About

Methodology

About & Methodology

TACO — Data Provenance & Methodology

TACO ("Trump Always Chickens Out") tracks documented reversals, pauses, and quiet walk-backs of Trump-administration policies: a loud cut, ban, tariff, firing, or rule that is later undone — often unannounced.

This document defines how an entry qualifies, how it is sourced, and the editorial standards the dataset holds itself to. The authoritative data shape is data/schema.ts; the seed dataset is data/reversals.json.

1. Purpose

TACO is a factual accountability record, not advocacy. Each entry pairs a dated, sourced original action with a dated, sourced reversal action. The point is carried by the dates and the citations, not by framing. A reader should be able to follow every claim to a primary or reputable secondary source and verify it independently.

This is a personal, non-commercial homelab project.

2. Inclusion criteria — what qualifies as a "reversal"

An entry qualifies when all of the following hold:

  1. A prior Trump-administration action exists and is dated. It was announced, enacted, ordered, or operationally begun — a cut, ban, tariff, firing/RIF, rule, grant termination, stop-work order, funding freeze, or withdrawal.
  2. That action is later undone, paused, partially walked back, or quietly restored, and that reversal is dated.
  3. Both the original action and the reversal are sourced (see §4).

The reversal can be driven by the administration itself (rescission, waiver, negotiated deal, market pressure) or forced externally (court order, congressional appropriation). Court- and Congress-forced reversals are in scope because they still represent the prior action being undone — the mechanism field records how.

Status taxonomy

  • fully-reversed — the original action was completely undone.
  • partial — only part of the action was walked back, or it was restored with caveats/limits.
  • paused — temporarily suspended; the underlying policy persists.
  • quietly-restored — undone with no public statement.
  • re-reversed — reversed, then the reversal was itself undone (whipsaw).

Re-reversals

When an action is reversed and then re-imposed, the entry uses status re-reversed and the notes field narrates the sequence (e.g. a court-ordered reinstatement later stayed by a higher court, or Congress restoring funding the agency then dismantles anyway). One entry captures the whole arc rather than splitting it.

3. Exclusion criteria — what does NOT qualify

  • Normal policy evolution. A new position with no prior contradictory enacted/announced action is just policy, not a reversal.
  • Courts blocking something never enacted. If a policy was enjoined before it ever took effect, the administration did not reverse itself — the court prevented the action. Example deliberately excluded: the birthright-citizenship executive order was blocked by multiple courts, but the administration never walked it back; it remained the administration's stated position. Not a reversal.
  • Ongoing litigation with no reversal yet. A pending challenge is not a reversal until funds/jobs/rules are actually restored or the action withdrawn.
  • Pure rhetoric. Statements, threats, or trial balloons that were never enacted, then dropped.

These exclusions keep the dataset defensible: every entry has a concrete, dated before-and-after.

4. Sourcing standard

  • Minimum: ≥1 credible source for the original action and ≥1 for the reversal. Most entries carry two or more on at least one side.
  • Each source records url, outlet, and date (publication date, YYYY-MM-DD), plus the headline in title where available.
  • Outlet hierarchy (prefer higher tiers where accessible):
    1. Primary documents — agency press releases (.gov), the Federal Register, court orders/filings, company statements of record.
    2. Wire services — Reuters, AP.
    3. Major national outlets — NYT, WaPo, WSJ, NPR, CNN, NBC, CNBC, Politico, The Hill, Al Jazeera, Newsweek.
    4. Specialist/trade and analyst outlets — Science/AAAS, KFF, Devex, Higher Ed Dive, Utility Dive, Electrek, law-firm and policy-org alerts — used where they have the most precise reporting on a niche action.
  • Paywalls are acceptable. If the full article is paywalled, the entry still records url, outlet, date, and the headline in title, so the claim is attributable even when the body is not freely fetchable.
  • Dates. date fields are day-precision where a specific event date is known. Where only a month is firmly established (e.g. a budget request, a multi-ruling restoration that played out over weeks), the first of the month is used and the notes field explains the span. Source date is always the article's publication date and may post-date the event it describes.

5. Neutrality standard

  • Describe actions factually: program names, dollar figures, dates, agencies, legal mechanisms.
  • No editorial adjectives, no characterization of motive beyond what a source states. The dataset does not editorialize about whether a reversal was a "climbdown" or "chickening out" — that read is left to the reader and to the aggregate pattern.
  • The project name is a framing device for the collection; individual records stay neutral.

6. Known gaps

  • Not exhaustive. This is hand-curated OSINT, not a comprehensive index. There is no single upstream dataset for "quiet policy reversals," so coverage reflects what curators found and could source.
  • High-profile bias. Widely covered reversals (tariffs, marquee court fights) are over-represented relative to low-profile agency actions that received little press — precisely the quiet walk-backs hardest to detect.
  • Domestic-media bias. International outlets sometimes surface reversals of U.S. policy that domestic coverage misses; the seed set leans U.S. outlets.
  • Recency/whipsaw risk. Several entries are contested or partially undone again after the cited reversal (e.g. NIH grant non-renewal, Supreme Court stays). status, notes, and updated_at flag these; entries may need revision as events develop.

7. Refresh cadence

  • Manual curation. Additions and corrections land via pull request.
  • Each entry carries added_at and updated_at; updated_at is bumped whenever a record is revised (e.g. a partial reversal becomes re-reversed).
  • No automated ingestion. Every entry is reviewed against §2–§5 before merge.

8. Non-commercial use & attribution

This is a personal, non-commercial homelab project. All cited reporting remains the property of its respective outlets; URLs are included for attribution and verification only. TACO reproduces facts (dates, figures, program names) and links to sources rather than republishing article text.