1.9.79: every LTV read answers money as numbers and counts as integers by tracking202 · Pull Request #313 · tracking202/prosper202

tracking202 · 2026-10-28T10:58:10Z

Backlog: N9MJ

Stacked on #305 (1.9.79-ltv-customer-numbers, backlog RKE4): this branch starts from #305's head (82b51c8) and must merge after it. Until #305 merges, this PR's diff against 1.9.79 also shows #305's commit; the change here is the second commit, d58d841.

What changed and why

After #305 the customer reads answered numbers and the rest of /ltv still answered mysqli's text, so a subscription's amount was 29.99 in GET /ltv/customers/{id} and "29.99000" in GET /ltv/subscriptions (CLAUDE.md #28: a field's JSON type is part of the contract). I enumerated the reads from api/v3/index.php's /ltv read group (18 GET routes) and measured every answer on a real server. These still served text:

Read	Fields that were text
`GET /ltv/subscriptions`	`amount`, `mrr`, `customer_ltv`
`GET /ltv/breakdown` by campaign, ppc_account, landing_page	`total_revenue`, `avg_ltv` (`"44.562500000"`), `aov`, `repeat_rate` (`"0.5000"`), `mrr`, `spend`, `cac`, `ltv_cac`, and `total_orders` (`"4"`)
`GET /ltv/breakdown?by=product`	the amounts, `units` (`"1.500"`), `orders`
`GET /ltv/products`, `PATCH /ltv/products/{id}`, its delete's `dry_run` record	`price`
`GET /ltv/companies`	`total_revenue`, `mrr`, `order_count`, `last_activity_time`
`GET /ltv/abm`, `/ltv/abm/company`	`total_revenue`, `mrr`, `avg_time_on_page`/`avg_scroll_depth`/`avg_video_pct` (`"30.0000000"`), and abm/company's `last_activity` (a `GREATEST()` that MariaDB answers as text)

The summary, cohorts, mrr, predict, fields, webhooks, deliveries and integrations already answered numbers.

One helper. 1.9.79: the LTV customer reads answer money as numbers #305's private decimals() is now Prosper202\Ltv\LtvNumbers. It has decimals() for money and measures, and integers() for counts and unix times.
Where the casts are. Each read goes through it where its SQL is, as 1.9.79: the LTV customer reads answer money as numbers #305 did: in the repositories, and in the controller for products, because that SQL lives there.
Every column is named. Each read lists every money and count column, not only the ones MariaDB happens to answer as text, because MySQL and MariaDB type COALESCE()/GREATEST() differently.
What is never converted. integers() leaves a value with a fraction unchanged rather than truncating it. NULL stays null: a product with no price, a subscription whose customer row is gone. Names, skus, product keys and references made of digits stay strings.

docs/openapi.yaml now gives every /ltv read a typed components schema:

LtvSummary, LtvAcquisitionBreakdownRow, LtvProductBreakdownRow
LtvCohort, LtvMrr/LtvChurnInputs, LtvPrediction/LtvPredictionInputs/LtvPredictionByGroup/LtvPredictionGroup
LtvProduct, LtvCompanyRollup, LtvCompany, LtvSubscription
LtvAbmAccount, LtvAbmContact, LtvField, LtvWebhook, LtvWebhookDelivery, LtvIntegration, LtvNextOffer

Before this, all but the customer reads were additionalProperties: true, and GET /ltv/abm's list was documented as an object.

Tests: #305's type tests now cover every read

All in tests/Ltv/LtvDatabaseIntegrationTest.php, on a real server.

testEveryLtvReadServesNumbers

The reads it covers are the GET routes parsed out of api/v3/index.php's /ltv group. A route added there fails the test until it is held to the same rule.
Each answer is walked for a numeric string outside a text field.
The newly numeric values are checked exactly: units 1.5, a null price, a company with no customer, a sku and an account name of digits that stay strings.

testEveryLtvAnswerHasTheSpecsKeysAndTypes

Each answer, plus the product/company writes and dry runs, must have its schema's keys and JSON types.
The check recurses into every schema a property $refs. 1.9.79: the LTV customer reads answer money as numbers #305's spec reader learned {$ref: ...}, items: {$ref: ...} and required: [...], the last so optional keys like fallback_reason and options_invalid are allowed.
The spec's 200 answer for each path must name that schema.

Plants. Each plant below made a named test fail, and each was restored with a verified backup:

Plant	Caught by
Subscription `amount` cast removed	both tests
Product `price` cast removed	both tests
Company `order_count` cast removed	both tests
ABM `last_activity` cast removed	both tests
`LtvSubscription.amount` typed `string` in the spec	spec test
An undocumented key added to `LtvAbmAccount`	spec test (key mismatch)
`GET /ltv/abm`'s answer put back to `additionalProperties: true`	spec test (path does not name the schema)
A GET route added to the router group	both tests (route set mismatch)
All of this PR's code put back to #305's	both tests

AccountAdministrationIntegrationTest compared the PATCHed price as the string '12.50000'; it now asserts 12.5.

Readers swept by field name (#28)

Go structs and their fakes: no Go struct decodes these answers, and no LTV fake answers money as text. TestStructsDecodeAPIMoneyAsJSONNumber's money names now include customer_ltv, avg_ltv, aov, spend, cac and avg_revenue_per_customer; no struct field carries one.
MCP tools: the catalogue has no LTV tool.
Eval checks: ltv.json and the administration product case read amounts with tonumber, which reads both types.
@group instance suites, tests/live/: none read /ltv.
PHP pages: they cast every value they read (diffed below).
Go and PHP CLI tables: these change; see below.

Live, before (#305 head `82b51c8`) and after (this branch)

The licence stub is down. I installed with install.php's steps without the licence form, into p202_n9mj_live, and served php -S on 8163 with every proxy variable pointed at the closed port 127.0.0.1:9. The licence check failed open; nothing reached the licence service. Then:

I seeded with tests/fixtures/agent-eval/seed.sh (Go CLI).
I added LTV data over the API: two click conversions with customer refs at company Acme, revenue with line items, a refund, two subscriptions, a priced product, a second company, engagement events, a select field, an integration, and a webhook with a delivery inserted by SQL.

Results:

API answers (23 reads + the PATCH/dry-run answers): every leaf's JSON type changed exactly where the table above says and nowhere else. With numeric strings read as numbers on both sides, every answer has the same values before and after.
Pages: byte-identical before and after. These were sort_ltv.php by campaign, ppc_account, landing_page, product and abm; ltv_subscriptions.php, ltv_companies.php, ltv_company.php, ltv_products.php and ltv_customer.php, all signed in.
Go CLI tables (p202 ltv subscriptions|products|companies|abm|breakdown): values the same. An amount now shows to the cent (9.50 / 27.50 where they showed 9.5 / 27.5), a zero repeat_rate shows 0 (was 0.0000), and units shows 2 / 1.50 (was 2.000 / 1.500).
PHP CLI tables (ltv:subscriptions|companies|abm|breakdown): the decimal text's trailing zeros are gone (29.99 where they showed 29.99000, 13.75 where they showed 13.750000000).
Agent-eval LTV cases (ltv.json, 6 cases) and ltv-product-001, on the after instance with the reference agent: 0 fail; every case reports needs_judge, because without --judge-cmd the rubric is not graded. Their mechanical checks passed.
Not exercised live: the landing_page breakdown answered [] there, because the seed's clicks come through the direct-link tracker. The integration test covers that read with a real row.

Verification

.claude/skills/p202-verify/scripts/verify.sh --since origin/1.9.79 --changed (scratch DB p202_n9mj_schema):

Scope report
------------
  go         PASS  ran under CI's go1.25.14 via GOTOOLCHAIN (local go is 1.24)
  golangci   PASS  ran under CI's go1.25.14 via GOTOOLCHAIN (local go is 1.24)
  openapi    PASS
  patterns   PASS
  phpcs      PASS
  phpstan    PASS
  schema     PASS
  syntax     PASS
  unit       PASS
Not covered by this script:
  live      stand up an instance, drive the path a user would take
  agent     add and run a case under tests/fixtures/agent-eval/cases/
  ci        a workflow still in_progress has not passed; read the run

Unit: 6104 tests, 5 skipped. The fifth skip is CronEntryPointsFailLoudlyTest's "this checkout has a 202-config.php", from the live instance's config in the worktree.
Integration group, run by hand against a scratch DB:
- tests/Ltv/LtvDatabaseIntegrationTest.php: OK, 47 tests.
- tests/Api/V3/AccountAdministrationIntegrationTest.php: OK, 13.
- LtvBodyValuesIntegrationTest: OK, 10.
- NestedBodyValuesIntegrationTest: OK, 8.
Did not run: UpgradePendingAnswerServedIntegrationTest skipped itself (it needs a setup this environment lacks; it does not read /ltv).
CI: not read yet.

For a reviewer to check

MySQL 8 is not exercised here. I could only run MariaDB 10.11 locally, and MySQL 8 may type GREATEST/COALESCE differently. Every count and money column is named either way, and CI's integration job runs the new tests on MySQL 8.0.
1.9.79: the LTV customer reads answer money as numbers #305's private helper moved. MysqlCustomerCrmRepository::decimals() became LtvNumbers::decimals(), with the same body. If 1.9.79: the LTV customer reads answer money as numbers #305 is amended, the merge may need that rename carried.
I moved one docblock. MysqlCompanyRepository::listWithRollups()'s docblock was stranded above search(); I moved it onto listWithRollups().
Dry-run answers have no spec schema. The dry-run 200s (product/company delete previews) are left described by their text alone, as every other DELETE's dry run in the spec is; the test still holds their record to LtvProduct/LtvCompany.
Merge conflicts with the other LTV branches. git merge-tree against 1.9.79: LTV webhooks go through the export webhooks' guard and sender #278, 1.9.79: a dead LTV webhook can be revived, and is sent what it missed in the last 7 days #289 and 1.9.79: LTV Settings reads the row a button names exactly #296 (and also 1.9.79: the minutely cron sends LTV webhooks, within a budget #273, 1.9.79: the cron job sweeps lapsed LTV subscriptions hourly and runs the rest of LTV maintenance daily #285, 1.9.79: the subscription sweep announces what it writes, for the last week #302, 1.9.79: company linking walks past names it cannot link #304, 1.9.79: Customer LTV's pages read the record a button names exactly #308) conflicts only in changelogs.txt and 202-config/changelogs.json. None of them changes a read's columns.

Generated by Claude Code

GET /ltv/customers/{id} answered mysqli's DECIMAL strings beside integer ids: "total_revenue": "54.12500", "mrr": "0.00000", each subscription's amount and mrr, each revenue event's amount, and each line item's quantity ("1.500"), unit_price and amount. GET /ltv/customers did the same for the rollups, and the engagement read answered SUM()'s "1" for a campaign's conversions and "75.000" for an event's value, while GET /ltv/summary has answered numbers since 17305ee: one field, two types, depending on the endpoint read. The casts are where 17305ee put the summary's, in the repositories, so every reader gets the same types: MysqlLtvRepository::customers(), MysqlCustomerCrmRepository::get() (a NULL unit_price stays null; quantity is a number, DECIMAL(12,3), 1.5 as well as 2) and MysqlEngagementRepository's customer reads. The pages that read them (the customer modal, the customer table, the download, the merge search) already cast; their output is byte-identical before and after on a seeded instance. Every write that answers a customer (POST/PATCH, merge, alias) and the erasure's dry_run record answer the same, through customerDetail() and crm->get(). next-offer already answered numbers. The same endpoint answered custom_fields as {} with a value and [] with none (PHP's empty array): it is an object either way now, at the API edge, since the page reads it as an array. docs/openapi.yaml types the answers (LtvCustomer, LtvCustomerDetail and its parts, LtvCustomerEngagement). LtvDatabaseIntegrationTest holds them on a real server: every customer read and the summary walked whole for a numeric string outside a text field (a DECIMAL column added later is seen too), the values exact, and each answer's keys and JSON types equal to the spec's schema. Each defect was planted and caught. The Go CLI decodes no struct from these answers; its money floor now names refunded_amount, mrr and unit_price, and the --all fake answers numbers as the server does. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRoS2SViuxaucFaKWsFnLr

After 82b51c8 the customer reads answered numbers and the rest of /ltv still answered mysqli's text, so a subscription's amount was 29.99 in GET /ltv/customers/{id} and "29.99000" in GET /ltv/subscriptions. Measured on a real server, the reads that still served text were: - GET /ltv/subscriptions: amount, mrr, customer_ltv - GET /ltv/breakdown by campaign, ppc_account and landing_page: every amount and rate (avg_ltv "44.562500000", repeat_rate "0.5000", spend, cac, ltv_cac) and total_orders ("4") - GET /ltv/breakdown?by=product: the amounts, units ("1.500"), orders - GET /ltv/products, PATCH /ltv/products/{id} and its delete's dry run: price - GET /ltv/companies: total_revenue, mrr, order_count, last_activity_time - GET /ltv/abm and /ltv/abm/company: total_revenue, mrr, the three engagement averages ("30.0000000"), and the company view's last_activity ("1793183200", a GREATEST() MariaDB answers as text) #305's private helper is now Prosper202\Ltv\LtvNumbers, with decimals() for money and measures and integers() for counts and unix times, and each read above goes through it where its SQL is: the repositories, and the controller for the products (whose SQL is there). Each read names every such column, not only the ones this server answers as text, since MySQL and MariaDB type COALESCE() and GREATEST() differently. integers() leaves a value with a fraction as it is, for the type tests to see, rather than cutting it. A NULL stays null (a product with no price, a subscription whose customer row is gone); a name, sku, product key or reference of digits stays a string. docs/openapi.yaml types every /ltv read with a components schema (LtvSummary, LtvAcquisitionBreakdownRow, LtvProductBreakdownRow, LtvCohort, LtvMrr, LtvPrediction, LtvProduct, LtvCompanyRollup, LtvCompany, LtvSubscription, LtvAbmAccount, LtvAbmContact, LtvField, LtvWebhook, LtvWebhookDelivery, LtvIntegration, LtvNextOffer); they were additionalProperties: true, and GET /ltv/abm's list was documented as an object. LtvDatabaseIntegrationTest holds every read on a real server: the reads are the GET routes of api/v3/index.php's /ltv group, so a route added there fails until it is held too; each answer is walked for a numeric string outside a text field; the new numbers are checked for value; and each answer has its schema's keys and JSON types, into every schema a property refers to, and the spec's 200 answer for the path names that schema. Planted, each failed a named test: the subscription amount cast, the product price cast, the company order_count cast and the ABM last_activity cast each removed; LtvSubscription.amount typed string; an undocumented key in LtvAbmAccount; GET /ltv/abm's answer untyped again; a GET route added to the group; and the whole change's code reverted to 82b51c8's. On an instance installed from 82b51c8 and then from this branch, every /ltv answer has the same values with the types above changed and no other; the Customer LTV pages (the breakdowns, ABM, subscriptions, companies, company, products, customer) are byte-identical; the agent-eval LTV cases' checks pass. p202's tables now show these amounts to the cent (9.50 where they showed 9.5) and a zero rate as 0 (it showed 0.0000), and the PHP CLI's 29.99 where it showed 29.99000. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRoS2SViuxaucFaKWsFnLr

chatgpt-codex-connector · 2026-10-28T10:58:17Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	⚠️ Failed 2026-10-28T10:58:13.812763Z We couldn’t complete this review. Comment `@codex review` to try again.	`d58d841`	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector · 2026-10-28T10:58:19Z

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

Copilot

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota.

github-actions · 2026-10-28T10:59:28Z

Code review: CLEAN

For d58d841: 0 issue(s). The review is encrypted for the maintainers in the review-findings artifact of this run.

github-actions · 2026-10-28T11:01:28Z

Adversarial review: NON-BLOCKING

For 83a6814: 3 finding(s). The report is encrypted for the maintainers in the review-findings artifact of this run.

Rewritten on every push; the check fails on BLOCKING.

Brings in 1.9.79's new commits. The conflicts were in 202-config/changelogs.json, changelogs.txt, where both sides added lines: the incoming lines are kept as they are and this branch's follow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRoS2SViuxaucFaKWsFnLr

Brings in ltv-customer-numbers (this branch's stacked base), and through it 1.9.79's new commits. The conflicts were in 202-config/changelogs.json, where both sides added lines: the incoming lines are kept as they are and this branch's follow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01VRoS2SViuxaucFaKWsFnLr

claude added 2 commits October 28, 2026 09:56

Copilot AI balanced review requested due to automatic review settings October 28, 2026 10:58

Copilot AI reviewed Oct 28, 2026

claude added 2 commits October 28, 2026 12:16

tracking202 merged commit fa85a2b into 1.9.79 Oct 28, 2026
23 checks passed

Navigation Menu

Repository navigation

1.9.79: every LTV read answers money as numbers and counts as integers - #313

1.9.79: every LTV read answers money as numbers and counts as integers#313
tracking202 merged 4 commits into
1.9.79from
1.9.79-ltv-numbers-everywhere

tracking202 commented Oct 28, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 28, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 28, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

github-actions Bot commented Oct 28, 2026

Uh oh!

github-actions Bot commented Oct 28, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

tracking202 commented Oct 28, 2026

What changed and why

Tests: #305's type tests now cover every read

Readers swept by field name (#28)

Live, before (#305 head 82b51c8) and after (this branch)

Verification

For a reviewer to check

Uh oh!

chatgpt-codex-connector Bot commented Oct 28, 2026

Codex Review Summary

Uh oh!

chatgpt-codex-connector Bot commented Oct 28, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Oct 28, 2026

Code review: CLEAN

Uh oh!

github-actions Bot commented Oct 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Adversarial review: NON-BLOCKING

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Live, before (#305 head `82b51c8`) and after (this branch)

github-actions Bot commented Oct 28, 2026 •

edited

Loading