Append-only · first frozen 7 August 2026
Disagreement is a result
IGRM measures the share of monitored coverage devoted to five India-relevant fault lines. Its comparators measure other things: AI-GPR scores risk intensity in a different newspaper corpus; IMF PortWatch counts ships. When their ranks separate sharply, this page preserves the date, the two readings and a route back to the evidence. It does not decide which instrument is right.
Reading rule
A rank gap says where to inspect. It does not identify a cause, establish predictive power, or show that the higher-ranked instrument is more accurate.
Current register
| Published | Period | Comparison | IGRM rank | Other rank | |gap| | Direction | Receipt |
|---|---|---|---|---|---|---|---|
| 2026-08-07 | week of 2026-07-27 | Bab el-Mandeb salience vs PortWatch transits | 99.7 | 1.3 | 98.4 | IGRM higher | inspect |
| 2026-08-07 | week of 2026-07-27 | Hormuz salience vs PortWatch transits | 95.4 | 0.4 | 95.0 | IGRM higher | W32 note |
| 2026-08-07 | 2020-02 | IGRM composite vs AI-GPR India_all | 2.8 | 79.6 | 76.9 | AI-GPR higher | inspect |
| 2026-08-07 | 2019-06 | IGRM composite vs AI-GPR India_all | 87.0 | 15.7 | 71.3 | IGRM higher | inspect |
| 2026-08-07 | 2019-02 | IGRM composite vs AI-GPR India_all | 23.1 | 94.4 | 71.3 | AI-GPR higher | inspect |
| 2026-08-07 | 2021-05 | IGRM composite vs AI-GPR India_all | 7.4 | 74.1 | 66.7 | AI-GPR higher | inspect |
| 2026-08-07 | 2023-01 | IGRM composite vs AI-GPR India_all | 86.1 | 20.4 | 65.7 | IGRM higher | inspect |
AI-GPR ranks use the 108 eligible months in the frozen 2017-07 to 2026-07 benchmark. PortWatch rows rank each series within its own complete 2019-present weekly history. Samples are not pooled across register families.
Receipts and reuse
The machine-readable register is divergence_register.json. The AI-GPR rows come from the verbatim registered output; its protocol, frozen script and interpretation are public. Only ranks and aggregates are republished from AI-GPR, never its raw values.
Append-only policy
Rows are never silently deleted. If a source revises, the original row stays and a dated correction is attached. New rows need a named comparator, a frozen sample, a rank rule, a period and a receipt. This keeps the register useful even when a comparison is awkward for IGRM.