Urd Atlas methodology

Verification & Evidence Pack v2

Worked examples showing what a careful reader can recompute from published artifacts, what should instead be checked against public chain evidence, and how to interpret edge cases that look inconsistent until the methodology is understood.

Verification classes used here

Class	Meaning	Typical examples
A	Directly reproducible from published artifacts.	`regime.determinism_hash`, driver-to-Gold consistency, confidence gate threshold, published label identity
B	Independently checkable against public chain evidence.	`tx_count_daily`, `block_count_daily`, fee direction, block-time behaviour
C	Publicly interpretable but not fully reconstructable from the trust layer alone.	Private calibration internals and full proprietary classifier implementation detail

One complete diligence path — sample pack to expected output

Use one coherent artifact set from the public sample pack, not six disconnected mini-checks. The path below is the intended end-to-end diligence flow for a technical buyer.

Step	File or action	Expected outcome	Verification class
1	`sample-pack/ethereum/2026-03-31/gold.json`	Gold row loads with `chain = ethereum` and `date = 2026-03-31`.	A / B
2	`sample-pack/ethereum/2026-03-31/derived.json`	Derived row loads for the same `chain` and `date`, exposing smoothed fields such as `__ma7` and `__ma30`.	A
3	`sample-pack/ethereum/2026-03-31/meta.json`	Meta row resolves to named state `CONGESTED` with public gate threshold `0.40`.	A
4	Recompute the named-row integrity anchor	Canonical hash recomputes to `81b295000696`.	A
5	Independently inspect Gold-level facts	Gold-level counts and fee behaviour should be independently checkable against public Ethereum evidence for the same date.	B
6	Interpret remaining classifier internals carefully	The full regime classifier remains deliberately non-public even though the published outcome is auditable.	C

Download these files → run this code → expect these results

Files to download first

Put these three technical files in one local folder: gold.json, derived.json, and meta.json from the Ethereum 2026-03-31 sample-pack row set.

import hashlib
import json
from pathlib import Path

base = Path("sample-pack/ethereum/2026-03-31")

gold = json.loads((base / "gold.json").read_text(encoding="utf-8"))
derived = json.loads((base / "derived.json").read_text(encoding="utf-8"))
meta = json.loads((base / "meta.json").read_text(encoding="utf-8"))

# 1) Basic artifact identity
assert gold["chain"] == "ethereum"
assert gold["date"] == "2026-03-31"
assert derived["chain"] == "ethereum"
assert derived["date"] == "2026-03-31"
assert meta["chain"] == "ethereum"
assert meta["date"] == "2026-03-31"

# 2) Gold -> Meta current-value consistency for named drivers
for driver in meta["regime"]["drivers"]:
    metric = driver["metric"]
    if metric in gold and "current" in driver:
        assert driver["current"] == gold[metric], (metric, driver["current"], gold[metric])

# 3) Derived row is present for the same observation date
required_derived_fields = [
    "tx_count_daily__ma7",
    "tx_count_daily__ma30",
    "median_tx_fee_native__ma7",
    "median_tx_fee_native__ma30",
]
for field in required_derived_fields:
    assert field in derived, field

# 4) Public confidence gate and named output
assert meta["status"]["label"] == "CONGESTED"
assert meta["regime"]["label"] == "CONGESTED"
assert meta["publish_confidence"]["threshold"] == 0.40
assert meta["regime"]["ruleset_id"] == "eth_l1_v1"

# 5) Recompute determinism_hash from the public named-row payload
payload = {
    "asof_date": meta["regime"]["asof_date"],
    "chain": meta["chain"],
    "drivers": meta["regime"]["drivers"],
    "label": meta["regime"]["label"],
    "ruleset_id": meta["regime"]["ruleset_id"],
}
canonical = json.dumps(payload, sort_keys=True, separators=(",", ":")).encode("utf-8")
sha = hashlib.sha256(canonical).hexdigest()[:12]
assert sha == meta["regime"]["determinism_hash"]
assert sha == "81b295000696"

print({
    "chain": meta["chain"],
    "date": meta["date"],
    "label": meta["status"]["label"],
    "gate_threshold": meta["publish_confidence"]["threshold"],
    "determinism_hash": sha,
})

Expected printed result	Value
`chain`	ethereum
`date`	2026-03-31
`label`	CONGESTED
`gate_threshold`	0.40
`determinism_hash`	81b295000696

Expected intermediate values and checks inside that same path

Layer	Field or check	Expected result	Why it matters
Gold	`chain`	ethereum	Confirms the artifact belongs to the intended chain.
Gold	`date`	2026-03-31	Confirms the three technical files are aligned on the same observation date.
Derived	`tx_count_daily__ma7` and `tx_count_daily__ma30` exist	Present in the same-date Derived row	Shows that smoothed trend context exists for the same observation row.
Meta	`publish_confidence.threshold`	0.40	Public gate threshold for whether a named label may be published.
Meta	`regime.ruleset_id`	eth_l1_v1	Pins the named output to a specific public ruleset identifier.
Meta	`status.label`	CONGESTED	Named public output for the worked example.
Meta	`regime.determinism_hash`	81b295000696	Integrity anchor proving the named-row payload identity.

What is directly reproducible vs independently checkable vs still black box?

Category	Items in this worked path
Directly reproducible from the downloaded files	File identity, same-date alignment, presence of Derived fields, public confidence gate threshold, named label, ruleset id, and `regime.determinism_hash`.
Independently checkable but not fully produced by this script	Whether the Gold-level daily counts, fee behaviour, and block conditions for 2026-03-31 match public Ethereum evidence.
Deliberately left black box	The full classifier internals that transform the entire evidence surface into the named regime, including non-public calibration detail and proprietary decision plumbing.

Important diligence boundary

This page is designed to show that the published output is auditable without pretending the entire production classifier is open-sourced. Reproducibility of the public trust layer and full disclosure of every proprietary internal are not the same thing.

Additional spot checks still worth keeping

Use the UNKNOWN/DEGRADED sample row at sample-pack/ethereum/2025-04-21/meta.json to confirm that sub-threshold rows do not publish a normal-confidence named state.
Inspect at least one ARB or BASE sample row to confirm that slower cadence and lag semantics are chain-specific, not pipeline failure.
Compare one Gold metric against a public explorer or independently gathered chain data for the same date.

Class

Meaning

Typical examples

Directly reproducible from published artifacts.

regime.determinism_hash, driver-to-Gold consistency, confidence gate threshold, published label identity

Independently checkable against public chain evidence.

tx_count_daily, block_count_daily, fee direction, block-time behaviour

Publicly interpretable but not fully reconstructable from the trust layer alone.

Private calibration internals and full proprietary classifier implementation detail

Step

File or action

Expected outcome

Verification class

sample-pack/ethereum/2026-03-31/gold.json

Gold row loads with chain = ethereum and date = 2026-03-31.

A / B

sample-pack/ethereum/2026-03-31/derived.json

Derived row loads for the same chain and date, exposing smoothed fields such as __ma7 and __ma30.

sample-pack/ethereum/2026-03-31/meta.json

Meta row resolves to named state CONGESTED with public gate threshold 0.40.

Recompute the named-row integrity anchor

Canonical hash recomputes to 81b295000696.

Independently inspect Gold-level facts

Gold-level counts and fee behaviour should be independently checkable against public Ethereum evidence for the same date.

Interpret remaining classifier internals carefully

The full regime classifier remains deliberately non-public even though the published outcome is auditable.

import hashlib import json from pathlib import Path base = Path("sample-pack/ethereum/2026-03-31") gold = json.loads((base / "gold.json").read_text(encoding="utf-8")) derived = json.loads((base / "derived.json").read_text(encoding="utf-8")) meta = json.loads((base / "meta.json").read_text(encoding="utf-8")) # 1) Basic artifact identity assert gold["chain"] == "ethereum" assert gold["date"] == "2026-03-31" assert derived["chain"] == "ethereum" assert derived["date"] == "2026-03-31" assert meta["chain"] == "ethereum" assert meta["date"] == "2026-03-31" # 2) Gold -> Meta current-value consistency for named drivers for driver in meta["regime"]["drivers"]: metric = driver["metric"] if metric in gold and "current" in driver: assert driver["current"] == gold[metric], (metric, driver["current"], gold[metric]) # 3) Derived row is present for the same observation date required_derived_fields = [ "tx_count_daily__ma7", "tx_count_daily__ma30", "median_tx_fee_native__ma7", "median_tx_fee_native__ma30", ] for field in required_derived_fields: assert field in derived, field # 4) Public confidence gate and named output assert meta["status"]["label"] == "CONGESTED" assert meta["regime"]["label"] == "CONGESTED" assert meta["publish_confidence"]["threshold"] == 0.40 assert meta["regime"]["ruleset_id"] == "eth_l1_v1" # 5) Recompute determinism_hash from the public named-row payload payload = { "asof_date": meta["regime"]["asof_date"], "chain": meta["chain"], "drivers": meta["regime"]["drivers"], "label": meta["regime"]["label"], "ruleset_id": meta["regime"]["ruleset_id"], } canonical = json.dumps(payload, sort_keys=True, separators=(",", ":")).encode("utf-8") sha = hashlib.sha256(canonical).hexdigest()[:12] assert sha == meta["regime"]["determinism_hash"] assert sha == "81b295000696" print({ "chain": meta["chain"], "date": meta["date"], "label": meta["status"]["label"], "gate_threshold": meta["publish_confidence"]["threshold"], "determinism_hash": sha, })

Expected printed result

Value

chain

ethereum

date

2026-03-31

label

CONGESTED

gate_threshold

0.40

determinism_hash

81b295000696

Layer

Field or check

Expected result

Why it matters

Gold

chain

ethereum

Confirms the artifact belongs to the intended chain.

Gold

date

2026-03-31

Confirms the three technical files are aligned on the same observation date.

Derived

tx_count_daily__ma7 and tx_count_daily__ma30 exist

Present in the same-date Derived row

Shows that smoothed trend context exists for the same observation row.

Meta

publish_confidence.threshold

0.40

Public gate threshold for whether a named label may be published.

Meta

regime.ruleset_id

eth_l1_v1

Pins the named output to a specific public ruleset identifier.

Meta

status.label

CONGESTED

Named public output for the worked example.

Meta

regime.determinism_hash

81b295000696

Integrity anchor proving the named-row payload identity.