Kairos · terminal red-team conclusion · 1 September 2026
NESTED
Count value only when a customer meant to continue and a payment, subscription, or entitlement failure broke that continuity.
The only retained-subscription value that qualifies
A failed payment is not automatically a lost willing customer. The large unpaid balance is real, but it includes cases we cannot control, cases current retries already recover, people who did not intend to continue, and value that may already sit inside Retention.
01 · The category boundary
The red team did not erase Payment Recovery. It put each kind of work where its value can be measured without counting the same money twice.
Customer Retention
The customer meant to continue. A payment, subscription, activation or entitlement failure stopped them.
Size and incremental durable value remain unmeasured.Outside Retention
False declines, retry cost, routing, fraud, risk and subscription-state integrity may create value without proof of renewal intent.
No accepted denominator, economics or accountable owner exists yet. If a material case appears, it needs its own category and must not inherit the unpaid balance.
Remove Payment Recovery from the separately ranked opportunity list. Keep the narrow retention mechanism. Keep reliability defects visible. Book no value until the cohort, counterfactual, durable outcome, cost and ownership are measured.
02 · Why standalone fails
A standalone opportunity must clear every gate below. Current evidence clears none of them completely.
The intent-positive technical-failure cohort is unmeasured.
We cannot yet separate lift from the recovery system already running.
No intervention has a mature controlled result against current practice.
Next-cycle paid use, retained use, refunds and later churn are not joined.
Fees, support, engineering, disputes and customer harm are missing.
No distinct accountable payment or entitlement owner is assigned.
Unrecovered invoices and retention losses are not reconciled in both directions.
≈51%
July 51.29%; August 50.54%. A serious diagnostic signal that still needs decomposition.
18–20%
Current practice already recovers part of first failures at the observed cutoffs.
≈6
The system is already retrying heavily. “Do more retries” is not an open thesis.
27.6
64,080 later attempts accompanied 2,318 observed recoveries across four advice-code classes.
≈$196/mo
The visible confirm_card_data surface before cost and durability haircuts.
31.5%
Directional warning only: 653 of 2,074 matched recovered cycles later churned.
The first-failure rate says “explain the system.” The unpaid face says “bound the diagnostic surface.” Neither says “this is the value we can win.” The earned number would be additional durable customer value above current practice, after all costs and harms.
03 · The rescue attempts
Some routes are dead. Some remain useful operating work. None currently earns a standalone Payment Recovery opportunity.
Current later-attempt volume is already heavy. Additional retries have no earned incremental case.
A timing test is valid only as a mature holdout against current practice after first-failure pathology is explained.
A customer-experience mechanism with a small visible proxy ceiling, not separate economics.
Looker and BigQuery point in different directions on unmatched populations. Neither supports a causal routing decision.
| Route | Current result | What would change it |
|---|---|---|
| Do-not-retry enforcement | Selection audit, not value. Selected retries recovered 38 of 56 invoices. | Joined cost, customer harm and recovery counterfactual. |
| Attempt-cost reduction | Unresolved and unpriced. | Per-attempt cost, retry ownership and outcome link. |
| Recovery communication | QA and measurement first. | Correct recipient, send tracking and causal recovery outcome. |
| Entitlement and webhook repair | High-priority integrity work, presently nested or reliability-owned. | Incidence, accountable owner and durable customer/economic effect. |
| Fraud, disputes, chargebacks | Unscored gap. | Joined cohort and preventable share. |
| Tax, FX, payout, settlement | Unscored gap. | Cohort economics tied to a controllable mechanism. |
| Measurement system | Necessary enabling work, not an opportunity thesis. | Evidence that it changes a material decision or outcome. |
Diagnosis, measurement repair, customer-state integrity and a bounded Retention question survive. Production changes to retries, routing, blanket suppression and new recovery messages remain blocked.
04 · The data-room answer
The audit did not declare every source exhausted. It converted the remaining unknowns into explicit reversal tests.
Decision-specific coverage
rooms dispositioned
Warehouse inventory
BigQuery objects dispositioned
What this proves
The obvious reversal paths have been named and checked against the current corpus.
It does not prove every room body has been queried, joined or content-exhausted.
Known evidence still sitting in rooms or waiting to be commissioned
The gap is no longer “maybe useful data is somewhere.” We know which missing joins, configurations and outcomes could change the decision, what each must prove, and which work stays outside Customer Retention.
05 · The cheapest decisive sequence
Do not commission a full customer-episode system first. Run the cheapest tests that can kill or reopen the thesis.
Set materiality, direction and stop rules before looking at reversal results.
Connect payer or invoice to account, subscription and UUID. Report unmatched rows, duplicates and identity conflicts.
Reconcile July and August billing-retry churn rows against unrecovered invoices at one mature episode or failed-obligation .
Split by invoice sequence, subscription tenure and rail. Add country or card fields only when a source-owned field exists.
Compare same-UUID re-subscription and privacy-safe post-failure use against successful renewals and recorded voluntary exits.
Then add technical cause, current-practice recovery, durable paid use, finance and harm before any intervention test.
Reopen toward standalone
Either the intent-positive technical-failure cohort proves material with a distinct action system, or a separate payment-operations pool clears its own economics and ownership.
Downgrade toward kill
Failures concentrate in acquisition or uncontrollable rails. Retain only justified reliability and integrity work.
Current state
Investigate the system, measure the cohort, assign reliability ownership and authorize no production intervention yet.
06 · Evidence and limits
The conclusion passed opposing verification with no remaining blocking or material findings. That certification applies to the current classification, not to every unanswered payment question.
Payment Recovery category red team. Integrated conclusion and current decision, 1 September 2026. Bound target SHA-256 begins baea5660.
All-room evidence receipt. Decision-specific coverage of 13 registered and 15 shadow rooms. Coverage is not content exhaustion.
Live BigQuery evidence receipt. Cohort, advice-code, censoring and warehouse-gap evidence from the authenticated live warehouse pass.
Terminal certification. High-confidence NESTED classification; opposing verification returned PASS with zero blocking or material findings.
Machine-readable decision record
The JSON record preserves the classification, evidence limits, production boundary and reopen rules without the page layout.