Matching FJC IDB docket numbers to CL opinions: gaps in docket search, empty clusters, and lead consolidated cases

[H-Hirsch]

Hi,

I'm building a pipeline that links FJC Integrated Database (IDB) appellate cases to CourtListener opinions by docket number. I'm using the Education Membership API. I have a few questions about cases I can't match.

1. Cases that exist in CL but aren't findable by docket number
Many cases return 0 results via /search/?docket_number=&court=&type=o and /dockets/?docket_number_core=, but ARE found by party name search. Examples in ca9:

• 17-55428 (core=17055428) — 0 results by docket, found by name
• 16-72186 (core=16072186) — 0 results by docket, found by name
• 21-17070 (core=21017070) — 0 results by docket, found by name
  Is there a known reason why some docket numbers aren't searchable by the docket_number or docket_number_core fields even when the case exists in CL? Is the core format (YY + 6-digit zero-padded sequence) consistent across all circuits?

2. Dockets found but clusters=[]
When I find a docket via docket_number_core, many recent cases have clusters=[] (the docket exists but no opinion is attached). Examples:

• 22-1284 (Duke Energy Progress v. FERC) — docket found, clusters=[]
• 19-1049 (NASDAQ v. SEC) — docket found, clusters=[]
• 18-1113 (Manhattan College v. NLRB) — docket found, clusters=[]
  Are these cases in PACER/RECAP only, not yet linked to the case law database? Is there a way to retrieve the opinion text for these cases via the API?

3. Lead consolidated cases not found
The IDB records a JOINAPP value for each case indicating its role in a consolidation. JOINAPP=2 cases are non-lead consolidated cases, and JOINAPP=3 cases are the designated lead case of a consolidated group. When I search CL using the lead docket number for JOINAPP=2 cases, they are found successfully. However, when I search for JOINAPP=3 lead cases directly by their own docket number, many return 0 results via both the search API and docket_number_core. Is there a known reason why lead consolidated dockets would be unfindable this way?

4. Match rate decline for recent cases
Match rates are strong for 2004–2016 (~85–95%) but decline for recent years (2020–2023: ~80%, 2025: ~59%). Is this a known coverage gap for recent appellate opinions, or likely a docket format issue?

Questions

1. Is there a more reliable way to look up cases by docket number that handles the gaps described above?
2. For dockets with clusters=[], is opinion text accessible anywhere in the API?
3. Is there a known reason why JOINAPP=3 lead consolidated dockets are unfindable via docket number search?
4. Is the recent-year match rate decline a known coverage issue?

Thank you in advance. I'm happy to provide more details or specific examples if that would help.

[mlissner]

1. No idea about this! Can you share some links that work or don't or should but don't?

2. Not sure about this, but it is, of course, possible to have dockets without decisions. That could be it. The other possibility is that we have duplicate dockets and the one with the case law attached to it isn't the one you're looking at. It's also possible we're missing the case law, but hopefully that's not the issue.

   You could work in the opposite direction though. Search the case law DB for the item, then grab the docket associated with it.

3. No idea about this, sorry!

4. Up until 2018, we benefit from the Harvard Case Law project, but not after that. Since then, we've scraped court websites, but this may miss things. We're scanning case law books now to fill in any real or perceived gaps, but that will take another six months or so to complete.

I hope this helps!

[H-Hirsch]

Thanks very much for responding, Mike!

1. Here are some concrete links as requested:

Working (found via docket number search): ca9 docket 03-74795 -> https://www.courtlistener.com/opinion/3052381/northwest-env-v-epa/

Not working (0 results via /search/?docket_number=&court=ca9&type=o): 17-55428, 16-72186, 21-17070 (no clickable URLs since CL returns no opinion for any of these).

2. I tried your suggested reverse approach – searching the case law DB by case name for 17-55428 (California Sea Urchin Commission v. Rachel Jacobson). This returns a different 2015 proceeding involving the same organization (v. Michael Bean, docket 15-56672: https://www.courtlistener.com/opinion/4487846/california-sea-urchin-comm-n-v-michael-bean/) but not the 2017 case. One complication with the reverse approach for my use case is that the IDB only stores truncated party names (e.g. “CALIFORNIA SEA URCHI” and “JACOBSON”), which aren’t reliable enough for a name search to consistently return the right case, especially when there are multiple proceedings involving the same parties. The docket number is a more precise identifier, which is why I rely on it.

I should note that the missing cases (17-55428 from 2017, 16-72186 from 2016) are actually before the 2018 Harvard cutoff, so the coverage gap explanation doesn’t quite apply here. Do you have any insight into why pre-2018 cases might be missing? Is Harvard’s coverage incomplete for certain court types or case categories?

3. No problem! I'll investigate further.

4. Good to know about the 2018 cutoff. I’ll document this as a known gap and revisit once the scanning is complete.

Thanks in advance for any clarity you can provide!

[H-Hirsch]

Hi @mlissner ,

I just wanted to check in and see if you had any answers to my previous follow-up questions (see above). My apologies for bugging you. I'd really appreciate any insights you can provide!

[johnhawkinson]

On this very narrow question:

Is the core format (YY + 6-digit zero-padded sequence) consistent across all circuits?

ACMS (the CM/ECF successor used by the 2nd Circuit and the 9th Circuit, initially rolled out only for immigration cases but then rolled out for all cases starting Oct. 1, 2023) does not pad its docket numbers. And indeed, historically CM/ECF docket numbers started at 1000 or 10000 and had structure/significance associated with the leading digit, and that is all gone. The first case is 26-1, then 26-2, etc., etc.

Or to put it differently: No.

Hopefully this is not too tangential.

[H-Hirsch]

I appreciate the response, John.

Understood on ACMS. Thanks for the heads-up!

[SamuelStentz]

Hey yall! Was digging into data as well and see a similar gap I think is mostly from the source group (likely from Harvard CAP dropoff).

@johnhawkinson Since then, we've scraped court websites, but this may miss things. We're scanning case law books now to fill in any real or perceived gaps, but that will take another six months or so to complete.

If there is a better central blog/update w.r.t. handling data completeness issues lmk!

Details

The gap — published clusters carrying their official reporter cite, by date_filed year:

filed year	Fed. Appellate w/ F.3d·F.4th	Fed. District w/ F. Supp.
2019	95%	93%
2020	~100%	8%
2022	99.5%	8%
2024	99.7%	1%
2025	99.8%	0%

Appellate reporter cites are essentially complete through 2024; published district cites fall off a cliff after 2019 (~90% → ~0%).

It tracks sourcing (opinions filed 2020–2025):

group	source C (court site)	source G (RECAP/PACER)	OCR'd	has page breaks (\f)
Fed. Appellate	100%	0%	0.5%	99%
Fed. District	1%	99%	55%	1%

Recent appellate comes from court-website scrapes with clean captions/dates; recent district is almost entirely RECAP/PACER. Every reporter cite in the snapshot was date_created in a single 2025-09 batch, which reached appellate through 2024 (~3–5K/yr) but only ~15 district clusters/yr after 2019 — so this reads as a matching/ingestion artifact tied to the RECAP-only sourcing (PACER dates ≠ decided dates, complex/ACMS docket numbers, truncated captions), not cites that simply don't exist yet.

Question

Is it likely resolving RECAP/PACER opinion -> official reporter cite will not be feasible for federal district courts?