[zriyansh]

What this adds

Five new /compare posts. Each comes with a quiz, its images and a social card. No existing post is changed.

Post	URL	Page date
Snowflake Alternatives and Competitors in 2026	/compare/snowflake-alternatives/	15 Oct 2026
Best Data Streaming Platforms in 2026, Compared	/compare/best-data-streaming-platforms/	16 Oct 2026
Best Big Data Tools in 2026, Sorted by Job	/compare/big-data-tools/	17 Oct 2026
Best Data Orchestration Tools in 2026, Compared	/compare/data-orchestration-tools/	18 Oct 2026
Best ETL Tools for Snowflake: 10 Options, Real Costs	/compare/snowflake-etl-tools/	20 Oct 2026

• Snowflake alternatives: eight options (Databricks, BigQuery, Redshift, Microsoft Fabric, ClickHouse, MotherDuck, Dremio, Athena) compared on cost model, limits and Iceberg support, plus how OLake Go works with Snowflake (reads through external catalogs, and writes Iceberg tables into Snowflake through the Horizon catalog from v0.11.0).
• Data streaming platforms: Kafka, Confluent, Amazon MSK, Redpanda, WarpStream, Kinesis, Azure Event Hubs, Google Pub/Sub, Pulsar and Striim, with dated prices, limits and what each says about Iceberg.
• Big data tools: eleven tools sorted by job: Hadoop, Amazon S3, Cassandra, MongoDB, Kafka, Spark, Flink, Trino, BigQuery, Snowflake and Databricks.
• Data orchestration tools: ten tools (Airflow, Dagster, Prefect, Kestra, Astronomer, Amazon MWAA, Google Cloud Composer, AWS Step Functions, Azure Data Factory, Databricks Lakeflow Jobs) with pricing and limits, and where OLake Go's own scheduling fits.
• Snowflake ETL tools: ten ways to load data into Snowflake with real costs, including OLake Go.

Review changes (after the first review)

• Screenshots of other vendors' pages are removed. The facts are dated text with a link to the vendor page, ending (checked Oct 2026). Our own diagrams and covers stay.
• Every competitor limit, price and licence claim links its vendor page.
• Negative-only and sharp review-site quotes are removed. Remaining quotes keep their date and link, and each section keeps the "one person's view" line.
• The Apache 2.0 mentions link the OLake repository, and the Kafka Upsert version requirement now matches the docs.
• SOP section 5.7 records these rules.

Notes

• Vendor facts come from each vendor's own pages with the date read. User quotes link to their source.
• Each post has a quiz (src/data/quizzes/) and a social card, with the cover entries added to scripts/og-images.manifest.json.
• Add five posts and update post dates #585 and Add five posts: MongoDB, Oracle, CDC #605 are merged, and this branch has master merged in.

[shubham19may]

Builds and typechecks clean, content only, no concerns on the code side. One editorial point before merge, and it applies to this PR and #605 (same comment there), plus the posts already live.

Review-site quotes. Across these ten posts there are 79 links to Hacker News and PeerSpot, roughly one "What users say" quote per competitor, almost all negative (cost, reliability, support). The nofollow plugin handles the link side. The content side is what I want changed:

1. Quote opinions and first-hand pricing experience only. Never an unverified factual defect ("it loses data", "it does not support X") from an anonymous commenter. If a defect matters, cite the vendor's own docs or issue tracker instead.
2. Balance: for each tool either one positive and one negative quote, or none. A page that quotes only the negatives about every competitor, with OLake underneath, reads as a hit piece, and it gets screenshotted when we later want a partnership or a joint customer with that vendor.
3. Drop the sharpest ones. The kind of line I mean, already live: "Fivetran is convenient but absurdly expensive".
4. Keep the date and the link on every quote, as you already do, and keep the "one person's view" line.

Please apply 1–3 to the ten posts in #604 and #605 before merge, and add the rule to the SOP so it holds for the next batch. The already-published posts can be swept in a follow-up PR.

[shubham19may]

Please remove the third-party screenshots before merge. Decision from our side: we do not want captures of other vendors' pages on olake.io. A screenshot of a vendor's landing, pricing, licence or docs page is a copy of their copyrighted page, including logo and design, and the remedy on their side is a copyright notice to us. We'd rather not carry that exposure across a hundred images, and pricing captures go stale the day the vendor changes a number while staying live on our domain.

What to do in this PR and in #605 (same comment there):

• Remove every image that is a capture of a third-party page: vendor landing pages, pricing pages and plan tables, licence pages, docs pages, limits pages, console views, Hacker News threads. By your naming that is roughly every <post>--<vendor>-*.webp file plus the landing-*, *-pricing*, *-plans*, *-limits*, *-licence*, *-free-tib*, *-rpu-price* files. In this PR that is around 40 of the 61 non-cover images.
• Replace each with the same information as text: a dated sentence or a small table with the price or limit and a link to the vendor page. The claim keeps its source, and we can edit it in a minute when the vendor changes it.
• Keep your own diagrams (*-which-fits, *-layers, *-three-routes, *-starts-vs-moves and the like) and any screenshot of OLake's own UI or docs. Those are ours.
• Remove the deleted images' entries from the image sitemap inputs if any, and check the build has no broken image warnings.

Please also add this to the SOP as a rule, and open a follow-up PR that sweeps the posts already live in /compare and /learn (I count roughly a hundred such captures there). That follow-up does not block this merge, but it should land this week.

[shubham19may]

Follow-up to the screenshot removal: please add a Claude Code skill that refreshes the pricing and limits claims monthly. Once prices and limits live as dated text with a source link, staleness becomes a maintenance job we can automate instead of a risk we carry.

What we want, in this repo (separate small PR is fine, not blocking #604/#605):

• A skill under .claude/skills/refresh-vendor-claims/SKILL.md (name is yours). Trigger: /refresh-vendor-claims, optionally with a post path or a vendor name to scope it.
• What it does: walk compare/ and learn/ posts, find every dated price, plan, limit or licence claim that carries a vendor link (the sentences and tables that replace the screenshots), fetch each linked page, compare, and propose the edit: new number, new date, same link. Where the page moved or the claim is no longer verifiable, flag it instead of guessing. Output one PR per run with a short table of what changed and what it could not verify.
• Make the claims findable: agree a consistent pattern for them, e.g. a (checked Oct 2026) suffix or a small <Checked date="2026-10-22" src="..."> component, so the skill does not have to guess which sentences are claims. The SOP should say every price or limit uses that pattern.
• A short section in CLAUDE.md (repo root) telling Claude what the pattern is, which folders are in scope, and to run the skill before publishing a new comparison post as well.
• Cadence: a GitHub Actions cron on the 1st of each month that opens an issue titled "Refresh vendor claims, ", assigned to you, so the run is not forgotten. The skill itself runs locally in Claude Code and opens the PR.

Please also add to the SOP that any new price or limit claim must use the pattern from day one.

[shubham19may]

Addendum with specifics, from a pass over the five posts against the docs.

OLake claims: all supported by the docs, nothing contradicted. Two nits: the Apache 2.0 licence mention in snowflake-alternatives.mdx:414 and big-data-tools.mdx:497 has no link (link the GitHub repo as the other posts do); best-data-streaming-platforms.mdx:498 ties the Go 0.12.1 / UI 0.7.1 requirement to Kafka null-value deletes only, the docs give it for Upsert mode as a whole.

Quotes to drop or replace under rule 3 above:

• big-data-tools.mdx:178 Cassandra "an operational nightmare"; :420 Snowflake "charges an arm and leg"; :454 Databricks "crazy expensive"
• snowflake-alternatives.mdx:227 Fabric "unreliable, unergonomic, and surprisingly expensive"; :292 MotherDuck "way too expensive"
• data-orchestration-tools.mdx:300 MWAA "the instance prices are ridiculous for what you get"
• snowflake-etl-tools.mdx:492 "Snowflake police"
• best-data-streaming-platforms.mdx:334 Event Hubs "almost no documentation and Java libraries that mostly do not run"

Unlinked competitor limits: about 60 lines across these five posts say "its docs say" or "per the X page" with no hyperlink (e.g. snowflake-alternatives.mdx:122,158,257,324; best-data-streaming-platforms.mdx:154,187,218,402; big-data-tools.mdx:340,374,451; snowflake-etl-tools.mdx:187,286,398,433). I know the ten-links-per-post cap is deliberate. A factual limit about a competitor is the one place the link is not optional. Either link it or cut the claim.

[zriyansh]

Review points applied on this branch (commit c8581d0, then master merged):

• Quotes: negative-only and sharp quotes removed (the ones you named included); remaining quotes are dated, linked and balanced.
• Screenshots: all 43 images of other vendors' pages removed and replaced by dated text with a link. Our own diagrams and covers remain.
• Unlinked competitor limits are linked (about 430 lines across the five posts), with (checked Oct 2026) on price, plan, limit and licence claims.
• Apache 2.0 mentions link the repo; the Kafka Upsert requirement (OLake Go 0.12.1, UI 0.7.1) is stated for Upsert mode as a whole.
• SOP section 5.7 records the quote, screenshot and claim rules.

Follow-ups: #609 applies the same changes to the posts already live (including #605's), #610 adds the refresh skill and the monthly reminder.

[zriyansh]

Re: review-site quotes. Done on this branch (commits c8581d0, 2148875). Each tool keeps one positive and one negative quote, or none; negative-only and positive-only quotes are removed, and so are the sharp ones, including the lines you named (Cassandra, Snowflake, Databricks, Fabric, MotherDuck, MWAA, Event Hubs, 'Snowflake police'). No quote states an unverified defect as fact. Every kept quote has its date and link, and each section keeps the 'one person's view' line. The rule is in SOP 5.7.

[zriyansh]

Re: third-party screenshots. Done. All 43 captures of other vendors' pages (landing, pricing, plans, licence, docs, limits, forum threads) are removed, and OLake's own home-page capture too, since it shows customer logos. Each fact is now a dated sentence or small table with a link and (checked Oct 2026). The branch adds only the five covers and nine of our own diagrams. The image sitemap is generated from the build, so it needed no edit, and the build has no broken-image warnings.

[zriyansh]

Re: unlinked competitor limits and the OLake nits. About 430 claim lines across the five posts now carry the vendor link; where a limit could not be sourced it was cut. Apache 2.0 mentions of OLake link the repo; the Kafka Upsert note (OLake Go 0.12.1, UI 0.7.1) now says it applies to Upsert mode as a whole. An independent read of every post against the live vendor pages found a few wrong figures (Confluent connector count, Databricks Lakeflow plan prices, ClickHouse prices, Snowflake table date); they are corrected in 2148875 and 7dbcfd7. Also fixed: FAQ answers are JSX so their links render (plain strings showed raw markdown).

[zriyansh]

Re: the refresh skill and SOP. SOP section 5.7 is in this PR (quotes, screenshots, the claim pattern, FAQ links as JSX). The skill, repo CLAUDE.md and the monthly reminder are in #610, which should merge after this one. #609 applies the same changes to the posts already live.