Real cross-pax synthesis. When independently authored pax share constructs, agents can run analyses no single pax could support.
You don't need a CLI, an MCP server, or any local tooling. Download global-supply-chain-risk.pax.tar.gz, extract it, open the included llm-context.md, and paste it into Claude or ChatGPT as the first message. The model now has 38 constructs, 113 findings, and 47 sources on supply-chain vulnerability sitting in its context.
Then ask it real questions. The agent cites by finding id (F041) and source id (baldwin_2020) — every claim traces back to a real paper. No hallucinated stats, no made-up effect sizes.
# step 1 — get the pax
$ curl -O https://pax-market.com/pax/global-supply-chain-risk.pax.tar.gz
$ tar -xzf global-supply-chain-risk.pax.tar.gz
$ cat global-supply-chain-risk/llm-context.md
→ 38 constructs · 113 findings · 47 sources
# step 2 — paste it into Claude (or ChatGPT)
USER > [contents of llm-context.md pasted as
the first message]
USER > a Tier-1 semiconductor supplier just
announced a 6-month outage. given the
constructs and findings in this pax,
what's the realistic downstream
impact window and what evidence
supports it?
CLAUDE > Single-source dependency triggers
the strongest cascade in the literature.
- F041 (baldwin_2020, n=312 firms):
average lead-time stretch is 4.2x
nominal for single-source nodes
(β=4.21, SE=0.78, p<0.001).
- F067 (miroudot_2021): inventory
buffers below 6 weeks correlate with
revenue loss > 15% within one quarter
of disruption.
- P004 (proposition): supplier
diversification compounds with
nearshoring to halve cascade duration.
Realistic window: 3-6 months for tier-2
exposure if buffers exhaust. Cite F041,
F067, P004 in your memo.Three pax cover related phenomena from different angles: Fearon & Laitin's civil war onset model focuses on structural opportunity (civil_war_onset, per_capita_income), the coup-prediction pax maps military-elite coordination (regime_durability_coup, power_sharing_institutions, patronage_capacity), and the state-fragility pax tracks institutional erosion (institutional_capacity, legitimacy_deficit, state_fragility_index).
None of these construct ids are registered bridges across the three pax — each pax was authored independently and uses its own vocabulary. But the concepts overlap. Pasted together, an agent can recognize that legitimacy deficit, fragility index, and elite-coordination metrics are measuring related things, and synthesize a cross-pax answer that no single pax was designed for.
# step 1 — pull all three pax
$ for p in fearon-laitin-2003 military-coup-prediction \
state-fragility ; do
curl -O https://pax-market.com/pax/$p.pax.tar.gz
tar -xzf $p.pax.tar.gz
done
# step 2 — concatenate the three llm-context.md
$ cat */llm-context.md > combined-context.md
$ wc -l combined-context.md
4,217 combined-context.md
# step 3 — paste combined-context.md into
# Claude as the first message,
# then ask:
USER > using ONLY the constructs and findings
in the three pax I just gave you, list
fragility indicators that show lead
time > 2 years before BOTH civil war
onset and coup attempts. cite by id.
CLAUDE > Three indicators meet the lead-time
bar across all three pax:
1. legitimacy_deficit
state-fragility F012 (n=78 countries):
OR 2.4 for civil-war onset, OR 1.9 for
coup, lead time 2-4 yrs.
Replicates fearon-laitin-2003 F003
at p < 0.001.
2. factionalized_elites
coup-prediction F031 (n=45 episodes):
precedes coup by median 3.1 yrs.
3. per_capita_income (low)
fearon-laitin F044, state-fragility F088:
same direction, both pathways, robust
across specifications.
cite: F012, F031, F044, F088, P007The first two stories show the paste-to-Claude path. This one shows the agentic path: when an analyst is running the Praxis MCP server, an agent can do more than recall findings — it can pull the underlying datasets registered with the pax, run a registered engine on them, and return real coefficients with diagnostics.
Below: an agent loads the taxation-fiscal-policy and income-inequality-gini pax, calls praxis_get_data to pull their shared OECD-revenue / WID panel, runs an OLS regression via praxis_run_engine, and reports the coefficient table inline. Bridge construct: redistributive_effect_taxes.
USER > using the registered datasets in
taxation-fiscal-policy and income-
inequality-gini, regress post-transfer
gini on top marginal rate plus controls.
return coefficients with diagnostics.
CLAUDE > [tool] praxis_install_pax
↳ { pax: 'taxation-fiscal-policy' } ok
[tool] praxis_install_pax
↳ { pax: 'income-inequality-gini' } ok
[tool] praxis_align_constructs
↳ { a: 'taxation-fiscal-policy',
b: 'income-inequality-gini' }
shared:
top_marginal_income_tax_rate
redistributive_effect_taxes
income_share_top_10_pct
[tool] praxis_get_data
↳ { dataset: 'oecd_wid_panel_v3',
years: [1990, 2018],
countries: 'OECD' }
loaded · 786 country-year rows · 42 countries
[tool] praxis_run_engine
↳ { engine: 'panel_ols_two_way_fe',
y: 'gini_post_transfer',
x: ['top_mtr', 'gdp_per_capita_log',
'union_density', 'trade_openness'],
fe: ['country', 'year'],
cluster: 'country' }
estimating · 786 obs · 42 panels...
──────── coefficient table ────────
coef se t p
top_mtr -0.084 0.024 -3.45 0.001
gdp_per_capita_l -2.11 0.71 -2.97 0.003
union_density -0.063 0.018 -3.50 0.001
trade_openness +0.012 0.009 +1.31 0.190
R²(within) 0.42 F(4,40) 18.7 p<0.001
fixed effects: country, year (clustered SE)
──────── reading ────────
A 10-point rise in the top marginal rate
predicts a 0.84-point reduction in
post-transfer Gini, controlling for income
level, labor power, and openness. Effect
is robust to bootstrap (95% CI [-1.31,
-0.36]) and consistent with F034 in the
taxation-fiscal-policy pax (β = -0.08).
[tool] praxis_run_engine
↳ { engine: 'sensitivity_breusch_pagan' }
heteroskedasticity p = 0.041
→ recommend cluster-robust SE (applied)
cite: F034, F021, P009