Gas-capture flag gates the site-type default in both trace paths

[HughRunyan]

Output

landfill_gas_collection now gates the site-type gas-capture default in both trace paths, not just one. A site with no recorded collection system is modelled at zero capture wherever it lives.

The defect

site_only_estimate_trace and citysite_estimate_trace carried copies of the gas-capture logic and the copies drifted. The citysite copy applied the site-type default (0.60 / 0.45 / 0.0) with no check on the flag, while still reading that same flag to pick ox_cap vs ox_nocap. The resulting Landfill was internally contradictory — constructed with gas_capture=False and gas_capture_efficiency=0.6 on the same object.

The rule it broke is not invented here. Two modules already state it:

• build_gas_capture_rates.classify_assets — "gccs NaN & lgc True → has a system, efficiency unknown. Emphatically NOT zero"; False → a genuine 0.
• uncertainty.recovery_and_oxidation_components — "Mirror SWEET's gate, not just its tables: city_params.py applies the site-type capture default ONLY when gas-capture presence is true."

Reproduction

10_05_26 submission archive, run_id 20260921T231117Z, SWEET 1c98d59 (city_params.py byte-identical to main, so this is the code that ran). Join site, site_year and city_site_year on asset_identifier at year = 2024:

	city-linked	flag	gce	ox	sites
BRA Controlled Dumpsite	✓	False	0.45	0.05 (ox_nocap)	959
BRA Sanitary Landfill	✓	False	0.60	0.10 (ox_nocap)	1,035
BRA Sanitary Landfill	✓	True	0.60	0.22 (ox_cap)	15
non-BRA Sanitary Landfill	✗	False/NULL	0.00	0.10	6,290

Expected: capture 0 wherever the flag is not a yes.
Actual: 1,994 sites at 0.45/0.60 while flagged False. Every one is city-linked; nothing outside that set is affected.

The symmetry is the argument: False is an unverified default in both Brazil and Mexico (the repo's own diagnostic says so), yet Brazil's 1,994 ran with full capture and Mexico's 1,938 ran at zero — decided only by whether the site is attached to a city record.

There was also a latent crash. citysite's gas_capture_presence == True raises TypeError: boolean value of NA is ambiguous on a null flag; it has not fired only because all 3,886 city-linked assets carry a real True/False.

The fix

Same history as the MCF table, same fix: one definition, two callers.

• _resolve_gas_capture is now the single home of the rule; both methods call it.
• GAS_EFF_OPTIONS / OX_OPTIONS move to module level; the two per-method copies are deleted.
• site_only's semantics are preserved exactly. citysite additionally picks up the pd.NA guard and the empty-gascap_df fallback.
• sinar_city_and_site and site_only_estimate are untouched — neither is on the Climate TRACE path.

Expected change in output

Derived by rescaling the archive against emissions = ch4 * (1 - gce*0.98) * (1 - ox); not yet confirmed by a run.

• 1,994 Brazilian sites: 2024 FOD 655 → 1,534 kt CH4 (2.34x)
• Global: +879 kt, +5.1% of the 17,323 kt all-site total; Brazil 1.51x
• Only 4 of the 1,994 are fusion-published, so the published change is essentially the FOD change
• Their published uncertainty also rises, 0.254 → ~0.446, because uncertainty.py already assumed the gated behaviour and was therefore charging no recovery-error term

Labelled model-output-change, not breaking-change: nothing stops running, the numbers move.

Acceptance criteria

• A flag that is not a yes (False, None, NaN, pd.NA, 0, "No") yields zero capture in both paths
• A yes yields the site-type default in both paths, with ox_cap
• A measured gas_collection_efficiency still overrides the flag in both directions
• A null flag no longer raises in the citysite path
• Capture data with no reported_emissions_year falls back to the gated default, not NaN
• Oxidation and capture can never disagree about presence
• Neither _trace method keeps its own copy of the tables (guard test)
• site_only's behaviour is unchanged

Definition of Done

• Acceptance criteria met
• Tests pass — 241 SWEET, plus 1,633 + 247 in the pipeline repo against this branch
• Docs updated — GAS_CAPTURE_LINKAGE.md in the pipeline repo carries the full investigation
• Reviewed & merged

Pairing

Must be merged and run with RMI/RMI_Climate_TRACE_Waste_Methane branch the-gas-capture-flag-gates-the-default, which carries the investigation write-up and the companion skiptop fixes.

🤖 Generated with Claude Code