# The denominator card

**The Operator's Map · GFF 2026 Special · 11 September 2026 · Beyond the Benchmark**

A one-page template any number must fill before it enters a decision. Every field is borrowed from a fetched page that already enforces it, so none of the ten questions is exotic. Write the number at the top, go down the list, and count the lines you cannot fill in from the page the number came from. The card returns that count. It never returns a verdict on the number's author.

**The number, as stated:** ____________________________________

**Stated by:** ____________________ **On the page at:** ____________________ **Fetched on:** __________

| # | Field | The plain question | Borrowed from | Filled from the page? |
|---|---|---|---|---|
| 1 | Population | What set is this a count of, or a rate over? | RBI Payment System Indicators, Note 4: "Only domestic financial transactions are considered." SEBI PR 56/2026: three named issuers. | |
| 2 | Denominator | Numerator over what? Successes over attempts; alerts over flags; frauds over transactions. | RBI PSI Part V column head: "One in every X payment transaction fraudulent". NCBI Bookshelf NBK557491: PPV = TP / (TP + FP). | |
| 3 | Window | Over which dated period, or as of which date? "Every month", "annually", "to date" and "since launch" are not windows unless a date is attached. | RBI PSI monthly columns. PIB factsheet: "As of 26th August 2026". SEBI PR 56/2026: "September 7, 2026". | |
| 4 | Event definition | What counts as one success, failure, fraud, call or account? What is excluded? | RBI PSI Note 4: "failed transactions, chargebacks, reversals, expired cards/ wallets, are excluded." Note 6: "Data does not include attempts to perpetrate frauds." | |
| 5 | Reporting population | Who reported the underlying rows? | RBI PSI Note 5: "as reported by Scheduled Commercial Banks (excluding RRBs), Non-bank PPI issuers and Non-bank Credit Card issuers in CPFIR". | |
| 6 | Provisional or final | Can the number still change, and who owns the revision? | RBI PSI Note 1: "Data is provisional." SEBI PR 56/2026: "Issuances under the first phase are ongoing." | |
| 7 | Trials and interval | How many runs, and how wide is the uncertainty? | Terminal-Bench leaderboard schema: `n_trials` and `accuracy_ci95_half_width` are required fields. | |
| 8 | Harness | For an AI claim: what was around the model? (The Harness Card in the next chapter of this lane expands this line into twelve questions, organized by the seven layers around a model.) | arXiv 2605.23950: "Until harness specifications are disclosed, leaderboard comparisons for long-horizon agents should be treated as incomplete and potentially misleading." | |
| 9 | Base rate | For an alert or detector: how rare is the event in the population where it will run? | RBI PSI Part V, July 2026: one in every 80,846.92. NCBI Bookshelf: "Disease prevalence in a population affects PPV and NPV." | |
| 10 | Source of record | Where is the page this number is read from, and does it resolve? | PIB factsheet References block naming npci.org.in/product/upi/product-statistics, which resolves. NPCI FiMI release naming its technical report as where results live. | |

**Blanks that apply:** ______ of ______

**Count the blanks. The card returns a count, not a verdict.**

---

## Worked pass 1 — a rate from a launch release (register row A31)

**The number, as stated:** "with 95%+ success rates" · Stated by: a cross-border payments firm, in its own wire release · Fetched: 11 September 2026

| # | Field | Filled from the page? |
|---|---|---|
| 1 | Population | No. Whose transactions is not stated. |
| 2 | Denominator | No. Successes over what attempts is not stated. |
| 3 | Window | No. |
| 4 | Event definition | No. "Success" (authorized? settled? net of reversals?) is not defined. |
| 5 | Reporting population | No. Who reported the transactions is not stated. |
| 6 | Provisional or final | No label. |
| 7 | Trials and interval | No. |
| 8 | Harness | Not applicable (not an AI claim). |
| 9 | Base rate | Not applicable (a throughput ratio, not a detector). |
| 10 | Source of record | No page named. |

**Blanks that apply: 8 of 8.**

## Worked pass 2 — the pilot aggregate (register row A9)

**The number, as stated:** "aggregating 1,025 crore" (₹; register row A9 carries the sentence in full) · Stated by: SEBI, PR No. 56/2026 · Fetched: 11 September 2026

| # | Field | Filled from the page? |
|---|---|---|
| 1 | Population | Yes: three named issuers. |
| 2 | Denominator | Yes: a count of money moved; the three components are listed. |
| 3 | Window | Yes: 7 and 9 September 2026, on the page. |
| 4 | Event definition | Yes: settlement is defined on the page (the bond and the money move together, at once); the FAQ linked from the release: "The token is the corporate bond." |
| 5 | Reporting population | Yes: the regulator, in its own release. |
| 6 | Provisional or final | Yes: "Issuances under the first phase are ongoing." |
| 7 | Trials and interval | Not applicable (a record, not a measurement). |
| 8 | Harness | Not applicable. |
| 9 | Base rate | Not applicable. |
| 10 | Source of record | Yes: the release and its FAQ, both fetched. |

**Blanks that apply: 0 of 7.**

---

Sources for every borrowed field are listed in the chapter's Sources block and in the register (`beyond-benchmark-gff-artifact-b.csv`). Quotations keep their source spelling.