Baseline, amendments, data refreshes and updates
Everything here is append-only. The baseline was frozen before the first tested release; nothing earlier is ever rewritten, and the commit hashes let anyone check.
Probability 30% · confidence LOW · status ACTIVE
The deflator turned positive in 2026Q2 and the price leg is more likely than not to hold through 2027Q1 (I-05); the September PMI output price index points to renewed factory-gate momentum. But the turn is concentrated upstream and coincides with the oil and metals shock (I-01). Private domestic demand has not turned (I-02): with the September PMI, new orders slipped below the expansion line on a three-month average, so fewer demand tests pass than the three of six that confirmation requires. Policy impulse is fading (I-03); exports help but do not satisfy the domestic-demand leg (I-04). Price leg roughly 65%, demand leg below even odds given the price leg: 30%.
Why: The baseline states the thesis is sustainable only if the implied GDP deflator is non-negative in BOTH 2026Q4 and 2027Q1 AND at least 3 of the 6 domestic-demand (H2) tests pass at the terminal release. The frozen CONFIRMED rule implements that exactly. The frozen FALSIFIED rule did not implement its complement: it fired only if the latest deflator was negative (T-H1-04 false), or if the deflator streak held while 5 of 6 demand tests failed. Two outcomes in which the stated success condition plainly fails therefore fell through to RETIRED (inconclusive) instead of FALSIFIED: (a) 2026Q4 deflator negative but 2027Q1 non-negative, so T-H1-05 fails while T-H1-04 passes; (b) deflator condition met but only 2 (or fewer, down to 1) demand tests pass. Separately, the decision window was left as null "to be set from the 2027 NBS calendar", which would have needed a later edit to the frozen case; it now opens when the first 2027Q1 GDP print is in the information set.
Effect: At the terminal release the outcome is now CONFIRMED exactly when the stated condition holds and FALSIFIED exactly when it fails. RETIRED is reserved for a genuinely inconclusive case: a required terminal input (the deflator streak, or enough demand tests to decide 3 of 6) has insufficient data. Early falsification (FALSIFIED_EARLY) and every evidence test are unchanged.
Amends 2026-10-04-baseline.yaml (commit bbd0b35); must be committed before September 2026 CPI and PPI (14 Oct 2026, 09:30 Beijing); informed by tested data: no. Changed: terminal_rules.FALSIFIED, terminal_rules.otherwise, horizon.terminal_window_opens. Probability, confidence, evidence tests and claims are unchanged.
The refresh job fetches it automatically after release, re-runs the tests and appends a data record here. The analyst then appends an update saying whether it changes the thesis.
No analyst update since the baseline yet: no tested release has been published since it was frozen. This is the expected state until the first release above.