I'll start by looking at the workspace and trying to fetch the primary sources named in the brief.The paper exists. Now let me pull the full HTML text and the other named sources in parallel.I have the full paper. Now let me pull the key independent sources in parallel: the Blockworks piece, the 0x "PropAMM shenanigans" post, the azfuller Ethereum memos, the Helius Solana overview, and WOOFi's sPMM docs.Good material. Let me read the truncated tails of the azfuller memo (capital efficiency, pricing curves, router details) and grep the other overflow files for remaining numbers.WOOFi's oracle design confirmed (mid-price p, spread s, coefficient k, Chainlink fallback). Let me grab the remaining mechanics/numbers from the Helius article tail.Let me grab the remainder of the 0x piece for Pattern 2 quantification and their recommended mitigations.I have enough material. Now I'll create artifacts/ and write the CSV first (primary deliverable), then the memo.Now the memo with the exact required headings.Validating the CSV structure and the memo headings.My check used the wrong column index (r[4] is unit, not value). Re-checking properly:CSV validates: 162 rows, all values numeric, no thousands separators, every row has source_url and as_of_date, estimate flags are clean booleans. Final check on workspace state:Done. Both deliverables are in artifacts/, untracked, and nothing outside the allowed paths was touched.

What I produced

• artifacts/mechanism-economics.md — decision memo using the exact required headings. All numbers traced to arXiv 2609.38056 (verified against the full HTML text), the 0x "PropAMM Shenanigans" post (2026-03-20), Fuller's Ethereum propAMM memo (data to 2026-08-31), Helius's Solana survey (Aug 2025), LambdaClass's Router/Factory posts, and WOOFi's Wooracle docs. WOOFi's sPMM is compared neutrally as a provider — notably, the paper shows Elfomo's Base propAMM actually reads WOOFi's Wooracle as one of its onchain anchors.
• artifacts/edge-and-cost.csv — 162 long-format rows covering markouts @2s/−5s/15s (pooled and per-venue), retail-proxy costs, route-class markouts, counterparty-tiered spreads, spoofing metrics, update gas/CU costs, priority-fee multiples, update cadence, inventory turns, and volumes — propAMM vs AMM across Solana, Base, Monad, and Ethereum.

Key findings for the build decision

• Update cost is 5–15× below swap cost on EVM chains and >10× on Solana (485–676 CU median; HumidiFi's optimized update is 143 CU / ~$0.0018); the 21k gas floor dominates EVM update cost.
• Repricing (one anchor+spread write reprices a whole ladder: 26.7k gas vs 152k for level updates) and onchain flow netting are the two gas tricks worth copying.
• PropAMM edge decomposes into repricing, counterparty-tiered fees (penalties alone flip Tessera's single fills from −1 to +1.29 bps), winning the cross-venue arb leg, and spoofing (~1.08 bps avg gap on Tessera; 5–10 bps typical per 0x).
• Retail pays less on propAMMs than AMMs on all three chains (0.26 vs 2.59 bps Solana), but the spoofing gap lands on aggregator-routed flow; an onchain requoting router is the deployed fix.

Flagged uncertainties (all marked in the files): the Blockworks report is paywalled — figures cited from its publisher summary with estimate=true; the Solana spoofing rate is second-hand via 0x (original removed); the paper has minor internal inconsistencies (data start date, 39% vs 41% identical-fill share); Helius vs paper swap-CU figures conflict (~150k vs ≥16.9k, different scopes).

Created both deliverables:

• Hostile audit
• Number reconciliation CSV — 21 comparison rows.

Corrected stale dates, separated incompatible share metrics, and marked missing evidence. Local structure, provenance, date, and arithmetic checks passed. Files remain untracked.

Created all three deliverables:

• Engineering memo
• Volume CSV: 475 rows
• Gas CSV: 23 rows

Local validation passed. Sources, conflicting totals, estimates and unverified gaps are documented. Files remain untracked for upload.

Created all three deliverables:

• Engineering memo
• Volume data — 84 rows
• Aggregator routing — 42 rows

Validated schemas, dates, arithmetic and source references. Unknowns remain explicit. Files are untracked.

Created all three deliverables:

• Engineering memo
• Provider volume CSV: 3873 rows
• Pair volume CSV: 89 rows

Local checks passed for schemas, dates, source attribution, arithmetic and required headings. Unavailable pair data and conflicting sources are explicitly marked. Outputs remain untracked.

I've written all three files: the decision memo, the gap matrix CSV (46 rows) and the chart pack (8 charts). I checked the CSV with a script: every row has source_url and as_of_date, and every headroom value matches the formula stored next to it. All ten required memo headings are present. Nothing was re-pulled from the web; the memo combines the five upstream reports and follows the hostile audit wherever they disagree.

Decision in the memo: build one shared pricing and risk engine with a separate execution adapter per chain. Copy the single write that reprices the whole book (median 26706 gas, against 152013 for touching each level, from the paper's decode of Tessera on Base). Ship a quote SDK whose output matches settlement exactly. Don't promise gas savings or market share until a replay test on identical trades exists.

Starting leads that turned out wrong:

• Blockworks date: the "Block Building Wars" article is dated 2026-02-17, not 2026-10-14. Its markout data only covers late January to early February 2026.
• LambdaClass dates: the PropAMM Router was introduced 2026-07-14; only the Factory was announced in October (2026-10-14).
• Solana share: the paper's ">50%" (one pair, SOL/USDC) and TDMM's 26.1% (all pairs, 30 days) measure different things. The memo explains the difference and does not average them.
• Provider list: the six named Solana providers are not the full list; ZeroFi, AlphaQ, Aquifer, QuantumAMM, Riptide, Taurusfi and Flint are also active.

Gap matrix: penetration is tracked PropAMM volume divided by each chain's total DEX volume over 2026-09-25 to 2026-10-24. Solana leads at 33.954%. The headroom estimates rank Ethereum (about 10.7B USD), Robinhood Chain (about 9.1B), Base (about 7.8B) and BSC (about 7.0B) highest.

• Caveat: headroom is a mechanical what-if that assumes every chain could reach Solana's mix, so use the ranking rather than the dollar amounts.
• WOOFi as a provider: each chain appears twice, once without WOOFi and once with it.
• Unknown cells: Sui, Sei, Aptos and Unichain have no tracked PropAMM volume, so those cells are empty, not zero. For stablecoin, tokenized-equity and long-tail pairs only the PropAMM side is known, so their penetration is empty too.

Still UNVERIFIED (labelled as such, never stated as fact):

• What share of each aggregator's flow goes to which venue.
• WOOFi's swap gas on Ethereum.
• Swap gas for El Zorro alone and for a typical LambdaClass Router trade.
• How much volume is wash or arbitrage.
• Whether any PropAMM operates on the chains with empty cells.

These questions are ranked by decision value in the memo.

The files are in artifacts/ and left untracked; nothing was committed. Several untracked shell and editor config files (.bashrc, .gitmodules, .vscode and others) were already in the repository root; I didn't create or touch them.

Files are in artifacts/:

• build-memo.md
• opportunity-sizing.csv
• chart-pack.md