The Best Temperature Is Taken · Validation bench

The bench · run in the page, nowboth hands as corrected

17 of 17 vectors pass · both hands agree everywhere

The first hand reproduces every frozen value across 55 declared checks, and the independently derived second hand agrees with it on the full output set for all 29 readings: the seventeen vectors, three probes beyond them, and the nine declared scenarios. Reinstate either defect below to see the bench go red.

Vectors17 / 1755 declared checks against frozen values

Probes beyond the vectors3 / 3readings the declared suite does not cover

Declared scenarios9 / 9the nine cases, cross-checked on every output

Two hands agree29 / 29full output set, every reading

The defect switchdefaults to the corrected arithmetic

Two defects were found in the research package’s relationships while building this world and corrected in the package on 13 September 2026. Either can be reinstated here, in the first hand only. The second hand is frozen and carries no switch, because a check you can bend is not a check.

R12A catalyst that lasted longer the hotter it ran

output · deactivation_run_hours

R31Advising a reader to wait out a term the register does not record

output · strategy_code

The seventeen frozen vectorsall pass

Each vector is run through the first hand and checked against its frozen expected values, then the full output set is compared with the second hand. Tolerances are the ones declared in the vector file.

CaseCoordinatevs frozenvs second hand

V-CENTRALthe declared defaultspassagree

V-HAZER-900900 °C · 1 bar(g) · iron orepassagree

V-COLD500 °C · non catalytic · series Bpassagree

V-HOT-NONCAT1200 °C · non catalyticpassagree

V-MOLTEN-ALLOY1065 °C · molten alloypassagree

V-LOW-TEMP-REGEN600 °C · nickel solidpassagree

V-DEACT-575575 °C · nickel solidpassagree

V-FREE-HOLE-C01700 °C · nickel solidpassagree

V-PRESSURE-GATE900 °C · iron ore · 150 bar(g)passagree

V-MAP-2031year 2031passagree

V-JUST-BEFOREyear 2035passagree

V-LITERATURE-BAND-700700 °C · literature bandpassagree

V-REGEN-AIR575 °C · nickel solid · regen airpassagree

V-REGEN-H2575 °C · nickel solid · regen h2passagree

V-REPLACEMENT-CLASS900 °C · iron orepassagree

V-2036-STRIPS900 °C · iron ore · year 2036passagree

V-OUT-OF-BASIS900 °C · iron orepassagree

Three probes beyond the vectorshands agree

Readings the declared suite does not cover, carried over from the second hand’s own cross-check. There are no frozen values here, so the only check is whether two independent implementations of the same evidence agree. This is where the deactivation defect surfaces: every in-basis vector sits exactly at the 575 °C reference temperature, and 550 °C does not.

CaseCoordinatevs frozenvs second hand

probe 1550 °C · nickel solidno frozen valuesagree

probe 2850 °C · iron oreno frozen valuesagree

probe 31100 °C · non catalyticno frozen valuesagree

The nine declared scenarioshands agree

The cases the package declares in prose, run as machine checks on the full output set. This is where the strategy defect surfaces: it only fires on a parcel that is held and undated, and no vector stands on one.

CaseCoordinatevs frozenvs second hand

centralthe declared defaultsno frozen valuesagree

cold_design500 °C · non catalytic · series Bno frozen valuesagree

hot_non_catalytic1200 °C · non catalyticno frozen valuesagree

pressure_gate900 °C · iron ore · 150 bar(g)no frozen valuesagree

map_2031year 2031no frozen valuesagree

just_before_hazer_opensyear 2035no frozen valuesagree

free_hole_nickel_700700 °C · nickel solidno frozen valuesagree

low_temp_regen_point600 °C · nickel solidno frozen valuesagree

literature_band_700700 °C · literature bandno frozen valuesagree

How the two hands differ

The second hand is re-derived rather than re-used. It shares no helper with the first: it carries the equilibrium anchors as parallel arrays and finds them by bisection where the first hand scans sorted pairs; it maps the register’s claimed element to the reader’s catalyst corners, which is the inverse of the first hand’s table; it quantises by its own route; and it reads the specific-electricity band as a midpoint of 6 and 10 where the first hand carries the figure 8. The only thing the two share is the register itself, exactly as both of the research package’s own hands read the same file.

Numbers are compared with a relative tolerance of 1e‑12 to absorb floating-point noise between two differently-grouped expressions. Both defects move their outputs by whole multiples, so nothing real hides under that floor.