Surface Consumption Rate, every computer dive

Methodology

Everything is computed from the raw per-sample records in the database — no summary fields were trusted without verification.

1 · Decoding. Each dive in log_data is a 4-byte length prefix + gzip blob. The 30 dives from 2020–21 decode to JSON (header + samples); the 40 dives from 2024–26 are Shearwater’s binary Petrel Native Format (PNF), parsed per the open-source libdivecomputer reference: 32-byte records; depth as a big-endian 16-bit value in tenths of a foot; tank pressure as 12-bit values in units of 2 psi (values ≥ 0xFF0 are “no comms / not paired” sentinels, top nibble is transmitter battery state); water density, surface pressure (mbar), sample interval and dive duration from the opening/closing records. A quirk found in the JSON exports: the per-record timestamp field is corrupted (it counts interleaved event records), so sample times were reconstructed as index × sample interval and verified against each dive’s recorded duration.

2 · Ambient pressure. For every 10-second sample, ambient pressure is rebuilt from depth: P_amb = P_surface + depth × ρ·g, using the dive’s own recorded surface pressure (997–1023 mbar) and the computer’s salinity setting (ρ = 1020 kg/m³ on all these dives). Because the computer derived logged depth from measured pressure with that same ρ, this inversion recovers the sensor’s true absolute pressure exactly — the result is immune to the salinity setting being “wrong” for fresh-water sites like the i90 Bridge.

3 · The GUE formula, integrated. The computation is exactly the cheat-sheet definition —

— with one upgrade: instead of a single eyeballed average depth, “Time × Average ATA” is computed exactly as ∫ ATA(t) dt, trapezoidally over every 10-second sample, so one minute at 2 ATA counts as precisely two surface minutes whatever the profile shape. Gas volume consumed is (P_start − P_end) × the tank factor, where P_start/P_end are the first and last valid transmitter readings inside the dive window; because the pressure differences telescope, brief transmitter dropouts between them cannot bias the result. Each log’s ~60-second post-dive surface tail (Shearwater’s end-dive delay) was cut using the computer’s own recorded dive length. (Terminology: GUE’s SCR is the same quantity other communities call RMV or volume-SAC; the psi/min “tank rate” column is the tank-specific pressure SAC.)

4 · Tank factors (psi → ft³). The SCR column uses the standard linear conversion: psi/min × tank factor (rated volume ÷ rated pressure): HP80 = 80/3442, LP80 = 80/2640, HP100 = 100/3442 ft³/psi. The SCR·Z column corrects two real-gas effects the linear factor ignores: air’s compressibility Z(P,T) (a psi near 3400 holds ~4% less gas than a psi near 1500) and gas density at the actual water temperature (~50 °F gas packs ~3–4% more per psi than the 70 °F rating). Moles consumed are computed with the NIST reference equation of state (CoolProp) at each dive’s logged average water temperature; per dive the correction runs −7% to +11% and it pulls the apparent outliers (top-of-tank short dives) back toward the pack. Tank identity comes from your own Shearwater Cloud gear notes (HP 80 / LP 80 / HP 100); the four 2026 twin-set dives are owner-confirmed 2×HP100. The two 2020-12-19 rentals and both Bandito 2021-04-30 dives have no recorded tank, so a psi-factor range is shown instead. Gas mix (air vs EAN32) doesn’t enter — SCR/RMV measure volume, not oxygen.

Validation. Four independent checks were run against this analysis. (1) The binary parser was audited offset-by-offset against libdivecomputer’s open-source Shearwater parser; every 70-dive scan matched, and no dive anywhere in the database carries a second transmitter channel. (2) Every dive was reconciled with Shearwater Cloud’s own summaries: durations exact, average depths within a foot, start/end pressures within the transmitter’s 2-psi step (endpoint picks differ ≤ 22 psi — ≤ 1% of ΔP). (3) A from-scratch independent re-derivation of three dives (one JSON, two PNF) reproduced the SCR values to 0.1 psi/min: 20.78 vs 20.8, 17.19 vs 17.2, 11.70 vs 11.7. (4) The 5-second AI event stream embedded in 7 of the 13 transmitter-equipped 2020–21 logs corroborates the sample-stream endpoints on every dive that has it (the other 6 predate that firmware feature).

Caveats. A freshly filled tank cooling into 46–54 °F water loses up to ~140–200 psi to thermal contraction with zero breathing; whatever hadn’t equilibrated before splash is counted as consumption, so absolute SCR carries a one-sided uncertainty of roughly +0–5% (worst on first dives with warm fills; second dives of the day are cleaner). Dives flagged † are short hops or same-tank re-entries whose windows are dominated by descent and surface swimming — their SCR is real but not representative. The 14-second Alki “dive” is a surface blip and was excluded. The first Teric’s files start at #3 — dives #1–#2 (very likely the Edmonds 2020-09-11 pair) were never synced to Shearwater Cloud, so no telemetry exists for them.