Every number on the site should survive a hostile reader. This file is where that reader lives. Each entry: the objection as sharply as a reviewer would put it, the answer, and what remains open. Resolved entries stay, with their resolution, so the arguing is on the record.
Rule: a new layer or meter does not ship until its skeptic entry is written. If the objection cannot be answered, the panel carries the caveat in plain sight or does not ship.
Objection. Copper's remaining-reserve figure changes every second, yet USGS publishes reserves once a year. The tick implies measurement that isn't happening.
Answer. The tick is interpolation: the January 2026 reserve figure minus the 2025 production rate applied to elapsed time. Every meter says so on its face ("ticking at the 2025 rate"), and the tier badge says ANNUAL. The interpolation is a faithful arithmetic consequence of the two published numbers; it adds no information and invents none. Its purpose is to make the rate legible, since a static "980 Mt" tells the reader nothing about how fast the account is draining.
Open. A reader skimming on a phone may see the moving number before the label. Consider a toggle to freeze meters at the published figure.
Objection. Static lifetime (reserves ÷ production) has been predicting imminent exhaustion for fifty years and has been wrong every time, because reserves rise with price and exploration.
Answer. Agreed, and the site says it in the methodology: static lifetime is a warning flag, not a countdown. That is why the meter shows four bands (mined, reserves, identified-not-economic, undiscovered) rather than one, and why the paper's argument is about gradient rather than volume: ore grade, energy per ton, and geographic concentration, none of which exploration reverses. The 43-year figure is reported because USGS reports the inputs; the page's claim is not "copper runs out in 2069" but "the economic band is this size and draining at this rate, and the next band costs more."
Open. The undiscovered band (3.5 Bt for copper) is a geologic model estimate from 2015. It should carry its own uncertainty, which USGS publishes as a range, and the meter currently shows only the mean.
Objection. The Energy Institute has not refreshed proved reserves since year-end 2020, so the oil/gas/coal ratios are 2020 reserves against 2020 production.
Answer. Correct, stated on the panel, and not fixable from this side: no public body has published newer global totals. The site reports the Energy Institute's own numbers with the Energy Institute's own date and flags the tier as "ANNUAL · reserves stale." 2025 production is known (record highs); applying it to 2020 reserves would mix vintages and is not done.
Open. OPEC's Annual Statistical Bulletin and EIA's International Energy Statistics publish newer country-level proved reserves. A cross-check panel showing those alongside the EI figure, with the discrepancies, would be more useful than silence.
Objection. The water layer shows single NOAA gauge readings per lake with no baseline, so the reader cannot tell high from low.
Answer. Also correct. The baselines were deliberately left blank rather than typed from memory. The USACE Detroit District monthly bulletin publishes the coordinated lake-wide means and the 1918–present averages; the fix is to pull that bulletin, not to guess.
Open. Unfixed as of this entry. The layer stays labeled "single-gauge readings, not lake-wide means" until the bulletin feed is wired.
Objection. The site presents itself as a running ledger. It is a set of JSON files regenerated once a day by a scheduled job.
Answer. The tier system exists so the reader never has to guess which it is: LIVE means the source updates at least daily and the job fetches it; PERIODIC, ANNUAL, STATIC say what they say. The daily cadence matches most of the sources; USBR, NOAA GML, and USDM publish daily or weekly, so fetching hourly would add nothing. "In motion" on the page refers to the meters, which are interpolated (entry 1), and that is labeled.
Open. Air stations, fires, and emissions sources do update hourly. When those layers are wired, the refresh for that subset should move to hourly, and the tier label should distinguish "hourly" from "daily."
Objection. SNOTEL SWE is near zero at most sites until November; a station reading 0.8 in. against a median of 0.3 in. shows "267% of median," which is noise dressed as signal.
Answer. The panel says so in its note. The number is shown because the station network is the asset, and the reader should see the sensors exist even when the signal is not yet meaningful.
Open. The map should suppress the percent-of-median color until the median exceeds a threshold (say 2 in.) and show plain SWE before that. Not yet done.
Objection. The auto-built meters carry USGS figures and no interpretation.
Answer. True. They were added so the account is complete rather than selective; a ledger that lists copper and silver but not nickel and zinc invites the question of why. The commentary is the author's job and is the part no pipeline can write. Until it exists, the meter says exactly that.
Open. 24 meters awaiting commentary.
Objection. Cumulative historical production is not in the USGS annual tables.
Answer. USGS's 2015 global copper assessment (SIR 2018-5160) states past production through 2015 as 600 Mt. Adding 2016–2025 world mine production from the annual MCS tables (about 215 Mt) gives about 815 Mt. The meter cites the basis. The 45% is 815 ÷ (815 + 980).
Open. The 2016–2025 sum uses rounded MCS world totals; the uncertainty is on the order of ±20 Mt, or about one percentage point on the bar.
Objection. A carbon budget is a policy construct with a probability attached (50% chance of 1.5 °C), not a measured stock.
Answer. The meter's label says "for a 50% chance of 1.5 °C" and its constraint line says "budget, not a reserve." It is on the commodity ledger because the framing fits: a stock, a rate, a date. The reader is told what kind of stock it is.
Open. The 2 °C budget and the 67%-probability variants belong beside it so the reader sees the sensitivity. Not yet added.
Objection. Much of the code and many of the fetchers were generated with an AI assistant. Why trust the numbers, and whose work is this?
Answer. The AI's part is not hidden; it is the method. This site was built by an engineering technician directing an AI assistant, the way a plant is built by a technician directing a crew: the crew runs the conduit, the technician decides where the conduit goes and signs off on the test. The pipeline, the fetchers, and the page code were written by the assistant at the author's direction. The decisions were the author's, and they are the part that makes the tool useful:
The numbers themselves are USGS's, NOAA's, USBR's, and the rest, fetched from the published files and linked on every panel, so anyone can recompute any meter by hand. The corrections log in METHODOLOGY.md records where the first draft was wrong and what fixed it, which is what checking looks like.
Steering is the skill on display. A reader who wants to evaluate it should read this file, the system map, and the corrections log, and ask whether the decisions were good ones.
Open. The 24 auto-built commodity meters carry figures the author has not yet checked against the source and commentary the author has not yet written. Until that is done, those meters are the assistant's plumbing without the technician's sign-off, and they say so.