feat: staleness sweep across the most visible verified programs - #70
Merged
Conversation
Round two of extraction, spending the last 80 free credits on the
verified programs ranked most prominently — the set where a wrong figure
costs the most, because verified is what the rankings weight and what an
agent trusts.
Ten programs checked. One correction, six confirmations, three left for
a human. Every outcome is recorded with quoted page text in
data/commission-candidates.json.
The correction is the largest error found so far:
digitalocean 50% -> 10%. The page says "10% commission each month,
for an entire year". The registry claimed five times the
real rate on a program the homepage rankings feature.
Confirmed current, last_verified_at bumped to 2026-08-06: convertkit
(50% first year, page adds a 10-20% indefinite tail the summary elides),
surfer-seo (multi-tier CPA matches; the extractor returned null numerics
because no single figure exists, which is correct behaviour), cal-com,
krisp exact matches, apify and instantly within their stated ranges.
Left alone on purpose:
grammarly page advertises the program but publishes no figure; 30%
can be neither confirmed nor denied from public pages.
jasper every known affiliate path redirects to the homepage —
the program page is gone. 25% has nothing to check
against.
copy-ai same shape; 45% unverifiable.
Those three stay in the candidates file as conflicts. Empty over wrong,
and unverifiable over guessed.
Running tally across both rounds: 16 verified programs checked, 4 were
stale or wrong (canva closed, customgpt 50->15-20%, descript 15%->$25,
digitalocean 50->10%), 3 unverifiable. A quarter of the checked verified
set was wrong — the April verification pass has aged out, and the
verified badge was quietly borrowing credibility it no longer had.
tsc 0, lint 0 errors, build 884 pages, and the corrected figure renders
on the page, the API and the .md twin with no trace of the old 50%.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
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
1 participant
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.
Round two, spending the last 80 free Context.dev credits on the verified programs ranked most prominently — the set where a wrong figure costs the most, because verified is what the rankings weight and what an agent trusts.
Ten checked: 1 correction, 6 confirmations, 3 for a human
The correction is the largest error found so far:
Five times the real rate, on a program the homepage rankings feature.
Confirmed current (bumped to
last_verified_at: 2026-08-06): convertkit — 50% first year, and the page adds a 10–20% indefinite tail by tier that the registry summary elides; surfer-seo — multi-tier CPA matches, and the extractor returned null numerics because no single figure exists, which is correct behaviour; cal-com and krisp exact matches; apify and instantly within their stated ranges.Left alone on purpose:
grammarly— page advertises the program, publishes no figure. 30% can be neither confirmed nor denied.jasper— every affiliate path redirects to the homepage. The program page is gone; 25% has nothing to check against.copy-ai— same shape; 45% unverifiable.All three sit in
data/commission-candidates.jsonas conflicts. Empty over wrong, and unverifiable over guessed.Running tally across both rounds
16 verified programs checked → 4 stale or wrong (canva closed, customgpt 50→15–20%, descript 15%→$25, digitalocean 50→10%), 3 unverifiable.
A quarter of the checked verified set was wrong. The April verification pass has aged out — the badge was quietly borrowing credibility it no longer had. The remaining ~33 verified programs deserve the same sweep; that needs the paid tier (~330 credits).
Verification
🤖 Generated with Claude Code