release: verification 19.97% per record / 33.69% per product - #152
Open
Seungpyo1007 wants to merge 14 commits into
Open
release: verification 19.97% per record / 33.69% per product#152Seungpyo1007 wants to merge 14 commits into
Seungpyo1007 wants to merge 14 commits into
Conversation
Every change has been landing straight on main, which is also the branch that publishes the site and tells TechEngine to pick up new data — so any merge was effectively a release. Adds develop as the integration branch and keeps main for released state. - validate-data now also runs on pushes to develop, so integration is checked the same way main is - bump-engine lands the submodule bump on develop instead of pushing to main. It switches branch *before* updating the submodule, so the change is staged on the branch that will carry it rather than being committed and then orphaned - README documents the model and points contributors at develop deploy-pages and notify-engine deliberately stay on main: publishing and the engine handshake should happen on release, not on every integration merge.
Brings the verification status aggregate that verify-status pushed to main after develop was branched. Under git-flow anything that lands on main is merged back so develop never starts from stale state.
84% of GPU records carried only a kaggle.com URL. The verification layer reads that as tier 3, which keeps a record yellow no matter how complete it is — GPU sits at 12% green while CPU, which picked up Wikipedia citations during the spec-fill batches, is at 65%. Wikipedia's per-vendor GPU lists document these same cards, so where a row could be matched AND a spec we already hold agreed with it, the page is added as a source. 1,006 rows parsed, 716 matched a record, 663 confirmed; 145 of those already had a tier-1 source, leaving 481 to cite. The spec agreement is the whole point: it makes the citation a claim about this card rather than a same-name lookalike. Confirmation came from base_clock_mhz (372) or tdp_w (146); the 53 rows that matched by name but agreed on neither were skipped rather than cited. No field values are changed — only source_urls gains an entry. Refs #1
Regenerates site/public/v1/gpus for the 481 records changed in the previous commit, with the engine at the pinned submodule commit. Only pages whose content actually differs are committed; 3,580 timestamp-only pages are left alone. Refs #1
Auto-promotions from the verification layer (green+live-source or crossref-confirm). Each flip is verified:false->true only; see data/_verify/ledger.jsonl. Refs #1
Auto-promotions from the verification layer (green+live-source or crossref-confirm). Each flip is verified:false->true only; see data/_verify/ledger.jsonl. Refs #1
Auto-promotions from the verification layer (green+live-source or crossref-confirm). Each flip is verified:false->true only; see data/_verify/ledger.jsonl. Refs #1
113 Qualcomm Snapdragon records carried a "YYYY-01-01" date invented at import time from the year of a device that used the chip, so it ran late and was precise to nothing. Each now carries the announcement date Wikipedia's Snapdragon SoC list states for that exact model, and cites the page. Only the model's own designation is matched, whole-token: "Snapdragon 8" never claims the 8 Gen 2, and a record naming two vendors is skipped as a mangled import. Sourced dates land 1-3 years earlier than the placeholder for 62 of the 113. Exynos, Kirin and the MediaTek list were dropped for want of the data, not the parsing: they mention 202/127/594 models between them and carry 7/6/1 dates. The Apple silicon page has dates but a row block mixes the launch dates of every product a chip shipped in, which dated the A15 Bionic to 2022-11 (really 2021-09), so it is not trusted either. Refs #1
Auto-promotions from the verification layer (green+live-source or crossref-confirm). Each flip is verified:false->true only; see data/_verify/ledger.jsonl. Refs #1
Most smartphone records are regional and RAM variants of one phone, and no source documents an individual SKU, so the per-record figure reads lower than how well the products in the dataset are actually sourced. The homepage now shows both readings of the same verifications, the per-product one as a secondary line under the headline. The block removes itself when the snapshot predates the field rather than rendering a zero. Refs #1
github-actions
Bot
added
ci
Member
🔎 Data verification — Tier 0 (offline existence/trust)Scored by Changed records in this PR6550 record(s) scored. %%{init: {"theme":"base","themeVariables":{"pie1":"#3fb950","pie2":"#d29922","pie3":"#f85149","pieStrokeWidth":"0px","pieOpacity":"1"}}}%%
pie showData
title Verification bands — all records
"Green" : 6360
"Yellow" : 190
"Red" : 0
Full-dataset baseline105742 record(s) scored. %%{init: {"theme":"base","themeVariables":{"pie1":"#3fb950","pie2":"#d29922","pie3":"#f85149","pieStrokeWidth":"0px","pieOpacity":"1"}}}%%
pie showData
title Verification bands — all records
"Green" : 21110
"Yellow" : 83431
"Red" : 1201
Hard violations (forced red):
green = authoritative source + complete + consistent · yellow = plausible, needs confirmation · red = sparse/weak source or a hard contradiction. Promotion to |
Member
TechEngine change review: PASS
Changed data
Changed record examplessoc modified
smartphone modified
tablet modified
watch modified
pda modified
gpu modified
cpu modified
Heuristic review
Changed site
Modified site files
|
Member
TechEngine validation stats: PASS
Data summary
Warning Tracked verified coverage is below 50% for watch 4.2% (18/433), tablet 6.3% (218/3455), soc 6.9% (146/2104), brand 11.6% (24/207), smartphone 18.6% (17384/93396), all 20.0% (21114/105742), gpu 32.7% (663/2030), pda 47.1% (66/140). Validation notes
Key output:
Homepage build: |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Release of the 14 commits integrated on
developsince the git-flow switch. Merging deploys the site and notifies TechEngine.Verification
The headline figure has been stale on the site since 2026-07-31:
verify-statuswas computing the aggregate correctly every run, then failing ongit push HEAD:mainafter the default branch moved to develop, sostatus.jsonnever updated. Fixed in GetTechAPI/TechEngine#56; the snapshot on develop is now current.The per-product figure is the same verifications counted per phone rather than per SKU — 57,216 smartphone records stand for 20,594 models, and no source documents an individual regional/RAM configuration. Both readings are now shown (GetTechAPI/TechEngine#55 + #151).
What is in this release
Verification — 14.07% → 19.97%. The bulk came from fixing a false consistency signal:
soc_not_after_devicecompared device and SoC years exactly, but 92.9% of SoC records carry a placeholderYYYY-01-01date that runs ~2 years late, so it was failing 5,488 sound phones (TechEngine#51 → 5,853 promotions). Also 481 Wikipedia citations on GPU records, 113 Snapdragon release dates replaced with sourced announcement dates, and a fix for rate-limit answers being cached as dead links (TechEngine#52, #53).Data — two weekly CPU ingests, 113 corrected SoC dates + refreshed
socsdump pages, refreshedgpusdump pages.Process — git-flow adopted; every automation now targets
develop(promotion PRs, weekly ingest, status refresh), leavingmainto move only through a release PR like this one.Closes #1