LAUNCHKEY — Methodology Annex. How the record was measured.

Methodology Annex.

How the historical record was measured, what the grading rule gets wrong, and what it cannot see. Annex draft 2 · Methodology v2 · Oct 21, 2026. The rules themselves live on the Grading Rules page.

What Changed in v2

Draft 1 of this annex published a blind spot: confirmed disposals on graduated coins were annotated on the record but never reached the grade, because the v1 rule counted harmful exits from one ending only. Methodology v2 closes it.

V2 separates the fact from the verdict. A measured exit is printed on the record as a fact, with its evidence, and moves nothing by itself — selling early is common, and on its own it is not a finding against anyone. An exit becomes an adverse item — the only kind that moves a grade — only when a published aggravator is also measured. Endings from the earlier backfill stand exactly as stamped; items live alongside them. Nothing was overwritten.

Grades are records, not promises. Nothing in this annex describes a coin as safe, verified, or trustworthy, and no grade is a claim about any person's intent or identity.

The Item Vocabulary

EXIT EVENT — A FACT

A measured disposal: a confirmed drawdown of the developer's birth token account to at most 10% of its sampled baseline with observed outflow, or a measured slow exit (below). Printed with its evidence. No grade consequence by itself.

ADVERSE EXIT ITEM — A FINDING

An exit fact that also carries at least one aggravator under the test below. One adverse item makes a key's record adverse — printed PROBATION · ADVERSE RECORD, a different label from plain Probation.

SLOW EXIT — A FACT

PROPOSED definition: a measured disposal of at least 50% of the sampled baseline that completes after the first 60 minutes and within 7 days of launch. A slow exit is adverse only through an aggravator, the same as a fast one.

COVERAGE, STATED ON EVERY ITEM

The platform definition measures holdings and disposal at cluster level — the dev wallet plus wallets linked by the printed criteria (shared funder, direct dev transfers). The historical desk data read the birth token account only, so every historical item records its coverage as birth account only. An item's absence is absence of a measurement, not a clean finding.

The Aggravator Test

Deception — the disposal contradicts a lock or hold position printed at launch. On the historical set: unchecked on all 194 measured coins — no position was printed at launch for a pre-platform coin, so there is nothing for a disposal to contradict. The test arms on platform launches, where the file prints the lock state at birth.

Extraction + abandonment — all three: cumulative disposal of at least 90% within 7 days of birth (timed at the disposal episode's completion point — the earliest observed reading at or below 10% of baseline, never above it again); no subsequent dev-wallet market activity on the coin; and a quiet-death or abandoned-with-holders ending. Fires only when all three are checked true.

Hidden position — the position printed at birth materially understated the cluster's true position, and the hidden share was disposed of. On the historical set: unchecked on all 194 — no position was printed at birth, and cluster disposal was not measured.

Two operationalizations inside B are PROPOSED, marked again in §7: "no subsequent activity" is read as no dev-signed transaction after completion that increases the position (remainder disposals and zero-change settlement transactions are the episode's tail; transactions the dev did not sign never count); and the abandoned-with-holders class is read through the desk's peak-collapse test (all-time high at or above $20,000; last read below the lower of 10% of the peak or $5,000).

What the Test Found — 194 Confirmed Disposals

Outcome	Coins
Aggravator B fired → adverse item	188
No aggravator established → exit event, fact only	6
Of the 120 dumps on graduated coins: now adverse	116 (4 fact only)
Of the 74 pass-2 HARMFUL EXIT stamps: stand as adverse	72 (2 convert to facts)
Devs invisible to the v1 grade, now carrying an adverse item	31 of 32
Slow completions (after 60 min, within 7 days)	7 — all 7 also adverse

The six fact-only items are the doctrine working as written: in each, the dev disposed of at least 90% of the sampled position and then came back — a dev-signed re-entry after the disposal completed — so the abandonment leg fails and no finding stands. The disposals remain printed facts. It must also be said plainly: on this set, two of B's three legs never discriminate — the conduct set was selected on collapsed coins with confirmed drawdowns of at least 90%, so those legs are true by the set's construction. The separation the test actually performed here is the abandonment leg: dumped and never came back (188) versus dumped but re-engaged (6). Half of all confirmed disposals completed within 12 seconds of the recorded creation time; the 90th percentile within about 11 minutes; the slowest took 139 hours.

Measured Performance of the Rule

An independent reimplementation of the v2 rule agrees with the published v2 snapshot on 3,335 of 3,335 wallets. The snapshot (Oct 21, 2026): PROBATION 3,181 (94 with adverse records) · PROVEN 153 · PILLAR 1 · 188 adverse items · 194 exit-event facts. The founder's key, graded by the same rule with no adjustment: Key #001 — PROVEN (9 launches, 1 graduated, 0 adverse items, 0 exit events).

As-of-birth — each coin graded only on its dev's prior record; shares of coins whose ending was determined:

Tier at birth	Determined	Graduated	Harmful-exit ending
PROBATION (clean)	419	10.5%	0.95%
PROBATION (adverse)	1,069	14.6%	5.1%
PROVEN	699	3.7%	1.0%
PILLAR	3	3 of 3	0 of 3 — counts only; no rate quoted

The adverse flag separates harmful-exit endings in the expected direction. It does not separate graduation: adverse-at-birth devs' coins graduated more (14.6%) than clean Probation (10.5%) or PROVEN (3.7%). An adverse record marks a dev with an aggravated exit on record; it does not mark a dev whose next coin will fail. Full-record grades — the kind the board displays — show zero adverse items for clean PROVEN and PILLAR keys by construction: an adverse item is what makes a key adverse. They must not be read as predictions.

Item rates computed over all determined endings are floors, not estimates: adverse items exist only on the 194 conduct-measured coins, a set selected by a price-shape label. The full matrices, including the item-axis floors and their denominators, are in the annex data file referenced by the desk.

Where the Grade Was Most Wrong

PROVEN at birth, the coin then carries an adverse item — 13 coins. E.g. mint 2bQcQFdqp62Wo99noGRYdWdLrCZ3SAVXtjTP2jUspump, dev Cm9BjYWLEUzwHQNV3Fn5dzghYpiqN8zuGW8soy7DoDQX (61 prior launches, clean at birth); the coin peaked at $107.4K, graduated, and the birth account was drawn to at most 10% with no re-entry.

PILLAR at birth, then an adverse item — 0 coins. The two cases draft 1 named under v1 no longer qualify: under v2 those devs were already adverse before the coins in question launched, because the graduated-coin dumps on their earlier records now count.

Adverse at birth, then graduated — 156 coins. E.g. a coin that peaked at $3.45M and graduated, launched by a dev who was adverse at birth under both rules. The record holds both sides of a wallet like that; the grade at any moment showed only one.

Known Limits

●

Undetermined is the majority. 21,213 of 24,047 backfilled coins (88%) carry no determined ending; every rate here is conditional on determinability.

●

Items are deep on 194 coins and silent on the rest. No disposal was measured for the remaining coins, so no item exists — that silence is not a clean finding.

●

Sampled, single-account conduct reads. At most 100 signatures on the birth account; exits through other wallets, or below the sampling floor, are not measured. A clean conduct check means no drawdown was observed in the sampled window — not that no exit occurred.

●

Two aggravators are untested on history. Deception and hidden position have never fired or failed on a historical coin; the data to test them does not exist before platform launches. The historical adverse count is an extraction-and-abandonment count.

●

Residual risk. A patient, well-funded actor can still rotate wallets funded through exchanges, so provenance reads unresolved forever — and unresolved is not adverse, by our own rule; farm a record by behaving honestly for a season, since farmed honesty and real honesty are the same evidence until the last act; then extract once, slowly, under every published line. V2 raises his price — runway, idle capital, fee toll, and a public record his one extraction burns in full. It does not catch him in advance. The missing link was memory, not omniscience.

PROPOSED — Awaiting Ratification

Not settled, and marked PROPOSED on every record they touch. Each defaults to the most conservative reading the desk could implement.

B2's reading of "no subsequent activity" — no post-completion increase in position; remainder disposals and zero-change settlement transactions count as the exit's tail. Under the strict variant (any post-completion balance change counts), 179 items would be adverse instead of 188.

The abandoned-with-holders proxy — the peak-collapse test read as that ending class for aggravator purposes.

Slow-exit thresholds — 50% disposal floor; completion window after 60 minutes, within 7 days.

The episode completion definition — the earliest observed reading at or below 10% of baseline after which no later observed reading exceeds 10%, within the observed segment.

Corrections

Corrections annotate the record; they never overwrite it. V2 is itself a correction layer applied through that policy: all 194 confirmed-disposal records were annotated with their aggravator-test outcomes — the original stamps' endings, evidence, and pass fields untouched — and every grade was recalculated under the v2 rule, with the deltas published above. If the evidence no longer supports a finding, the finding is WITHDRAWN, clearly marked, and no longer counted; affected grades are recalculated.

LAUNCHKEY — METHODOLOGY ANNEX v2