Inertia

Inertia means tracking how much a case or wave is moving, from the record, over time. It is shown first as measured series, plain counts of what happened, and second as a Repleader index derived from those series.

The measured series come first

Every matter, court, case type, firm and company page leads with counts we observed in public records, each linked to its source, charted from the first date we hold to today:

New filings per week.
Docket entries per week.
Distinct plaintiff-side firms joining, per month.
Related cases spreading across districts: the number of distinct courts over time.
Rulings and agency actions, each dated on a timeline.

These are counts, never modeled. They are the headline numbers and the headline chart on every page.

The Repleader index comes second

Under the measured chart, pages show a line labeled "Repleader index (derived from the series above)", with its components table and a link to this page. The index is a number between 0 and 100 computed from the same records, labeled modeled, and versioned; this version is inertia-1.0. Leaderboards may sort by the index, but every row also shows the raw 7-day and 30-day counts behind it. Nothing on Repleader presents the index as a prediction.

Index backtest pending. The index has not yet been tested against past MDL centralizations. Until the backtest results are published on this page, the index stays unpublished and pages show the measured series only.

The rest of this page defines the index exactly.

Notation

t is the UTC as-of date.
Half-life weight: hl(d, H) = 2^(-(t - d) / H) for a date d on or before t, else 0. An event H days old counts half.
Weighted count: ew(dates, H) = sum of hl(d, H).
Log scale with a cap: log_norm(x, cap) = min(ln(1 + x) / ln(1 + cap), 1).

Docket events

Each docket entry is classified by the first matching pattern below (case-insensitive). Weight and direction feed the momentum part of the index. Terminal events end a case's momentum unless a later appeal is filed.

Event	Weight	Direction
`settlement_final`	terminal	ends
`dismissal_prejudice`	terminal	ends
`judgment`	terminal	ends
`jpml_cto`	6	up
`mdl_transfer`	6	up
`jpml_motion`	4	up
`class_cert_granted`	5	up
`class_cert_denied`	5	down
`class_cert_motion`	3	up
`mtd_granted`	6	down
`mtd_denied`	5	up
`mtd_filed`	0.8	up
`msj_granted`	5	down
`msj_denied`	4	up
`msj_filed`	0.8	up
`trial_set`	5	up
`settlement_notice`	4	up
`appeal`	1.5	up, clears a terminal event
`stay`	2	down
`dismissal`	1	down
`amended_complaint`	1.5	up
`removal`	1	up
`complaint`	2	up
`other`	0	none

Index for a case

For a docket i in matter m:

Part	Weight	Definition
A, activity	0.20	`log_norm(ew(entries of i, 14), 20)`
M, momentum	0.20	`clamp(sum(weight x direction x hl(entry date, 14)) / 10, -1, 1)`
S, spread of filings	0.15	`log_norm(ew(filing dates of dockets in m, 30), 25)`
F, plaintiff firms	0.10	`log_norm(ew(first appearance of each plaintiff-side firm in m, 30), 15)`
J, MDL stage	0.15	stage / 3, where 0 is none, 1 a transfer motion, 2 centralized, 3 trial set in the MDL
N, news	0.05	`log_norm(ew(linked headlines, 14), 10)`
R, regulators	0.10	`log_norm(ew(linked agency actions, 60), 3)`
Q, judge pace	0.05	1 minus the judge's percentile of median days to a first ruling on a motion to dismiss, summary judgment or class certification, among judges with 10 or more such dockets; 0.5 if unknown

raw = 0.20A + 0.20M + 0.15S + 0.10F + 0.15J + 0.05N + 0.10R + 0.05Q

After a terminal event, or a termination date with no later appeal, the index is multiplied by 0.2. The case index is round(100 x clamp(raw, 0, 1) x T, 1).

Index for a matter

C is the mean of the top 10 active case index values, divided by 100.

raw_w = 0.30C + 0.20 log_norm(ew(member filing dates, 30), 50) + 0.10 log_norm(ew(first appearance per court, 30), 15) + 0.05 log_norm(ew(first appearance per defendant, 30), 15) + 0.10 log_norm(ew(first appearance per plaintiff firm, 30), 20) + 0.10J + 0.05 log_norm(ew(headlines, 14), 20) + 0.10 log_norm(ew(agency actions, 60), 5)

After a final settlement approval or the end of the MDL, the index is multiplied by 0.3. The matter index is round(100 x clamp(raw_w, 0, 1) x T_w, 1).

Labels

The 7-day change in the index is labeled rising at +5 or more, falling at -5 or less, and steady otherwise. The raw 7-day and 30-day counts are shown beside every index value, and every input is stored with each daily value.

Backtest plan

Positives: groups of cases, built only from records dated on or before the as-of date, that include a docket transferred into an MDL within 180 days. JPML orders from 2019 to 2024.
Negatives: three per positive, matched on nature of suit and year, never centralized within 365 days, each with 5 or more entries.
Snapshots 30, 60 and 90 days before centralization. Tuning on 2019 to 2022 only; a single holdout report on 2023 to 2024: AUC at 60 days, top-decile precision, and median lead time.
If the holdout AUC is below 0.65, the index stays unpublished and pages show the measured series only. Even above that bar, the index is described as a summary of recorded activity, not a prediction.

The backtest results will be published on this page exactly as computed.