Skip to main content

The SignalRecord

A monthly measurement of technology coverage

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.

Attention

Which Subjects and Fields drew the coverage?

Stage

Was it an Announcement, a Launch, Speculation, or a Reported Result?

Framing

Was the coverage Neutral, Promotional, Critical or Skeptical?

Change

Which subjects and fields Moved or Emerged against their own recent record?

These are the terms to watch, in two kinds. Outlined terms name a thing being counted or tested. Filled terms are values coverage is sorted into, and each carries the colour that value wears in its own section’s chart — so the chip and the bar match. Every figure also states its own legend.
How to read this issue

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.

The scale of the month — and what each thing counted is
29,256
Articles collected
An article is one published piece from one outlet, gathered from the standing panel during the month.
47,462
Coverage items classified
A coverage item is one distinct development inside an article. An article can yield none, one, or several.
138
Outlets
An outlet is a publication in the panel. These are the ones that carried at least one measured article.
10,611
Subjects measured
A subject is a company, product, person, technology, standard, organisation or event that coverage names.
59
Fields reported
A field is a subject area, such as cybersecurity. Subjects belong to one or more; fields group into clusters.
3,585
Reported results
A reported result is a coverage item describing something already completed and measured — not planned, announced or shipped.
August 2026 · the month in summary

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.

The month in four numbers
Attention
29.9%

share taken by the largest field cluster

Stage
1 in 13

items reporting a completed result

Framing
2.1 : 1

promotional to critical-plus-skeptical

Change
17 of 48

eligible fields breaking their own pattern

01

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.

Figure 1
Attention and reported-result share, by field cluster
August 2026 · full panel · ordered by share of coverage
SHARE OF THE MONTH'S COVERAGE REPORTED A RESULT 0 010% 5%20% 10%30% 15%Devices, hardware & robotics 29.9% 3.5%Platforms, internet & media 15.9% 3.5%Compute & infrastructure 13% 7.6%AI — applied & ecosystem 9.5% 8.5%Deep tech, industry & policy 7% 5.9%AI — core technology 6.7% 14.1%Security & trust 5.9% 5.8%Software & developer ecosystem 5.4% 6.7%Power, networks & space 3.6% 3.8%Fintech & web3 3.1% 11.7% ALL COVERAGE 6%
Share of coverage Reported-result share — top quartile of fields Reported-result share — middle half Reported-result share — bottom quartile All-coverage reported-result share
Read this as

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 →

02

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.

Figure 2
Every classified coverage item, by reporting status
August 2026 · full panel · 47,462 items · ordered by size
ANNOUNCEMENT · 60.9%LAUNCH · 27.7% REPORTED RESULT · 7.6% · 3,585 ITEMS SPECULATION · 3.8% · 1,825 ITEMS
Announcement — stated or demonstrated, not shippedLaunch — shipped or deployed, not yet measuredReported result — already completed and measuredSpeculation — reported but unconfirmed

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 →

03

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.

Figure 3
Promotional against critical and skeptical framing
August 2026 · share of each group's own classified coverage items
4% 4%8% 8%12% 12%0 ◀ CRITICAL + SKEPTICAL PROMOTIONAL ▶All sources 47,462 items 3.8% 8.1%
Promotional Critical + skeptical Neutral — 41,793 items (88.1%), excluded from this 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.

04

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.

Figure 4
Fields that moved, and subjects that emerged
August 2026 · every comparison is against the field's or subject's own twelve-month record
AFields whose reported-result share moved 17 of 48 eligible fields · descriptive movement, not a significance test · the 10 largest moves shown
08%16%24%SHARE OF THE FIELD'S COVERAGE THAT REPORTED A RESULTAI Inference & Serving 20.3%Energy & Power for Compute 4.1%AI in Healthcare & Life Sciences 11.6%Identity & Access Management 8.3%Databases & Data Infrastructure 3.2%AI Chips & Accelerators 13%Enterprise Software & SaaS 7.6%Open Source Software 8.2%Data Centers & AI Infrastructure 8.7%Robotics & Automation 6.5%
Twelve-month reported-result share August 2026 — rose August 2026 — fell
BSubjects that cleared all three emergence tests 24 this month · 60+ items · 5× their own twelve-month share · 5+ distinct outlets · the 12 furthest above their own baseline shown
Subject Coverage items · axis 0–398 ItemsOutletsAgainst 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×
Cleared all three tests — reported
One case, up close
AI Evaluation & Benchmarks — a busier field with a different result record
× 3.3
August 2026 volume against this field's own normal month
63.5% 53.6%
reported-result share, twelve-month record to August 2026

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.

Before interpreting any of this

What this record cannot establish

Coverage, not reality

Compass measures the technology press. It does not measure adoption, revenue, product quality or market performance.

Reported, not verified

A reported result is something coverage says has happened. Compass does not determine whether it is true or well evidenced.

A panel, not the internet

An absence means not observed in this measured panel. It is not proof that no reporting exists elsewhere.

From the whole field to your subject

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 Record
Research Record

Inspect 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.

Cumulative share of coverage, by subject rank · log scale
0%25%50%75%100% CUMULATIVE SHARE OF COVERAGE HALF THE MONTH'S COVERAGE BY RANK 500 4.5% 1 16.6% 10 22.7% 25 35% 100 54.5% 500 100% 10,611SUBJECTS, RANKED BY COVERAGE (LOG SCALE)
Cumulative share of the month's coverage Halfway reference
Cumulative concentration
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
What kind of subject drew coverage
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%
Most-covered subjects
SubjectSubject-itemsShare
Apple 3,339 4.48%
Google 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.

015%30%45%60%SHARE OF THE FIELD'S OWN COVERAGE THAT REPORTED A RESULT Q1 MEDIAN Q3 LARGEST FIELD BY COVERAGE — CONSUMER DEVICES & WEARABLES · 18,551 attributions
Top quartile Middle half Bottom quartile Interquartile range

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.

FieldAttributionsResults Share · axis 0–54%ShareBand
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.

What kind of story ran · share of classified items
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%
Story type most likely to carry a measured result
Largest fields · by attributions
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
In the top quartile for reported-result share
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.

04%8%12% 13-MONTH MEAN 6.3% OBSERVED RANGE 5.6% – 7.5% AUG 26 7.5%AUG 25NOV 25FEB 26MAY 26AUG 26
Matched-panel reported-result share Observed range Mean of the measured record
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.

0100200300400 COVERAGE ITEMS IN THE MONTH 081624DISTINCT OUTLETS CARRYING THE SUBJECT BREADTH GATE · 5 OUTLETS VOLUME GATE · 60 ITEMSCyberleekGLM-5.3ColdcardAstraPixel 11 ProPixel 11 Pro XLPixel 11Apple Watch Series 12Pixel 11 Pro FoldPixel Watch 5M5 UltraM6Ted LassoZ.ai
Cleared both gates — reported Failed the breadth gate — excluded 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.

Field12-monthThis monthReading
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.
Emerging subjects
SubjectItemsOutlets Share of the monthPrior 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%
Mac Studio 144 20 0.19% 0.03%
Mac mini 195 22 0.26% 0.04%
iPhone Ultra 173 10 0.23% 0.03%
GPT-5.6 Sol 71 23 0.1% 0.02%
Unitree 83 18 0.11% 0.02%
Rockstar Games 127 15 0.17% 0.03%

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.

MONTHLY VOLUME · ATTRIBUTIONS SHARE REPORTING A RESULT Normal month (12-month average) August 2026 Twelve-month record August 2026 42 140 ◀ GPT-5.6 SOL ▶ ◀ REST OF FIELD 63.5% 53.6% 075% 23 OUTLETS CARRIED THE LEAD SUBJECT; THE LARGEST OF THEM ACCOUNTED FOR 11.5%
Twelve-month reference Attributed to GPT-5.6 Sol Rest of the field
The source trail
The Register · 26 Aug 2026
OpenAI explains how its naughty AI agents attacked Hugging Face
stated or demonstrated, nothing yet measuredCompass record →

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.

FigureUnitPanelDrawn fromFilter DenominatorTotal
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.

The Signal Record · Issue 02 · August 2026 · Compass by Enginerds Attention · Stage · Framing · Change — every month, the same four questions