Methodology

Observed

Read from a source by a validated capability, with its own as-of time.

Derived

Computed by a published method from observed values; the inputs are in the bundle.

Stale

A value older than its maximum age at the release cutoff. Shown with its age, never refreshed silently.

Insufficient data

The method applies but its inputs are incomplete. Never shown as zero.

Unavailable

No validated, enabled capability provides the input. Never shown as zero.

Not applicable

The measure does not apply (for example the bottoming test outside a drawdown).

Availability time

When an observation was first committed to the research record. Historical views use only observations available at their cutoff, so later data never rewrites what was known.

Releases

Each release is immutable and content-addressed; the website and the API read the same release. A correction is a new release, and weekly editions get a new revision.

Summary

An auditable three-condition state machine evaluated point-in-time on UTC-day intervals. It describes what has been observed in a drawdown; it does not predict or confirm a bottom.

Interval

Seconds

86400

Source

daily OHLC candles (codex.token-bars resolution 1D, or a validated x4o2 candle capability)

Note

Daily snapshot prices are point samples, not candles: they can open an episode and measure drawdown, but they never establish a held low (no observed lows).

Episode

Peak Lookback Intervals

30

Open Drawdown

0.5

Definition

At evaluation time t, peak = max(high) over the lookback intervals that ended at or before t (point-in-time). An episode opens when close(t) <= (1 - openDrawdown) x peak. Its id is <tokenId>@<peakStart>. The episode low is min(low) since the peak, using only intervals ended by t.

Close

The episode ends when a later interval's high exceeds the episode peak (superseded by a new high) or when no adequately observed interval exists for staleIntervals consecutive intervals (stale).

Stale Intervals

7

Conditions

Capitulation

Name

Material realized loss-taking during the drawdown

Window

From the episode peak to t.

Formula

lossSoldShare = soldValueUsd(loss-making sales with known cost) / soldValueUsd(all sales with known cost); realizedLossUsd = sum(cost - proceeds) over loss-making sales with known cost.

Met

costCoverage >= 0.8 AND lossSoldShare >= 0.6 AND realizedLossUsd >= max(25000, 0.05 x mean circulating market cap over the window)

Cost Coverage

soldValueUsd with known acquisition cost / total soldValueUsd. Coverage is by sold value, not by number of wallets returned.

Accounting

Cost Basis

FIFO per wallet within the token.

Transfers In

Unknown cost (never zero, never inferred from the sender).

Starting Inventory

Balance held before the first observed trade has unknown cost.

Pricing

USD value at transaction time from the trade record; current prices are never used.

Fees

Proceeds and costs net of fees when the trade record states them; otherwise gross, disclosed.

Source Reported Pnl

Provider P&L, cost basis and unrealized P&L are context only and never substitute for reconstructed realized losses.

Insufficient

costCoverage < 0.8, no wallet-resolved trades, or an unvalidated trade capability.

Low holds

Name

The episode low holds

Formula

Count consecutive adequately observed intervals after the interval that set the episode low whose low >= episode low.

Met

count >= 3

Adequate

An interval with high, low and close present and USD volume >= 1000. A missing or thin interval resets the count; it never counts as holding.

Insufficient

No candles (only snapshot prices) or fewer than 3 intervals after the low.

Absorption

Name

Buying absorbs selling and first-seen buyers participate

Baseline

The drawdown leg: intervals from the peak to the interval that set the low.

Formula

absorption = buyVolumeUsd / sellVolumeUsd; firstSeenShare = buyers never seen buying this token before / all buyers (wallet-resolved).

Met

absorption(after low) >= 1.2 x absorption(baseline) AND firstSeenShare(after low) >= 1.1 x firstSeenShare(baseline), each over at least 3 adequately observed intervals with flow coverage >= 0.8 of volume.

Insufficient

No buy/sell split or no wallet-resolved buyers.

Stage

Counted in order: stage = number of leading conditions met. A later condition never advances the stage while an earlier one is unmet; every condition is still shown independently.

States

Not applicable

No open episode: the coin is not in a qualifying drawdown at t.

Insufficient data

An episode is open (or candles are missing) but condition 1 cannot be evaluated. Never shown as 0/3.

Stage 0

In a qualifying drawdown; condition 1 evaluated and not met.

Stage 1

Realized loss-taking observed.

Stage 2

Realized loss-taking observed; the low has held.

Stage 3

Realized loss-taking observed; the low has held; buying absorption and new-buyer participation improved.

Reset Policy

A new lower low (low < episode low) resets condition 2 and condition 3 to unmet and starts their windows at the new low. Condition 1 is re-evaluated over the whole episode window. Each reset is recorded as an invalidation in the condition history.

Look Ahead

Every evaluation uses only intervals that ended, and observations that were available, at or before t. Later lows or recoveries never relabel earlier evaluations; histories are append-only per evaluation time.

Labels

Descriptive only: 'Capitulation observed', 'Low holding', 'Absorption improving'. Never 'bottom confirmed', never a return probability.

Changelog

• Version

  bottoming-v1.0.0

  Summary

  Initial method.

• Version

  bottoming-v1.0.1

  Summary

  secondsSinceLow is measured from the close of the low's interval and is zero while that interval is still open (sample-based episodes could report a negative value). States and stages are unchanged.

Changes

Holder-analysis-v1.1.0

Components 1, 2, 4-7 compute from stored holder panels, wallet-resolved trades, provider candles and address classes when those inputs exist; values for meaningful-holder growth and whale accumulation are now defined as ratios; trade coverage rules added. Results are unchanged while no such input is collected.

Summary

Nine components in four areas. Raw values, histories, peer comparisons and coverage come first; a composite is a separate, versioned, gated framework that is disabled until its inputs are defensible.

Inputs

Panels

Holder panels are stored with the population the source declares (holder_panels, balance_checkpoints). A panel is complete only when the source says it covers every holder; no x4o2 panel does. Only token-unit amounts are used; USD values are never converted back into balances.

Trades

Wallet-resolved trades come from one tape per coin whose responses are complete between the oldest returned trade and the read time (trade_coverage). Coverage of a window is the share of it inside those spans. Filtered feeds (FOMO activity) are context, never flow inputs.

Supply

Circulating supply = circulating market cap / price from the same Codex snapshot.

Address Classes

docs/data/address-labels.json (evidence-backed) plus the pool addresses LAMBLE tracks for the coin. Unclassified pools and programs stay in the population and the coverage note says so.

Checkpoints

Two checkpoints about 7 days apart, each within checkpointToleranceHours (default 24) of its target; the later within the component's maxAgeSeconds.

Exclusions

Policy

Excluded from holder populations only when classified with evidence: liquidity pools, burn addresses, custody/exchange wallets, lockups/vesting contracts and the token's own program accounts. A contract address is not excluded merely for being a contract (smart wallets hold tokens). Heuristic ownership clusters are reported separately and never merged into verified labels.

Sensitivity

Concentration is reported with and without each exclusion class so the effect of every exclusion is visible.

Populations

All holders

Every balance of the token on its chain at a checkpoint. Required for any measure labelled 'all holders'.

Top holders sample

The largest positions a source tracks (for example pump.holders, at most 100). Valid for top-N concentration when total supply is known; never for holder counts or retention of all holders.

Fomo users

Holders who trade on FOMO (fomo.holders). Reported only as 'FOMO users'.

Pump users

Positions held by pump.fun accounts (pump.holders). Verified 2026-10-20: for PIERRE the route listed 526 positions against 3,890 on-chain holders and a top-10 share of 10.6% against 25.2% on chain. Never a top-holder sample; reported only as 'pump.fun users'.

Recent trades sample

The most recent trades a tape returns. Valid only for the window it covers.

Components

• Id

  persistence.drawdown_retention

  Area

  persistence

  Label

  Holdings retained through drawdowns

  Formula

  For a fixed cohort (holders at the episode peak checkpoint), sum(min(balance at trough, balance at peak)) / sum(balance at peak). Endpoint retention: a wallet that sold and rebought between checkpoints counts as retained only up to its trough balance; continuity is reported separately.

  Requires

  all_holders balance checkpoints at peak and trough

  Min Coverage

  0.9

  Max Age Seconds

  172800

• Id

  persistence.balance_retention

  Area

  persistence

  Label

  Balance retention and uninterrupted holding

  Formula

  endpointRetention = sum(min(b1, b0)) / sum(b0) over the cohort at t0. uninterruptedShare = share of the cohort's t0 balance held by wallets whose wallet-resolved history shows the balance never fell to zero between t0 and t1. Buy 100, sell 100, buy 100 is endpoint-retained but not uninterrupted.

  Requires

  all_holders balance checkpoints, wallet-resolved transfer or trade history for continuity

  Temporal Resolution

  Checkpoint spacing is disclosed; exits and re-entries between checkpoints are invisible without transfer history.

  Min Coverage

  0.9

  Max Age Seconds

  172800

• Id

  persistence.portfolio_allocation

  Area

  persistence

  Label

  Allocation to the token within the measured wallet portfolio

  Formula

  token value / value of all measured assets in the same wallets, on the measured networks only. Never described as a share of anyone's whole portfolio.

  Requires

  wallet balances across measured assets (no validated capability yet)

  Min Coverage

  0.8

  Max Age Seconds

  172800

• Id

  demand.meaningful_holder_growth

  Area

  demand

  Label

  Meaningful-holder growth

  Formula

  entrants = wallets at or above usdThreshold at t1 that were below it at t0 AND whose token balance increased. priceCrossers (balance unchanged or lower, crossed only because price rose) are reported separately and never counted as entrants. value = (after - priceCrossers - before) / before, where before and after count wallets at or above usdThreshold at each checkpoint (prices: snapshots within 12 hours of each checkpoint).

  Parameters

  Usd Threshold

  100

  Requires

  all_holders balance checkpoints, price at both checkpoints

  Min Coverage

  0.9

  Max Age Seconds

  172800

• Id

  demand.whale_accumulation

  Area

  demand

  Label

  Accumulation by a fixed whale cohort

  Formula

  Cohort = top 20 non-excluded wallets of a top-holder or all-holder panel about 7 days before the release, fixed for the comparison. netPurchases = sum(buy token amounts) - sum(sell token amounts) from wallet-resolved trades only; value = netPurchases / the cohort's starting balance. Transfers, airdrops and rebases change balances but are never purchases; balance change is reported separately. Requires trade coverage of the window >= minCoverage and token amounts on every cohort trade.

  Parameters

  Cohort Size

  20

  Requires

  wallet-resolved trades, holder ranking at window start

  Min Coverage

  0.8

  Max Age Seconds

  172800

• Id

  demand.dip_absorption

  Area

  demand

  Label

  Buying that absorbs selling during dips

  Formula

  Over daily intervals in the last lookbackDays whose close fell >= dipThreshold versus the prior day's close: sum(buyVolumeUsd) / sum(sellVolumeUsd) from wallet-resolved trades. A dip day counts only when its flow coverage (time inside complete trade spans, capped by traded USD against the day's total volume) is >= minCoverage, and at least minCoverage of dip days must count. No dip day: not_applicable.

  Parameters

  Dip Threshold

  0.05

  Lookback Days

  30

  Requires

  buy/sell volume split per interval

  Min Coverage

  0.8

  Max Age Seconds

  172800

• Id

  health.concentration

  Area

  market_health

  Label

  Concentration under documented exclusions

  Formula

  top10Share = sum(top 10 non-excluded balances) / circulating supply. Valid from a top-holders sample when supply is known; reported with each exclusion class toggled.

  Requires

  top_holders_sample or all_holders balances, supply, address classifications

  Min Coverage

  1

  Max Age Seconds

  86400

• Id

  health.liquidity_to_size

  Area

  market_health

  Label

  Liquidity relative to size

  Formula

  liquidityUsd / circulatingMarketCapUsd from the same snapshot (codex.token-snapshot). Both are the source's figures for the token's top pair; circulating market cap, not FDV.

  Requires

  codex.token-snapshot

  Min Coverage

  1

  Max Age Seconds

  93600

• Id

  relative.peer_performance

  Area

  relative_strength

  Label

  Performance against explicitly defined peers

  Formula

  excessReturn7d = token 7-day price return - median 7-day return of the peer set. Peers = the other open members of the token's narratives on the evaluation date, each listed. Returns come from snapshot prices taken 7 days apart (+/- 12 hours); a token or peer without both prices is excluded and the peer count is reported.

  Parameters

  Horizon Days

  7

  Tolerance Hours

  12

  Min Peers

  3

  Requires

  codex.token-snapshot

  Min Coverage

  0.6

  Max Age Seconds

  93600

Composite

Version

holder-composite-v0.1.0-experimental

Enabled

false

Disabled Reason

Seven of nine components have no validated input capability yet (no holder panels, wallet-resolved trades or portfolio data are collected; the store and methods accept them once a capability is enabled). A score over two market-health/relative components would not measure holders.

Normalization

Persistence.drawdown retention

Zero At

0.3

Hundred At

0.95

Persistence.balance retention

Zero At

0.4

Hundred At

0.98

Persistence.portfolio allocation

Zero At

0.01

Hundred At

0.3

Demand.meaningful holder growth

Zero At

-0.2

Hundred At

0.3

Demand.whale accumulation

Zero At

-0.1

Hundred At

0.1

Demand.dip absorption

Zero At

0.6

Hundred At

1.6

Health.concentration

Zero At

0.8

Hundred At

0.2

Health.liquidity to size

Zero At

0.02

Hundred At

0.3

Relative.peer performance

Zero At

-0.5

Hundred At

0.5

Weights

Persistence.drawdown retention

0.15

Persistence.balance retention

0.15

Persistence.portfolio allocation

0.05

Demand.meaningful holder growth

0.15

Demand.whale accumulation

0.1

Demand.dip absorption

0.1

Health.concentration

0.1

Health.liquidity to size

0.1

Relative.peer performance

0.1

Gate

Min Components

7

Required Areas

persistence, demand, market_health, relative_strength

Max Bound Width

15

Missing Policy

Missing components never raise a score: the score is published only when the gate passes, and then as a bound [missing at 0, missing at 100]; the point value is shown only when the bound is narrower than maxBoundWidth.

Interpretation

A relative description of holder structure on fixed references. Not a return probability and not a recommendation.

Lifecycle

States

Collected

LAMBLE's daily run collects market data for the narrative's members.

Collection paused

Collection stopped (for example a replacement at the six-narrative cap). Memberships stay open; pages stay reachable; gaps after the pause are shown as gaps.

Retired

The defining source is gone or contradicted. Memberships close with the reason.

Featured

Featured placement (homepage, at most six) and collection priority are separate from lifecycle and from factual membership.

Trend

For collected narratives: 'rising' when 7-day volume of the same basket is >= 25% above the prior 7 days, 'fading' when <= -25%, otherwise 'steady'; 'new' when first observed within 7 days. Computed on the current basket and labelled so.

Membership

Factual

A coin's membership is an evidence-backed classification of what the coin is (name, metadata or official source naming the story). It is not affiliation with a named project or person.

Intervals

Each interval has a unique id (narrative:token@start), its effective start, when it was recorded, and (only for factual exits) its end. Collection pauses and inactivity archive a member without ending the interval.

Reconstructed

Members added with historicalMembershipBasis=reconstructed_today were classified when first observed; volume before that date is today's classification applied backwards.

Metrics

Volume Share

Share of the narrative's measured volume by member (all indexed pools per token, Codex getTokenBars). Market-volume share, not social attention.

Concentration

top1Share and Herfindahl-Hirschman index (sum of squared shares, 0-10000) of member volume over the window.

Breadth

Members with a positive 7-day price return / members with prices at both ends; the denominator is shown.

Attention

X results are a top-results convenience sample: counts of posts are reported only as 'posts found in N searches', never as mindshare.

Evidence

Project claims (official pages and posts), independent evidence (news, third-party reporting) and observed on-chain behavior are listed separately. Reposts or quotes of one claim count once.

Dedup

Market totals across narratives count each token once; the overlap (tokens in more than one narrative) is listed with its volume.

Capital Flows

Turnover and market-cap changes are not net capital inflows; no inflow figures are published.

Calendar

ISO weeks: Monday 00:00 UTC to the following Monday 00:00 UTC, end exclusive. Edition id = ISO week id (2026-W42). The comparison week is the immediately preceding calendar week.

Views

Calendar

A complete calendar week, frozen when it passes the completeness checks.

Developing

The current calendar week so far (Monday 00:00 UTC to the data cutoff), always labelled partial.

Rolling7d

The seven days ending at the latest data cutoff; never mixed with calendar weeks.

Historical Modes

As known then

Only observations with available_at <= the cutoff and memberships recorded by the cutoff. The cutoff of a frozen edition is its data cutoff; for a past week it can be set to the week end to show what was knowable then.

Same cohort

A declared fixed set (the members open at the comparison week's start, plus entrants named separately) measured in both weeks; members that died or were archived stay in the set with their (possibly zero or missing) values.

Current basket

Today's open members applied to both weeks retrospectively. Labelled as reconstructed; it is not what was known then.

Completeness

Fees

Every included venue (fee policy of the edition's financial run) has a settled value for all 7 days of the week; otherwise the fee total is partial and the missing venue-days are listed.

Narrative Volume

Every member in the mode's basket has a daily volume bar for every day it existed in the week; a missing bar makes that narrative's total partial.

Runs

At least 5 daily research runs available within the week plus one run after it whose financial anchor reaches the week end.

Publication

An edition is published as complete only when the fee and run checks pass; otherwise it is published as partial with the failing checks listed, or delayed.

Aggregation

Flows

Sum daily flows over the 7 days: venue fees, revenue, supply-side revenue, token volume.

Stocks

Never summed over time: holder counts, market caps, liquidity, prices. Reported as end-of-week (last observation at or before the week end) and start-of-week values.

Rolling

Rolling-24h and rolling-7d provider observations are never summed into weekly totals. Indexed launch counts are reported as the provider's rolling 7-day value nearest the week end, labelled with its own window.

Ratios

Period completions / period creations is a period ratio, not a cohort graduation probability, and can exceed 100%.

Fees

Covered fee subtotal uses the run's fee policy (non-overlapping streams only). Excluded venues are listed with reasons and shown individually, never added.

Multi Label

A token in several narratives counts once in deduplicated market totals; per-narrative totals overlap and the overlap is disclosed.

Scope Changes

When the included venue set or a narrative basket differs between the two weeks, the comparison is shown on the intersection and the scope change is listed; the denominator is never silently shrunk.

Revisions

A frozen edition is never edited. A later build of the same week whose content differs (provider restatement, late data, correction) is written as revision r+1 with a list of what changed and why; earlier revisions stay published.

Watch

Next-week watch conditions are generated deterministically from rule thresholds (registry narrativeRules and method thresholds), not from forecasts.

Comparison

Week-over-week narrative volume change is computed on the comparable basket (members with complete daily data in both weeks) and reported only when that basket has at least 3 members and at least half of the members present in both weeks; otherwise the change is null and the basket is still shown.

Changelog

• Version

  weekly-v1.0.0

  Summary

  Initial method.

• Version

  weekly-v1.1.0

  Summary

  Narrative change requires an adequate comparable basket (>=3 members and >=50% of members present in both weeks); watch subjects carry a short token id so coins sharing a symbol are distinguishable. Earlier editions are revised (r+1), not edited.

0 bottoming signal(s) at stage 1 or higher so far.

• Horizon 1 day(s)

  No performance claim: 0 outcome(s), below the 30 needed.

  Baseline “drawdown-episode-open”: n=112, median net return −15.4%, failure rate 46.4% (confidence moderate).

• Horizon 7 day(s)

  No performance claim: 0 outcome(s), below the 30 needed.

  Baseline “drawdown-episode-open”: n=32, median net return −54.5%, failure rate 78.1% (confidence low).

None of these figures is a forecast or a recommendation. The bottoming test describes what has been observed in a drawdown; see sources for which inputs exist today.