# How critical are the missing records?

Research date: 2026-09-23 UTC. The earlier report was too undifferentiated about uncertainty. We already have a good public-record estimate of average **absolute exports**. Unresolved internal accounting does not equally impair all downstream estimates.

## Main conclusion

- **Whole-project exported AF:** strong recorded estimate, 420,676.9294 AF / 18 = **23,370.9405 AF/year, WY2007–2024**, from the Approved total tunnel class. Unresolved native runoff does not prevent reporting this. It remains measured/administrative discharge, not city consumption; meter uncertainty and record reconciliation are separate questions.
- **Individual intake captured AF:** publish the existing matched collector means with their ledger/2016-zero caveats. Missing bypass chiefly prevents the percentage denominator, not reporting the intake ledger's captured AF.
- **Historical Red Cliff percentage:** the matched WY2007–2011 screen is informative about the order of magnitude. Real endpoint storage and recorded evaporation adjustments are small in this pooled period. A clearly conditional, partially naturalized cumulative balance stays close to **43%**. This is evidence for a useful approximate historical answer, not a full-mouth or present-day percentage.
- **Gold Park through WY2024:** there is a complete observed downstream gauge series and real storage endpoints. A provisional cumulative balance is feasible for its **35.6-square-mile** basin, separately from Red Cliff/full mouth. It suggests about **52%**, under the explicitly incomplete corrections below. The upper basin figure cannot be relabeled full valley.
- **Each individual colored basin:** missing native/bypass flow remains fundamental. A known captured volume alone supplies no denominator. One combined Missouri/Sopris gauge cannot divide their individual contributions, and confluence polygons include land below intake catchments.

## Why the cumulative balance is easier

At a downstream gauge below the entire existing collection system:

`N ≈ Q_downstream + E_tunnel + (S_end − S_start) + L_net − X_external + other corrections`

`E/N` is **export intensity relative to estimated native flow**.

`(N − Q)/N = (E + ΔS + L_net − X_external + other corrections)/N` is **estimated net flow reduction at that gauge**.

The five collector transfers, reservoir pass-through bookkeeping and reservoir-to-creek releases are **internal** at this larger boundary. They cancel or are already reflected in measured downstream Q. Therefore the disputed HOMOUTCO channel combination and French/Sopris source-account duplication do **not** independently block a cumulative Gold Park/Red Cliff balance. They still matter for E alone, short reaches, seasonal routing and source attribution. A downstream physical replacement inflow must be evaluated by where water actually enters; an administrative exchange is not automatically a physical imported inflow into Homestake.

Missing mid-period reservoir-storage readings also do **not** prevent a pooled balance when the beginning and ending storage are available: the intermediate changes telescope to `S_end − S_start`. Averaging cannot similarly make unknown bypass disappear for individual basins, and a long record does not cancel a systematic gauge bias.

## Exact pooled arithmetic (all Oct–Sep water years)

| Term | Red Cliff WY2007–2011 | Gold Park WY2007–2024 |
|---|---:|---:|
| Approved tunnel E | 139,942.6551 AF | 420,676.9294 AF |
| Observed creek Q | 185,166.8700 AF | 386,968.0066 AF |
| Beginning storage, 2006-09-30 | 41,254.0313 AF | 41,254.0313 AF |
| Ending storage | 39,696.6900 AF (2011-09-30) | 42,265.5300 AF (2024-09-30) |
| Actual endpoint ΔS | −1,557.3413 AF | +1,011.4987 AF |
| Mean ΔS | −311.4683 AF/year | +56.1944 AF/year |
| Recorded evaporation-use accounts | 1,315.7251 AF | 4,321.7130 AF |
| Of which stored-water evaporation | 466.4755 AF | 3,272.7042 AF |
| Of which stream-source evaporation | 849.2496 AF | 1,049.0088 AF |
| E/(E+Q), original two-path screen | 43.0448% | 52.0869% |
| E/(E+Q+ΔS), storage-adjusted export comparison | 43.2520% | 52.0217% |
| E/(E+Q+ΔS+recorded evaporation), conditional export comparison | 43.0768% | 51.7452% |
| (E+ΔS+recorded evaporation)/(same denominator), conditional net gauge-flow reduction | **43.0024%** | **52.4012%** |
| Conditional mean denominator | 64,973.5818 AF/year | 45,165.4527 AF/year |

The actual storage endpoints are Approved DWR contents records. For Red Cliff, the Q column retains the preserved DWR annual totals matching the original screen. Gold Park uses USGS daily means × 86,400 / 43,560; all 6,575 dates in WY2007–2024 have Approved observations, including 2,876 estimated daily values. Red Cliff's matched 1,826 dates include 1,311 estimated daily values. Approval does not remove estimation uncertainty.

**Worked Red Cliff calculation:**

`N_partial = 185166.87 + 139942.6551 − 1557.3413 + 1315.7251 = 324867.9089 AF`

`100 × 139942.6551 / 324867.9089 = 43.0767864%` export comparison.

`100 × (139942.6551 − 1557.3413 + 1315.7251) / 324867.9089 = 43.0024127%` conditional net flow reduction.

Storage alone moves the original Red Cliff export screen by **+0.2072 percentage point**; adding the recorded evaporation accounts then moves it **−0.1752 point**. For Gold Park WY2007–2024, the corresponding movements are **−0.0652** and **−0.2765 point**. Thus these known pooled adjustments do not overturn the scale of the original comparisons.

**Limits on the provisional figures:** the recorded evaporation classes are not yet verified as the correct physical *net* reservoir loss for the no-project counterfactual. Precipitation on the reservoir, the land's counterfactual evapotranspiration/runoff, seepage returning above/below the gauge, other net depletions, physical external inflows, and gauge/volume biases need scoped checks. The figures are not a calibrated natural-flow estimate with an error bar. They show how far the known corrections go, without waiting for unnecessary internal-account closure. Do not print four decimals as substantive accuracy.

A sensitivity of ±1,000 AF/year in the **remaining net denominator correction** moves Red Cliff's export comparison to **42.4238–43.7501%** and Gold Park's to **50.6243–52.9168%**. This is a ruler showing sensitivity, **not an asserted missing-term bound or confidence interval**. No evidence here establishes that all missing terms fit within that range.

## Model corroboration, kept separate

The official legacy model gives Gold Park native flow of 264,026 AF in WY2007–2011. The actual-gauge/export/storage plus recorded-evaporation subtotal above is 262,918.8439 AF for Gold Park in those same five years, about 0.42% lower. This is useful consistency evidence, not independent validation: the model uses substantially the same observed gauge and administrative records. Its inferred remaining modeled balance terms are 2,892 AF after legacy Q/export/ΔS, **not a separately documented evaporation measurement**; do not add this residual to the evaporation classes. Its individual source allocations are fixed aggregate model weights and cannot resolve every intake. The model input record ends in 2013, so it cannot simply provide the 2014–2024 native denominator.

## Why E remains more uncertain

For WY2024, the storage-account residual **2,578.0447 AF** equals **14.20% of tunnel export** and **15.35% of the unaccepted 16,793.5071 AF E diagnostic**. The collector/source mismatch **1,367.0698 AF** is **7.53%** and **8.14%**, respectively. The second outlet-channel volume is **1,274.5190 AF**, **7.02%** and **7.59%**. These are scale comparisons, not independent random errors; do not sum them or combine in quadrature. Their existence does not imply the tunnel export is wrong by those percentages. They affect a small internal branch's inferred balance much more directly than the cumulative downstream balance.

## Focused additional public search

New searches targeted historical USGS gaging terminology, station-number variants, NOAA owner CCSUWD and the actual NWS station identifiers. No additional validated tributary bypass cfs series was recovered in this short follow-up. The new NOAA current Colorado DCP metadata download confirms EFCC2/FACC2/MOCC2/SOCC2; it does not list FHCC2. Search-indexed NWS/IEM broadcasts reference FHCC2, but an IEM station table places that identifier near Colorado Springs, outside this basin. This conflicting-location lead must be resolved before treating French broadcasts as Homestake measurements. It is not proof that no French data exist. A broader USGS site-catalog query returned HTTP400; that failure does not expand the earlier successfully searched inventory. Historical USGS annual reports continue to identify Gold Park and Missouri, with no new collector-specific historical station recovered from these results.

The useful next request is narrower than asking for all internal accounts: first verify physical net-loss/external-inflow/other-depletion terms for the **cumulative Gold Park and historic Red Cliff boundaries**, and obtain the operator's existing collector sensor/rating archive for individual basins. The current full-mouth denominator still requires current downstream data or a validated lower-reach reconstruction. No external messages were sent.

## Replication and sources

`python analysis_update/followup_criticality_2026-09-23/basins/calculate.py`

Outputs: `cumulative_balance_sensitivity.csv`, `goldpark_matched_annual_inputs_WY2007_2024.csv`, `arithmetic.json`, and input hashes. These calculations use preserved datasets whose primary source URLs/retrieval timestamps are in `analysis_update/upper_balance/retrieval_manifest.json`, `analysis_update/basin_runoff/source_manifest.json`, and `analysis_update/statemod/raw/retrieval.json`. New web downloads have their own `source_manifest.json` in this folder.