Methodology — read before quoting any number
1. What this site is
Smart Glasses Data is a public data ledger for smart glasses: privacy cards for every model, verified specs, search interest, and the analyzed coverage of each model. One fact, one home: release tracking lives on the companion site aiglassesdata.com; real-user experience records live in the UX Ledger (uxledger.com) and are linked, not duplicated here.
2. Sources
Vendor product pages and announcements (Meta, Snap, Samsung/Google), regulator reports (Hamburg HmbBfDI, EDPB), and launch coverage. Every card, event and spec row carries a source link. Where a claim rests on secondary coverage rather than a vendor page, the note says so.
3. Not stated is a real value
If a vendor has not published a figure, or it has not been verified against a citable source, the cell says Not stated. No number on this page is estimated, interpolated or filled in from a similar product. Averages are computed over non-null values only; null counts are shown next to them.
4. Privacy heatmap rule
Codes are derived from the card text by keyword rules — “Not stated” → 0; “on by default” → 3; “No continuous recording” → 1; LED/light present → 2; “not been independently verified” → 1; “On-device processing” → 1; “Meta servers” → 2. The full card text is shown directly below the map, so any code can be checked against its sentence. Read the map as a map, not a score: a red cell means “on by default — check the card”, not “bad product”.
5. Trends
Google Trends index points (0–100), not search counts. A 24-hour average is a noisy single-day snapshot; the weekly figure is the stable read. The two screenshots (Canada and global, Nov 5, 2026) are the source.
6. Articles and analysis
Coverage is collected from Hacker News, Bing News, tech blogs, YouTube and Reddit (via Bing News). Google News RSS returns 503 from Worker egress, so Bing News is the news funnel; direct Reddit RSS is 403 from the same egress. Each article is analyzed once by Workers AI (Llama 3.1 8B, @cf/meta/llama-3.1-8b-instruct-fp8): a 1–3 paragraph summary plus structured fields. Analysis is based on the title, snippet and retrievable body excerpt; where the full text cannot be fetched it is marked “snippet”. Anything undeterminable is null — nothing is invented. A story only enters when its headline names a tracked model (title gate: Display first, then Ray-Ban, Oakley, Galaxy Glasses, Specs). The gate is imperfect: one known miss is a 9to5Google article about the Motorola Signature 27 phone filed under snap-specs; its battery figure is excluded from the Snap battery row with a note, and stays in the raw ledger for audit.
7. The scoring rule — what the indexes are and are not
The positive sentiment index (0–100), the recommendation index (0–100), and the author verdict (buy / wait / skip / mixed, with the author’s own quoted words) are kept, by a decision of the site owner on 2026-11-07. They measure the article content and its author’s own stance — how positive this piece reads, how strongly it leans toward recommending, and what its author concluded. They do not measure the product, and they are not this site’s rating of any product. This site never rates products: there is no site score, no ranking, and no “best” pick anywhere on it. That is the line: indexes that describe coverage are kept and shown with their method; any number that would present as our judgement of a product is not published. Where a ranking appears on the site it is the source’s own (e.g. a Google Trends index), and the page says so.
8. The three ledgers
All three are generated directly from the per-article analyses — not from experience_records, which is 0 rows (real-user experience lives in the UX Ledger).
- Claims vs measured (/api/ledger/claims): every key_claim carrying a measured_value, listed with its source article. measured_value is the number as stated in that article, not an independent lab measurement. Rows flagged “tested” come from hands-on/owner pieces or articles stating a usage duration; the rest restate a vendor or announcement figure. The battery comparison on /metrics/battery/ puts the vendor claim from the specs table next to these stated values.
- Scores (/api/ledger/scores): sentiment and recommendation distributions (buckets 0–19 / 20–39 / 40–59 / 60–79 / 80–100), plain-mean averages over non-null values, and verdict counts per model and site-wide. Articles with no verdict are counted as “no verdict”, never assigned one.
- Failure modes (/api/ledger/failures): issues[] grouped by exact issue string within each model. Near-duplicates are kept separate and shown as written, not merged. Counts are mentions across analyzed articles.
9. Evidence checks
Two deterministic checks run on every analysis before it is stored, independent of the model’s wording. First, compared_products keeps only names that appear verbatim in the evidence text (title + snippet + fetched body); compares_with is recomputed from the surviving list, never trusted from the model. Second, a verdict_quote is kept only if its normalized text appears in the normalized evidence text; otherwise it is stored as null. These checks only affect new analyses; older analyses that predate them may still carry invented items and are cleaned on re-analysis.
10. Live data / snapshot
Numbers come from the API (D1) when it is reachable. If the API cannot be reached, the page renders the frozen snapshot generated from the same database at build time and says so in the header. The snapshot is never presented as live. Charts are rendered live from the API or snapshot data (ECharts). No static screenshot charts. The raw data table sits under each chart.
11. Corrections log
- 2026-11-07 — Meta Ray-Ban Display battery. Before: “Up to 6 hours (per buying guide; Meta figure not yet confirmed)”. After: “Up to 6 hours per charge (mixed use); charging case adds up to 24 hours; Meta Neural Band up to 18 hours”. Source: Meta Help Center, https://www.meta.com/help/ai-glasses/1116302600085081/. The older “with case 30 hours total” wording is superseded and is not used.
- 2026-11-07 — Oakley Meta HSTN microphone. Before: not listed (sources conflicted). After: “5-mic array (2 per arm + 1 near nose pad)”. Sources: Abt and Android Central. The “dual beam-forming” wording is not used. The specs table gained a microphone column; Ray-Ban Meta Gen 3 shows “6-mic array” (per this site’s own privacy card) and the remaining models show Not stated.
12. Limitations
- Analysis is AI extraction from titles, snippets and partial bodies; most analyses are snippet-based where full text could not be fetched. Treat every extracted number as “as stated in the article”, not as a verified fact.
- The title gate misses and mis-files occasionally (see §6, Motorola case). Mis-filed articles are excluded from the affected comparison with a note, not silently deleted.
- experience_records is 0 rows on this site by design; claimed-vs-measured here is coverage-derived, not lab-measured. Independent measurement would need the UX Ledger’s experience records.
- Search interest is two Google Trends screenshots (Nov 5, 2026), not a continuous series.
- Charts are rendered in the browser; if the chart library fails to load, the data table underneath carries the same numbers.
13. Past, present, future — three complete segments; the past is the baseline
The data on this site comes in three segments, and all three are needed: the past (what has already happened), the present (the current value) and the future (each change to come). Never start from a present zero and wait only for the future to change: the past and the present are the baseline against which a change has meaning. Without a baseline, change means nothing. Three rules. First, every dimension is backfilled — price history, specs and launch history, review and experience-record history, mention and event history; everything that can be taken from a public source must be taken. Second, the articles already collected are themselves history: the claimed values, measured values, issues and conclusions in this site's 233 analyzed articles are already structured and entered into the history ledger (see /history/) — we do not wait for future articles to begin. Third, zero is an acceptable starting point only when there truly is no history, neither in public sources nor in the collected articles, and that must be written on this page. Every future change is measured against the baseline so rebuilt: archiving and sentence-level diff start from that baseline. Current status: independent laboratory measurements (experience_records) remain at 0 rows — there is as yet no independent structured measurement record in public sources under our criteria; hands-on measured values cited in coverage are in the history ledger (marked "tested"), kept separate from independent measurements and never mixed with them.