The standard · Standing since August 2026

Find an error in our numbers and we will publish the correction with your name on it.

Anyone can produce confident prose now, which means confident prose proves nothing. The only claim PILLAR can make that a competitor cannot copy is that our figures survive being checked. So here is the log of every time they did not.

The offer, in full

If a published PILLAR figure is materially wrong, we correct it in public, name the person who found it, and say what we changed.

No form, no submission portal, no discretion about whether it is embarrassing. Send the figure, the page it is on, and what you think it should be. If you are right, it appears below with your name unless you ask us to leave it off.

01
What counts as material
Any figure that would change a reader's conclusion. A wrong dollar amount, a wrong district, a wrong direction of travel, a method that does not do what we said it does. Typography and prose are welcome too, but they do not go in the log.
02
Scope: everything, with no exclusions
Every figure PILLAR has ever published, on any page, going back to the first one. The Ledger, The Holdings, the growth strategies, the essays, the proposals we sent you. Nothing is carved out for being old, internal, or awkward. If a number is wrong, it is wrong whenever we wrote it.
03
What we do
Correct the figure at source, rebuild every page that carries it, and publish an entry here saying what it was, what it is, and how it was found. We do not quietly edit.
04
What you get
Attribution, and the correction in writing. There is no cash prize. If you want your name left out, say so and we will log it as found externally.
Send a correction

The log

Every material correction, newest first. Entries found by our own review are marked as such, because a log that only contains other people's catches is not honest about where errors come from.

8 August 2026 · The Ledger, Issue No. 001 FOUND INTERNALLY

Maryland figures restated against the reconciled vendor registry

Found during vendor identity work in PILLAR Vertical Intelligence, which surfaced roughly $45 million of Maryland payments that had not been matched to a vendor. The largest single cause was a misspelling in the source file itself, "curriculum associate llc", worth $15.8 million in Baltimore City. Holding-company and brand-only name strings accounted for the rest.

Every affected figure in the issue was restated rather than patched, and the two meta descriptions that had drifted were corrected at the same time. The direction of the issue's argument changed: FY2025 became the peak spending year rather than FY2022, which makes the post-stimulus story reallocation rather than contraction.

FigureAs publishedCorrected to
Headline disclosed total$1.04B$1.12B
Curriculum Associates, statewide$57.6M$73.4M
Pearson, statewide$91.6M$93.5M
Baltimore City, total wallet$124.7M$150M
Prince George's, total wallet$287M$297M
Supplemental decline from peak45%43%
Payment records counted555,855555,880

A limit was added to the method section at the same time: a district checkbook systematically under-observes assessment spending, because statewide summative contracts are signed by state agencies rather than districts. Cambium Assessment receives $0 across all 555,880 district payment records, which is a fact about the record and not about the company.

No externally reported corrections yet. This offer went up with the cut zone instrument in August 2026. If that line is still here in six months, either the figures are holding or nobody is checking, and only one of those is good news.

What we know about our own corpus

Making this offer retroactive with no exclusions means saying what shape the back catalogue is actually in. We audited it before publishing the offer rather than after.

What it isFiguresTraced to a public sourceBasis declaredNeither, yet
Published analysis
The Ledger, The Holdings, growth strategies		5,783100%—0%
Framework essays
The insights library		44650%50%0%
Client documents
Proposals, implementation plans, demos		1,67844%56%0%

The published analysis is the part we ask you to rely on, and it is the part that holds up. Everywhere else, a figure now says what kind of number it is. The essays declare their benchmarks as calibration from experience rather than measurement. A proposal says which figures we inferred about you from outside and which came from public records. An implementation plan says its account numbers are yours, read back to you and not independently verified. A product walkthrough says the deal in it is invented.

A declared basis is not a source, and the table keeps the two apart on purpose. Saying "this is our estimate" tells you how much weight to put on a number. It does not tell you the number is right. That is what the offer above is for, and it applies to every row.

As of 9 August 2026 there are none left. Every figure PILLAR publishes either traces to a public source or says on the page what kind of number it is. Published analysis is sourced outright. The essays, the newsletter and our own product pages declare their basis. Client documents say whose numbers they are. Four pages are excluded and say why: three are unpublished working drafts with no inbound links, and one reply essay's only figure is the title of the piece it answers.

Source: PILLAR published-figure provenance audit, run against the full site on 9 August 2026, covering 7,907 numeric claims across 407 pages. This page is excluded from its own counts, because reporting the numbers would otherwise change them. The auditor is a script; it checks whether a reader can find out where a figure came from, and it does not check whether the figure is true. That second question is what the offer above is for.

Why this exists

Generative writing tools made polished prose free to produce. A well-argued page is no longer evidence that anyone did the work, which is why long-form analysis now gets dismissed unread as machine output. That dismissal costs the reader nothing and is often wrong, but complaining about it changes nothing.

What does change something is making the work checkable. "A human worked hard on this" is a defence that anyone can offer, including someone who did not. A number that reconciles against a public file is a defence that only holds if the work was actually done.

So we publish the source, the vintage, and the method next to every figure, and we run an open invitation to prove us wrong. Fabricated numbers do not survive a spot check. That is the entire argument.