assembly-economics/workplans/ASSEMBLY-WP-0018-gadm05-grounding.md

id	type	title	domain	repo	status	flavor	depends_on	owner	topic_slug	created	updated	state_hub_workstream_id
ASSEMBLY-WP-0018	workplan	GADM-0.5 grounding: dematerialization, ecological classes, non-rival output	communication	assembly-economics	finished	residual	ASSEMBLY-WP-0007	claude-code	coulomb-social	2026-10-13	2026-10-13	a34af97f-0000-5186-8abb-c44bc3fd197e

GADM-0.5 grounding

Residuals of ASSEMBLY-WP-0007 (models/GADM-0.5.md). The structural extensions work and pass their grounding tests, but most of their parameters are priors. This workplan grounds them in data, and then decides whether GADM-0.5 replaces GADM-0.4 as the reference model.

Dematerialization: demand split and efficiency trend

id: ASSEMBLY-WP-0018-T01
status: done
priority: high
state_hub_task_id: "ba4ac94a-0351-52d7-8dbf-64fd837f8303"

Done 2026-10-13: on the right operator (material draw per output) the model reproduces the observed trend without an efficiency trend (the first comparison used total throughput, which was wrong). The data constrain φ_r: EU 0.66, DE 0.15. At the EU fit, decoupling disappears (BR-011). Weakly identified, because 2010–2024 is pre-generative-AI. E5 is wired into the full checkpoint (sources.json). dematerialization.py, models/GADM-0.5.md § 5. The efficiency parameter eff remains available (default 0).

Modeled throughput intensity falls 0.1–0.5 % per year; observed EU dematerialization is about 1.6 % per year (resource productivity, Eurostat env_ac_rp / env_ac_mfa, 2000–2022; UNEP IRP for US and AR). Add an autonomous efficiency trend for thr, and fit it with dm_eps and dm_lam to the observed intensity series per jurisdiction. Re-run demand.py. Do the decoupling and sinks-breach readings survive?

Ecological classes: starting levels, loads, replenishment

id: ASSEMBLY-WP-0018-T02
status: done
priority: high
state_hub_task_id: "3691dd92-6ba0-51f2-8a95-6c569cbe1aab"

Done 2026-10-14: starting levels from control variables (renewables 0.12, sinks 0.30), compute load grounded (IEA: e2 0.03 → 0.0015), physical load on rival capacity (structural fix), material replenishment supported (copper reserves). The generative regime's sinks breach disappears; the BR-011 conflict sharpens (fits EU 0.79 / DE 0.57 against task data ≤ 0.25). models/GADM-0.5.md § 7. Earlier note: Started 2026-10-13, interrupted (search limit): one control variable so far: atmospheric CO₂ 423 ppm, boundary 350, high-risk line 450 (PHC 2025, via a secondary source). That puts the sink class 73 % into the zone of increasing risk. Next: the full PHC 2025 table (PIK executive summary PDF) for all 13 control variables; IEA Energy and AI (data-center electricity and emissions) for the compute load shares; USGS reserve histories (for example copper) for material replenishment.

Ground the class starting levels (per-boundary control variables, Planetary Health Check), the load shares by class (energy, material and emission intensities of physical capital vs. compute; IEA for data centers), and material replenishment (reserve-to-production ratios over time, USGS). Which class binds next depends on the replenishment.

Non-rival output parameters

id: ASSEMBLY-WP-0018-T03
status: done
priority: medium
state_hub_task_id: "92de5938-90df-5268-b937-e1ea033e0308"

Done 2026-10-13: φ_r 0.15 (0.05–0.25) from Anthropic Economic Index occupational shares; xn 0.6 from GitHub's 18.5 % public share of 2025 contributions. GADM-0.5 defaults updated, experiments re-run, readings unchanged in direction (models/GADM-0.5.md § 6). The task data make the EU dematerialization fit (φ_r 0.66) implausible as an AI effect (BR-011).

Ground φ_r (rival share of organism output; task-level data from the Anthropic Economic Index, L4) and xn (commons feed; open-licensing shares of AI-generated output, for example code on public repositories).

Selection weights: speed and links

id: ASSEMBLY-WP-0018-T04
status: done
priority: low
state_hub_task_id: "60940505-b8f9-55f6-b55f-aa4075fe10ff"

Done 2026-10-14: the speed (0.05/yr) and strengths are consistent with Bessen et al. and Kang 2016; the capture link o_pd was tested (with Pd passed to the base auxiliaries): VFAC effect ≤ 0.0003. Capture of selection is not material (models/GADM-0.5.md § 8).

Ground the weight adjustment speed (0.05 per year) and the capture strengths. Test the candidate link O = Q (1 − Pd) (organism selection quality falls with discovery concentration). Only that would make capture of selection material for VFAC.

GADM-0.5 as reference model

id: ASSEMBLY-WP-0018-T05
status: done
priority: medium
state_hub_task_id: "3aace583-cc0b-58f7-9eb0-e241dcfddc1e"

Done 2026-10-14 (decision): GADM-0.4 stays the reference model; GADM-0.5 is the module for ecology and demand questions. Revisit when post-2024 E5/L4 data decide BR-011 (models/GADM-0.5.md § 9).

After T01–T03, decide whether GADM-0.5 (classes, demand split, non-rival output) becomes the reference model for the band probe, the calibration (WP-0013/0015) and the cross-border world (WP-0011/0016). Re-run the key readings and record what changes in the break register.

Close-out 2026-10-14

T01–T05 are done. The open question, BR-011 (decoupling vs. φ_r), is tracked in the break register, with the trigger "post-2024 E5 and L4 data". E5 is wired into the full checkpoint, and L4 is collected at reviews (WP-0015-T01). No new workplan.