,420 | 116 | 8.2% | Middle half | |
| Enterprise Software & SaaS | 794 | 60 | 7.6% | Middle half | |
| Supply Chain & Manufacturing Tech | 294 | 22 | 7.5% | Middle half | |
| AI Policy & Governance | 2,998 | 224 | 7.5% | Middle half | |
| AI Assistants & Chatbots | 3,296 | 246 | 7.5% | Middle half | |
| Generative & Multimodal AI | 314 | 23 | 7.3% | Middle half | |
| Robotics & Automation | 414 | 27 | 6.5% | Middle half | |
| Cybersecurity | 3,289 | 209 | 6.4% | Middle half | |
| Autonomous Vehicles & Mobility | 594 | 35 | 5.9% | Middle half | |
| Startups & Venture Capital | 576 | 33 | 5.7% | Middle half | |
| Clean Energy & Climate Tech | 406 | 23 | 5.7% | Middle half | |
| Social Media & Creator Economy | 1,071 | 60 | 5.6% | Middle half | |
| Developer Tools & Platforms | 1,491 | 83 | 5.6% | Middle half | |
| DevOps & Platform Engineering | 385 | 21 | 5.5% | Middle half | |
| Cloud Computing & Platforms | 6,265 | 327 | 5.2% | Middle half | |
| Programming Languages & Frameworks | 237 | 12 | 5.1% | Middle half | |
| Privacy & Data Protection | 851 | 43 | 5.1% | Middle half | |
| Gaming & Interactive Entertainment | 2,956 | 148 | 5% | Middle half | |
| Networking & Connectivity | 985 | 49 | 5% | Middle half | |
| Energy & Power for Compute | 148 | 6 | 4.1% | Bottom quartile | |
| E-commerce & Retail Tech | 924 | 37 | 4% | Bottom quartile | |
| Consumer Devices & Wearables | 18,551 | 668 | 3.6% | Bottom quartile | |
| Space & Satellite Systems | 1,850 | 59 | 3.2% | Bottom quartile | |
| Databases & Data Infrastructure | 158 | 5 | 3.2% | Bottom quartile | |
| AR/VR & Spatial Computing | 4,754 | 136 | 2.9% | Bottom quartile | |
| Ransomware & Cyber Threats | 504 | 14 | 2.8% | Bottom quartile | |
| Streaming & Digital Media | 2,480 | 66 | 2.7% | Bottom quartile | |
| Operating Systems & Platforms | 5,721 | 148 | 2.6% | Bottom quartile | |
| Tech Regulation & Antitrust | 920 | 19 | 2.1% | Bottom quartile | |
| Defense & Aerospace Tech | 497 | 10 | 2% | Bottom quartile | |
| IoT & Edge Devices | 546 | 9 | 1.6% | Bottom quartile | |
| Cloud Cost & FinOps | 2 | 1 | 50% | Below floor | |
| On-Device & Edge AI | 14 | 6 | 42.9% | Below floor | |
| AI Safety & Alignment | 63 | 10 | 15.9% | Below floor | |
| Data Engineering & Analytics | 47 | 6 | 12.8% | Below floor | |
| AI Training & Model Development | 26 | 3 | 11.5% | Below floor | |
| AI Search & Answer Engines | 118 | 12 | 10.2% | Below floor | |
| Retrieval, RAG & Vector Search | 65 | 6 | 9.2% | Below floor | |
| Search & Digital Advertising | 32 | 2 | 6.3% | Below floor | |
| Cryptography & Post-Quantum | 59 | 3 | 5.1% | Below floor | |
| Security Regulation & Compliance | 30 | 1 | 3.3% | Below floor | |
| AI Data, Copyright & Licensing | 31 | 0 | 0% | Below floor |
Stage · what kind of story ran, and the largest fields by volume
Story type and field are independent classifications: a story type describes what the coverage was about, a field describes the subject area it was attributed to.
| Report | 51.9% | |
| Launch | 29.2% | |
| Regulation | 4.2% | |
| Benchmark | 3.9% | |
| Funding | 3.4% | |
| Partnership | 3.4% | |
| Lawsuit | 1.6% | |
| M&A | 1.2% | |
| Restructuring | 1% |
| Consumer Devices & Wearables | 18,551 | |
| Cloud Computing & Platforms | 6,265 | |
| Operating Systems & Platforms | 5,721 | |
| AR/VR & Spatial Computing | 4,754 | |
| Foundation Models & LLMs | 4,102 | |
| AI Assistants & Chatbots | 3,296 | |
| Cybersecurity | 3,289 | |
| AI Policy & Governance | 2,998 | |
| Gaming & Interactive Entertainment | 2,956 | |
| Enterprise AI Adoption | 2,667 | |
| Streaming & Digital Media | 2,480 | |
| Space & Satellite Systems | 1,850 |
Stage · the two panels and the matched-panel trend
Full panel — every outlet measured in the month: 138 outlets, 47,462 classified coverage items, 3,585 reported results. It carries the month's cross-section. Matched panel — the 102 outlets read in every month since August 2025: 36,278 items this month. It carries all month-to-month comparison. The two answer different questions and are never compared with each other.
Change · how a subject qualifies as emerging, and which candidates failed
Every candidate at five times its own twelve-month share or more, plotted against the two gates it must also clear: enough coverage in the month, and coverage from enough separate outlets. The shaded region is the rejection zone.
Change · every mover and every emerging subject
Movement is measured against each field's own twelve-month reported-result share. Language stays descriptive until a formal field-variance test is implemented and documented.
| Field | 12-month | This month | Reading |
|---|---|---|---|
| AI Inference & Serving | 13% | 20.3% | Rose into the top quartile, on a smaller base. 52 of 256. |
| Energy & Power for Compute | 11.2% | 4.1% | Fell into the bottom quartile, on a smaller base — fell by more than half. 6 of 148. |
| AI in Healthcare & Life Sciences | 6.4% | 11.6% | Rose into the top quartile, on a smaller base. 15 of 129. |
| Identity & Access Management | 3.2% | 8.3% | Rose within the middle half, on a smaller base — more than doubled. 14 of 168. |
| Databases & Data Infrastructure | 8.2% | 3.2% | Fell into the bottom quartile, on a smaller base — fell by more than half. 5 of 158. |
| AI Chips & Accelerators | 8.6% | 13% | Rose into the top quartile, on a smaller base. 225 of 1,729. |
| Enterprise Software & SaaS | 3.4% | 7.6% | Rose within the middle half, on a smaller base — more than doubled. 60 of 794. |
| Open Source Software | 4.7% | 8.2% | Rose within the middle half, on a smaller base. 116 of 1,420. |
| Data Centers & AI Infrastructure | 5.3% | 8.7% | Rose into the top quartile, on a smaller base. 43 of 496. |
| Robotics & Automation | 3.3% | 6.5% | Rose within the middle half, on a smaller base. 27 of 414. |
| AI Policy & Governance | 4.3% | 7.5% | Rose within the middle half, on a large base. 224 of 2,998. |
| AI Agents & Autonomy | 5.6% | 8.6% | Rose within the middle half, on a smaller base. 51 of 594. |
| Supply Chain & Manufacturing Tech | 4.5% | 7.5% | Rose within the middle half, on a smaller base. 22 of 294. |
| Generative & Multimodal AI | 4.8% | 7.3% | Rose within the middle half, on a smaller base. 23 of 314. |
| Social Media & Creator Economy | 3.5% | 5.6% | Rose within the middle half, on a smaller base. 60 of 1,071. |
| Cloud Computing & Platforms | 3.4% | 5.2% | Rose within the middle half, on a large base. 327 of 6,265. |
| E-commerce & Retail Tech | 2.3% | 4% | Rose into the bottom quartile, on a smaller base. 37 of 924. |
| Subject | Items | Outlets | Share of the month | Prior monthly share | Against baseline |
|---|---|---|---|---|---|
| Cyberleek | 67 | 8 | 0.09% | — | no prior record |
| GLM-5.3 | 60 | 10 | 0.08% | — | no prior record |
| Coldcard | 83 | 7 | 0.11% | <0.01% | 77× |
| Astra | 66 | 19 | 0.09% | <0.01% | 55× |
| Pixel 11 Pro | 183 | 16 | 0.25% | <0.01% | 44× |
| Pixel 11 Pro XL | 92 | 15 | 0.12% | <0.01% | 38× |
| Pixel 11 | 398 | 20 | 0.53% | 0.02% | 24× |
| Apple Watch Series 12 | 61 | 9 | 0.08% | <0.01% | 23× |
| Pixel 11 Pro Fold | 101 | 16 | 0.14% | <0.01% | 23× |
| Pixel Watch 5 | 145 | 18 | 0.19% | <0.01% | 23× |
| M5 Ultra | 67 | 15 | 0.09% | <0.01% | 19× |
| M6 | 115 | 18 | 0.15% | <0.01% | 18× |
| Ted Lasso | 102 | 10 | 0.14% | 0.01% | 13× |
| Z.ai | 66 | 17 | 0.09% | <0.01% | 13× |
| OpenRouter | 65 | 16 | 0.09% | <0.01% | 12× |
| Grand Theft Auto VI | 295 | 19 | 0.4% | 0.03% | 11× |
| Flock | 141 | 17 | 0.19% | 0.02% | 10× |
| CXMT | 103 | 18 | 0.14% | 0.02% | 7× |
| Mac Studio | 144 | 20 | 0.19% | 0.03% | 7× |
| Mac mini | 195 | 22 | 0.26% | 0.04% | 7× |
| iPhone Ultra | 173 | 10 | 0.23% | 0.03% | 7× |
| GPT-5.6 Sol | 71 | 23 | 0.1% | 0.02% | 6× |
| Unitree | 83 | 18 | 0.11% | 0.02% | 6× |
| Rockstar Games | 127 | 15 | 0.17% | 0.03% | 5× |
The qualifying multiple is a share-over-share test: this month's share of all subject-items against the subject's own mean share across its prior twelve months. Because both sides are shares of their own month, a month that simply carried more coverage overall cannot manufacture an emergence on its own.
Change · the AI Evaluation & Benchmarks case, with its source trail
AI Evaluation & Benchmarks drew 140 attributions against a twelve-month average of 42, and its reported-result share moved from 63.5% to 53.6%. 71 of the 140 concerned GPT-5.6 Sol.
Every cited item links to the source article and to its Compass record.
How every figure in this issue was produced
For each figure: the unit it counts, the panel it was drawn on, the measurement set it came from, the filter that selected the rows, and the denominator it was divided by. A figure can be reconciled against this table without recomputing anything.
| Figure | Unit | Panel | Drawn from | Filter | Denominator | Total |
|---|---|---|---|---|---|---|
| Figure 1 · Attention and reported-result share | field attributions | full | the monthly coverage measure, joined to the subject-to-field map | the August 2026 period, with each subject folded into every field it belongs to | the month's field attributions | 83,165 |
| Figure 2 · Reporting status | coverage items | full | the classified coverage corpus | articles published in August 2026, one record per classified development | the month's classified coverage items | 47,462 |
| Figure 3 · Framing | coverage items | full | the classified coverage corpus | articles published in August 2026, framing recorded per development | each group's own classified coverage items | 47,462 |
| Figure 4 · Movement and emergence | field attributions (fields) · coverage items (subjects) | full | the monthly coverage measure, month by month | each field or subject against its own twelve months before the reported one | the field's or subject's own record | 83,165 |
Method version 1.0 · corpus as of 3 September 2026
Method, limits and publication rules
The panel. A standing panel of technology publications: major newsrooms, the specialist press, aggregation desks, and labs and platforms in their own words. August 2026 collected 29,256 articles from 138 of them.
The units. An article yields zero or more coverage items — one per distinct development Compass can classify. Each item is attributed to the subjects it names (subject-items) and, through them, to fields (field attributions). §01 counts attributions and subject-items; §02 and §03 count coverage items. Every figure names its unit.
A reported result. An item qualifies when the coverage describes something already completed: a benchmark run, a measured behaviour, a decided case, a published figure. A plan, preview or claim about the future does not. A launch is a fact but not a measured one, and is counted separately. Compass does not assess whether a reported result is true or well evidenced.
Status, not sequence. The four reporting statuses are mutually exclusive labels for what a given item reported. They are not a maturity ladder: speculation can precede or follow a launch, and an item is never promoted from one status to another after the fact.
Two panels. The full panel carries each month's cross-section. The matched panel — the 102 outlets read in every month since August 2025 — carries the longitudinal series. Reported separately, never compared with each other.
Floors and bands. A field needs 120 attributions in the month before its reading is used in a finding. Bands are monthly quartiles of the eligible fields, not fixed editorial thresholds.
Movement. A formal field-variance rule is specified but not yet implemented. Until it is, movement is reported descriptively against a screen of at least one percentage point and at least half the field's own twelve-month rate.
Emergence. Three tests: 60+ coverage items in the month, five times the subject's own twelve-month monthly share, and coverage from at least five distinct outlets. The breadth test exists because a subject can appear to surge when one publication enters the panel. Candidates that fail it are named and excluded.
Publication gate. An issue is released only after its month is fully classified. Articles collected, classified and reported results are counted and labelled separately. A partial month is held, not published with a provisional denominator.
Exclusions. Cryptoasset price and market coverage is screened from public rails at story level. Machine-translated republications are deduplicated, so one story published in eight languages counts once.
Not measured. Market adoption, revenue, product quality, the quality of a source's evidence, or whether any claim is true.
Who writes this. Nobody. Every sentence in the issue is templated from the measurements by a deterministic rule, and which template fires is decided by what the data supports. The same month regenerated from the same run produces the same words.
Citation, provenance and corrections
Suggested citation — Compass, “The Signal Record, Issue No. 02: August 2026,” Enginerds. Published 3 September 2026.
Measurement window 1–31 August 2026 · corpus as of 3 September 2026 · method version 1.0 · every cited item resolves to both the source article and its Compass record · corrections are published against the issue they affect.