A blank is not a bad score

California keeps a list of job training programs that public workforce money can pay for, the Eligible Training Provider List, or ETPL. A federal law, the Workforce Innovation and Opportunity Act (WIOA), requires every state to keep one and report what happened to the people who took each program. The U.S. Department of Labor publishes those reports as a scorecard at trainingproviderresults.gov (opens in a new tab).

Here is what that scorecard holds for California, as counted on September 18, 2026 by the coverage page (opens in a new tab) of Afterward, an independent project of mine that reads that record. It lists 3,266 California programs. For 1,209 of them it shows no completion rate, no employment rate, and no median earnings figure. For 42 of those it shows a count of people served or exited beside those empty cells. For the other 1,167 it shows no performance figures at all.

Workforce Pell is a new federal grant for short training programs. For its first three award years, it admits a program only if at least 70 percent of its students were employed in the second quarter after they left, as shown by administrative data such as wage records, the quarterly earnings reports employers file with the state. The scorecard already carries that measure, for a different group of students. In California it is filled in for 1,760 of 3,266 programs.

Those numbers are not a list of bad programs, and a blank is not a score. But a person choosing between two programs is working without that number, and so is an agency deciding where to spend a reporting effort. California's Employment Development Department (EDD) sends the public (opens in a new tab) to a search screen inside CalJOBS, the state's job-search site, where state policy says performance information can be searched. I found no file to download there or on the state's open data portal as of September 18, 2026, so the federal scorecard is the record I can count.

An earlier essay was about software that records a missing answer as a no. This one is about a public record that leaves the answer out.

Two gaps, and they are not the same gap

A cohort is the group of people a figure describes. A program whose record shows a cohort and an empty employment cell has a hole in its record. Where that hole is not a privacy rule or a federal quality hold at work, the fix is a reporting rule. A program with no cohort count and no measure has no record, and, unless the program is new to the list, the fix is getting a report into the federal record at all.

For 144 of the 581 providers Afterward counts in the record (its grouping of the names as filed), nothing is published for any program, and they account for 278 of those 1,209 programs (coverage page (opens in a new tab), September 18, 2026). Every other empty row belongs to a provider with something published for another program, so a reporting relationship already exists.

The measures split by what it takes to know them

The counts come mostly from a provider's own records. People served are published for 2,099 programs, people who left for 2,084, people who completed for 2,047, and credentials earned for 1,800 (coverage page (opens in a new tab), September 18, 2026).

Employment and earnings describe what happened after people left, which a provider can learn only by following up with them, or the state by looking them up in wage records (a wage-record match, run on the roster of people who left). The count employed two quarters after leaving is published for 1,766 programs, the employment rate for 1,760, and median earnings in that quarter for 1,384 (same page and date).

Every employment and earnings measure is published less often than every other measure: the lowest of the others, credentials earned at 1,800 (a count that itself runs up to a year past exit), sits above the highest employment count, 1,766. That describes the published data, not how well any match is done.

California's current ETPL policy, directive WSD25-02 (opens in a new tab), dated February 23, 2026, allows either route. It replaced a November 2021 directive that was in force when these figures were filed. I have not read that earlier one. A provider can send totals for each measure, or send a list of Social Security numbers and let EDD run the wage match. The scorecard does not mark which route any figure took.

Workforce Pell defines placement the same way, for now

Workforce Pell is a Pell Grant, the federal grant for low-income students, for programs of at least 8 but less than 15 weeks at schools approved for federal student aid. Congress created it in section 83002 of Public Law 119-21 (opens in a new tab), and the first award year began July 1, 2026.

To qualify, a program needs verified completion and job placement rates of at least 70 percent each. The statute measures placement "180 days after completion." For award years 2026-27 through 2028-29 the rule defines it as the share of students "employed during the second quarter after exiting the program, using administrative data, including wage records," certified by the Governor (34 CFR 690.94(a)(2)(i)(B) (opens in a new tab)). The final rule's preamble (opens in a new tab) says an institution "cannot self-certify job placement," and accepts a state contract with a private outcomes-tracking platform.

That interim measure is defined the same way as the scorecard's employment column. The Department of Education said (opens in a new tab) it proposed to "align directly with the WIOA indicator." The wording differs (WIOA says unsubsidized employment, 20 CFR 677.155(a)(1)(i) (opens in a new tab)), and the group of students differs. After 2028-29, only employment in the occupation the program trains for, or a comparable in-demand one, counts, measured from completion.

Employment is the one scorecard column that shares a measure with a Workforce Pell test, and in California it is filled in for 54 percent of programs (1,760 of 3,266 as read September 18, 2026; my arithmetic). That predicts nothing about any Workforce Pell program. I do not read the completion or earnings columns against Workforce Pell: its completion test and its price test, a cap on tuition, are defined differently (34 CFR 690.95 (opens in a new tab)).

California wrote the dependency into law

Education Code section 69872 (opens in a new tab), added by Senate Bill 135 (opens in a new tab) and in force since July 13, 2026, has the California Student Aid Commission (CSAC) make the Governor's part of the eligibility decision; the Department of Education still has to approve. CSAC (opens in a new tab) says its application portal opens October 1, 2026.

Section 69872(d)(2) says institutions "shall use data available under the authority provided in Section 1095 of the Unemployment Insurance Code." Subdivision (as) of section 1095 (opens in a new tab), the law on who may use the information EDD holds, lets the central offices of the state's three public college systems "access any relevant quarterly wage data" for Workforce Pell compliance.

An institution without a data-sharing agreement under section 1095 may only file a letter of interest, with CSAC and the Office of Cradle-to-Career Data, by December 15, 2026. Its application "may not be accepted" until legislation is adopted saying how equivalent data will reach it.

Nothing published shows whether missing agreements empty the scorecard's employment and earnings columns. The statute shows only that California makes access to wage records the precondition for applying.

Not every provider owes the same report

Apprenticeship programs registered under the National Apprenticeship Act are "not required to submit" performance information (20 CFR 677.230(b) (opens in a new tab)). On the coverage page as read September 18, 2026, 137 of the 160 programs filed under National Apprenticeship are among those 1,209, which is what a voluntary rule would produce.

The current directive exempts nobody else from reporting. It excuses some programs from its performance thresholds, which bind only private postsecondary providers, and that is an exemption from the standard, not from the filing. So a blank at a college is not that exemption being used. It may be a program too new to the list to owe figures yet (the directive asks for nothing from before a program was listed), or it may be something else, and the record cannot tell you which.

The reason is how the scorecard writes a blank. Wherever an outcome measure is missing it writes one sentinel, a placeholder value of -1, and its data dictionary (opens in a new tab), the file that defines each field, gives that value three causes: a sample too small to protect personal information, no data reported for the program, or "significant data quality issues" the Department of Labor found in the state's submission. The statute's privacy rule names no minimum group size (29 U.S.C. 3141(d)(6)(C) (opens in a new tab)), and I could not find the rules the Department says it applies before publishing.

Where privacy stops explaining

If small groups are behind an empty cell, the blank rate should fall as the group grows. That is the one test I can run without inventing a reason.

As read September 18, 2026, among the 117 programs whose own cohort had 1 to 10 leavers, the completion rate is blank for 11 (9 percent), the employment rate for 55 (47 percent), and median earnings for 70 (60 percent). Among the 361 with 251 or more, completion is blank for none, employment for 25 (7 percent), and median earnings for 91 (25 percent). The coverage page (opens in a new tab) counts only cohorts filed against a single program, so the 103 programs whose cohort it could not attribute to one (coverage.json (opens in a new tab)), and every program with no exit count, sit in no band.

Completion and employment behave the way a privacy rule predicts. The earnings column does not come down to meet them. The earnings figure is a median for the leavers who were employed (20 CFR 677.155(a)(1)(iii) (opens in a new tab)), a smaller group than the band a program sits in, so size may still explain part of it. I do not think it explains a quarter of the 361 programs in the largest band. Whatever else is keeping that column empty, the record does not say, and neither will I.

What Afterward refuses to do

Afterward (opens in a new tab) is built from public data and labeled beta in its README (opens in a new tab). It is not affiliated with the State of California. It reads the federal record and measures nothing itself.

A withheld or unreported measure is stored as empty and shown as "not reported," never as zero.

The project also publishes the programs in CTDL, the Credential Transparency Description Language, an open vocabulary kept by the nonprofit Credential Engine, with which Afterward is not affiliated. Nothing has been published to the Credential Registry, Credential Engine's public database. A separate validator I also wrote could check 4 of the 7 record types the export (opens in a new tab) uses (validation statement (opens in a new tab), August 7, 2026 snapshot). The other three are the outcome-statistics layer, which went unjudged, not approved.

Limits

The Workforce Pell population is not the scorecard population. Workforce Pell's completion and placement rates describe the students of one program of 8 to under 15 weeks, at a school approved for federal student aid, for each award year. Its price test narrows that to the program's Pell recipients who completed it. The scorecard describes everyone who took a WIOA-listed program over four program years. The claim here is about the measure, and about the administrative data it takes to know it, which in California means wage records. It is not a claim that one dataset feeds or predicts the other.

Afterward reads the scorecard's search feed, the data behind the site's own search. The Department of Labor's bulk download is an older file with different California rows, so counts from it will not match these.

There is no state file to check the federal record against, so I do not claim the two lists hold the same programs. The program-year window behind every figure, July 1, 2021 through June 30, 2025, is a sentence on the scorecard's About page (opens in a new tab), not a field in the data, and the data dictionary's cover still names program year 2022. Every source linked here was read on September 18, 2026, and the scorecard refreshes. I cannot separate the three causes behind -1 for any one program.

Where Afterward disagrees with the source, the source wins. I have read the directive and the regulations, and I have not sat in the office that runs the match.

What a state could publish

A state could publish, without waiting for anyone:

• Its ETPL as a downloadable file.
• In that file, the program year each figure covers.
• The denominator behind each rate. The bulk download carries it; the search feed does not.
• A reason beside every blank (small group, nothing reported, or held for quality), so a privacy rule at work stops looking like a missing filing.
• In California, which institutions a section 1095 data-sharing agreement covers, directly or through their system office, since that now decides who can apply.

The program year and the reason code are also the Department of Labor's to add to its scorecard. Certifying a 70 percent placement rate takes the same facts: how many people were on the roster, and how many of them the administrative data could find.

If you work on this data at EDD, CSAC, the California Workforce Development Board, or the Department of Labor, and I have misread an obligation or a field, tell me. The corrections link (opens in a new tab) on the coverage page opens a public issue.