Growth ideas for PROB-0004, checked against evidence

IDEA-0009 to IDEA-0016 against what the repo, the logs and Firestore show. Numbers: raw/2026-11-02-cloud-logging-scans-and-quota-counts.md and raw/2026-10-31-rival-reviews-and-keywords.md.

Check this before any channel: photo scans may be failing

From 10-04 to 10-18, 22 of 23 photo scans succeeded. From 10-21 to 10-31, 1 of 18 did, and 16 answered "no food". Those images were also about a quarter of the earlier size, and many repeat.

• Could be harmless: the same images sent again look like tests of non-food, and 10-03 had 4 of the same answer. The one 200 on 10-30 shows the path can still work.
• Could be a break: the drop starts on 10-21, the day 1.4.2 was cut and days after the SDK 58 move on 10-17. A very small image is what a blank frame would send. Build 18 is what Play now serves.
• Not settled here: nothing here can take a photo on a phone. One real meal photo on the Play build answers it. If it fails, this is p0 and every idea below waits.

The main finding: most of the funnel can be read today

Step	Source	Needs a build?
Install	Play Console, by UTM	No
Tried a scan	count of trial_ docs in Firestore, 64 today	No
Paid	RevenueCat Charts	No
Third meal	nothing: the log stays on the phone	Yes

metrics/definitions.yml said no source counted the free-to-paid denominator. The count works and takes one query (source-firebase). It counts people who tried, so a failed first scan still counts.

Verdict per idea

Idea	Verdict	Why
0010 post on Play	First, after the scan check	Assets exist, no money, and the table above reads it
0009 PostHog	Narrow it, and do it second	It adds one step of four, at a real cost
0016 weigh and publish	Keep, owner's call	The only idea that can unlock the message; no thesis dishes yet
0013 switcher page	Later, unproven	Cheap, but no web search data exists
0012 languages, listing tests	Wait for traffic	No demand data, and a test needs installs
0011 Apple Search Ads	Wait, and fix the stop rule	App Store not live; the rule reads about 3 buyers
0015 lower price	Wait, needs ten times the traffic stated	About 10,600 first scans, not 1,000
0014 share card	Drop for now	No share count, cannot be judged

IDEA-0009, funnel

• Confirmed: no analytics package in apps/mobile/package.json.
• Cost the idea understates: the published privacy policy says "No analytics or tracking SDKs", and listing/listing.md declares no such data. Both change, on both stores.
• A build is slow here: metrics/flow.json has a median of 345 hours from release to store review, and the App Store has rejected this app at least three times.
• Do it once there are installs to read and the third-meal step is the open question.

IDEA-0010, post now

• Confirmed: images approved 2026-10-31, Shorts and build scripts exist, and tracked Play links are written. The search bar in images and Shorts still names both stores.
• Play's UTM report and the trial_ count give install to tried. App Store Campaigns hide a campaign under 5 installs, but Play is the only store this would use.
• The other apps' post results: no install numbers are saved in their raw/ folders, so this channel stays unproven.
• Take the trial_ count on the day of the first post. It already holds tests.

IDEA-0011, Apple Search Ads

• Earning back 30% of ₹25,000 is ₹7,500, about 3 yearly plans (at an assumed ₹88 to $1 and a 15% store fee).
• At the starting targets (60% × 3%), that takes about 167 installs, so ₹150 an install at most. Earning back all of it takes about ₹54 an install.
• 3 expected buyers is noise: about 1 run in 5 would show 0 or 1 by chance. The stop rule cannot tell a bad channel from bad luck. Set a floor on taps or installs first.
• The current cost per tap for these keywords is not known. Read Apple's own suggested bid range before approving the spend.

IDEA-0012, languages

• The app has no translation package, so Spanish listings lead to an English app.
• From memory, Play's code is es-419, not es-MX. Check before writing T029's step.
• The claim that these locales add search reach in the US is not checked. No download data by country exists yet to pick languages from.
• The English listing changed on 10-31, and another edit restarts re-indexing.

IDEA-0013, switcher page

• Rival reviews support it: being charged is 26 of 139 bad Cal AI iOS reviews and 22 of 172 on Android, and paying before use is 22 and 19. These are keyword counts.
• Those keyword scores are for store search. Web search demand is unmeasured.
• The site does have a page at /products/askcal/, so the home exists.

IDEA-0014, share card

• Confirmed: "Counted with AskCal" ships, in share-card.tsx. No share is counted anywhere.
• It costs a design and a store release for an effect that cannot be seen.

IDEA-0015, price

• $29.99 needs 33% more buyers just to tie $39.99: 3% must become 4%.
• Telling 3% from 4% at the usual 95% and 80% power takes about 5,300 first scans per price. The idea's 1,000 is about a tenth of that. Metered scans ran at about 73 a month, tests included.
• Whether the RevenueCat plan in use includes price tests is not checked.

IDEA-0016, weigh and publish

• Confirmed: T007 is still blocked on the owner, and the eval has only Nutrition5k dishes.
• It retries experiments/first-accuracy-gate-run on the food the product is for. That run could not settle the thesis, because its dataset could not answer the questions.
• It brings no installs alone. What it can change is the claim every post may make.

Limits

• Web search, web fetch and the mcp-appstore tools were not permitted in this run. So there is no fresh store ranking, rating, rival price or install band, and no outside benchmark. Keyword data is from 2026-10-31.
• The RevenueCat connector needs authorising, and Play Console and App Store Connect numbers are read by a person. Nothing here is a real install or payment count.
• App Store state is from ops/release.yml, read from the API on 10-31: waiting for review.
• All counts include the owner and possibly store reviewers. Play has been live for days at most, so none of this describes real users.