NEAR / Methods & reproduction

A short path from source to claim.

The website is a static research product. It has no accounts, transaction signing, paid API dependency or backend database. The scenario engine runs in your browser. The frozen analysis runs with Node alone.

Run a fresh checkout

Use Node 20 or later. No dependency installation is needed for the website or analysis. Python 3 is used only as a local static server.

git clone https://github.com/0xpg/near-evidence.git
cd near-evidence
npm test
npm run build
python -m http.server 4187 --bind 127.0.0.1 --directory dist

Open http://127.0.0.1:4187. Repository access requires authorization because the source repository is private; the website is publicly deployed. Serve dist/, which contains only the intended website assets and public-source evidence. Do not serve your home folder or credentials.

Reproduce the headline without a model

Take the 30 saved UTC date buckets from 23 September through 22 October. Sum the vendor revenue_usd field: $2,225,003. We label it tracked collection-wallet receipts because it does not establish net income or purchases.

annualized receipts = 2,225,003 × 365 / 30
                    = $27,070,869.833333
historical reference value = 1,309,898,856.237708 × $4.44
                           = $5,815,950,921.695423
hypothetical ratio = annualized receipts / reference value × 100
                   = 0.4654590487%

This ratio assumes 100% purchase execution, 100% retention and no additional unmeasured costs. Those assumptions are hypothetical. It is not an observed buyback yield, dividend, fair value or price target. The zero-purchase default credits no unverified purchases; it does not claim none occurred.

The evidence chain

1. Source: public API responses, pinned RPC block/configuration, saved transaction receipts and executed Dune results.
2. Extraction: inherited public acquisitions are preserved verbatim in the SHA-256 manifest, with source provenance. Original acquisition code lives in the preserved prior research repository; this repository contains the relevant frozen inputs and inspectable query references.
3. Snapshot: dated operating and market inputs; raw responses in data/snapshots/. A hash proves preservation, not source accuracy.
4. Transformation: analysis/reproduce.mjs verifies every aligned daily cell against the saved raw API arrays, rejects duplicate date buckets, rebuilds monthly sums, checks supply conversion and issuance configuration, and runs model consistency checks.
5. Calculation: the shared calculation engine runs in both the browser and Node. Validated outputs feed the website and report figures.
6. Claim: the claim ledger records the source, extraction, period, method, limitations and potential falsifier beside each substantive claim.

What each time window means

Definitions that prevent false conclusions

Notional volume measures reported routed activity. Gross fees combine channel and protocol charges. Collection-wallet receipts are a distinct vendor proxy, without a complete expense, source-period or disposal reconciliation. Verifier fees are asset-denominated collector balances; they do not establish automatic market buys or burns. Accumulated inventory can be distributed or sold. Staking issuance is not external customer income.

April–September native DeFi stocks, if inspected in the source pack, must not be combined with cross-chain Intents flows on one scale. Selected-date origin-account recurrence measures accounts, including relayers; it does not measure unique customers or 30/60/90-day retention.

The 27 September–26 October Dune/API volume totals agree within $1 but share upstream lineage. Dune fees exceed the API by $145,994.83 (2.237% of Dune), unresolved. The supplied Q3 report’s $10.41bn volume differs from the API’s $9.786bn. Neither discrepancy is hidden or resolved by averaging.

Rerun calculations; refresh sources separately

Changing Scenario Lab assumptions runs only local arithmetic. Reset restores the documented snapshot. Exports contain model version, snapshot identifier, all inputs and outputs. Missing or invalid values are errors, not measured zero.

Source refresh is an analyst operation: create new raw filenames, preserve UTC period and retrieval timestamps, inspect schema, units, grain, nulls, duplicate conventions and source coverage, then hash the new bodies and reconcile before creating a new snapshot version. Public RPC data can be fetched with read-only calls; provider rate limits and archival coverage can prevent historical reconstruction. Dune execution may need a user account or paid quota; frozen results need neither.

SQL under data/queries/ pins the saved comparison dates where the source permits. Source queries and downstream provider tables can change; preserving SQL does not guarantee independent reindexing or permanent hosted results. No automatic refresh job is configured.

Regenerate the report

The checked-in PDF report and SVG charts are generated with Python’s ReportLab package, using the same checked Node outputs. To rebuild after an analytical change:

node analysis/reproduce.mjs
python -m pip install reportlab
python scripts/report.py
npm run build

The ordinary website build needs no Python package installation. It checks report-source.json and rejects a PDF whose analytical inputs, engine or ledger changed. It does not silently regenerate a report with a different cutoff.

Analytical review and limits

An independent reviewer reproduced the 0.465459% result from saved inputs without using the engine, challenged the alternative theses and checked accounting distinctions. Public documentation and current protocol configuration were cross-checked on 2 November. The site is not an independent network reindex or an audit of NEAR consensus, contracts or custodians.

Still missing: a complete external fee-origin conversion custody disposal reconciliation; economic-customer cohorts and actual incentive cost; confidential partner margins; independently reproduced deployed WASM; a complete period mint/burn census; and evidence of market mispricing. No claim of profitability or excess investment return follows from this work.

Prior work and official requirements

The original private NEAR investment research repository is preserved. This separate product reuses dated public research snapshots and source-query work, with new narrative, shared browser calculation, report generation and interface. Earlier research conclusions were reconsidered; the selected thesis remains conditional.

Official rules checked on 2 November: development closes 14 November 2026 at 23:59 UTC (15 November at 06:59 Jakarta). The application page and extended rules allow work since the 19 October announcement; earlier commits are baseline. AI-generated code is allowed with disclosure. The linked Investment Thesis prompt is an example, not a required technical architecture.

This product, with its source repository kept private, is not submitted or submission-ready. An eventual entry requires a public repository, README disclosures and reproducible instructions, a license, and the human’s public demo post; a local-only demo needs video. Those publication and submission actions require separate authorization. No README or agent-instruction files are included in this repository under the current private-build instruction.

Design references

Finishing review follows Unslop by Damir Sokolovsky (CC BY 4.0). Navigation icons are from Lucide; their license is preserved with the SVG assets.

The interface uses annotated editorial pacing inspired by the FT’s visual journalism, visible definitions and chart downloads from Our World in Data, narrative layering from Kontinentalist, and evidence-led exploration from Truth & Beauty and Visual Cinnamon. The landing illustration is an original illustrative field of simplified human figures with shallow-depth focus, adapted from the composition and pacing of the supplied 30-second video. It does not represent observed people, entities or transaction flows. Connections appear in short bursts with quiet intervals; three labels select different illustrative people every two to three seconds with a brief text glitch. The scene honors system reduced-motion preferences and stops when hidden, out of view or keyboard-focused. It has no visible pause control. Its geometry self-check runs with node scripts/check-scene.mjs. The fallback illustration is an original deterministic ASCII surface, inspired by the glyph-density approach in Codrops' ASCII shader study and the textmode.js community. It is decorative, not a measured economic flow. Regenerate it with node scripts/ascii.mjs; no runtime renderer or GPU dependency is needed. Saans fonts from the supplied archive are included for this noncommercial research website at the owner’s request. The supplied font files are preserved under assets/fonts/.