Reconciling the portfolio contradiction
Generated by build_scheme_reconcile.py.
Generated by build_scheme_reconcile.py.
Options found three rivers beat one by 3.9x. Synthesis found the portfolio loses. Both cannot be right. This changes the three candidate variables one at a time from the options baseline, for one scheme and for three.
| variant | basin each | injection each | settling | n=1 | n=3 | n=3 / n=1 |
|---|---|---|---|---|---|---|
| options baseline | 5/n | 40k/n | 0 d | 1.24 | 0.80 | 0.64x |
| + settling 2 d | 5/n | 40k/n | 2 d | 1.21 | 0.79 | 0.66x |
| + basin 1.0 Mm3 | 1.0 | 40k/n | 0 d | 1.06 | 0.81 | 0.77x |
| + injection 150k | 5/n | 150k | 0 d | 2.02 | 0.88 | 0.44x |
| synthesis config | 1.0 | 150k | 2 d | 1.40 | 0.89 | 0.64x |
The options baseline does not reproduce, and that is the finding. At n=1 it matches exactly (1.24 against the 1.10 options reports). At n=3 this reimplementation gives 0.80 where options reports 4.32 - on what is, as far as line-by-line comparison shows, the same configuration.
build_scheme_options.py still produces 4.32 when re-run, so this is not a stale number. It is a discrepancy between two code paths that should agree, and it has not been isolated.
The 3.9x portfolio result is suspended, not confirmed and not explained. Two independent implementations - this one and synthesis - agree the portfolio LOSES on this geometry in every variant tested. One says it wins decisively. Until the difference is found, no portfolio conclusion should be drawn in either direction, and the 3.9x figure must not be quoted.
The single-scheme numbers are unaffected: all three implementations agree at n=1, which is why the rest of the layer stands.
No mechanism is claimed for the discrepancy because none has been isolated. What is established is that the number must not be used.
What the variants do say
Across every configuration the ratio stays below 1 (0.44 to 0.77), so on this geometry one larger scheme beats three smaller ones sharing the same plant. That is consistent with the physics the aquifer chapter established - leakage goes as 1/r^2, so dividing a mound is expensive and the extra intakes do not recover it.
The largest single effect is injection capacity: giving each site the full 150,000 m3/day lifts the single scheme to 2.02 while barely moving the portfolio (0.88), because three small mounds saturate on headroom while one large mound absorbs the rate. That reinforces the design chapter - wells beat earthworks, but only where there is mound enough to take the water.
What this means for the earlier chapters
Options section 3 remains correct as stated - its control was sound and its numbers reproduce - but its conclusion needs the condition attached. The 3.9x figure describes a scheme where the plant is shared out between sites, not one where each site is independently equipped.
Synthesis is also correct as stated, for its own configuration. The contradiction was never between the models; it was between two different schemes wearing the same name.