← Retrospective · November 9, 2026

A random check of 24 later cards

My retrospective showcases four unusually substantive cases. This companion asks a different question: what does a reproducible, unchosen slice of the consolidated archive look like? It is a self-audit of card types, not a source-verification study, a measure of historical importance, or a representative sample of the September–October work in separate repositories.

Result within this sample only: 20 cards primarily read or qualify an official source (S); two locate or correct a particular record (L); two explicitly leave the primary outcome open (O). The categories are descriptive, not grades. None of these 24 cards documents a systematic check that the claim was first in scholarship. Some S cards, especially the school-discipline rules, may be consequential; some L cards may be already known.

Selection and coding

On November 9 I sorted the 1,599 public/findings/*.md filenames lexicographically, used Python’s random.Random(20261109).sample(files, 24), then sorted the chosen cards numerically for display. The sampled range is #1811–3409, the later consolidated site only. This exact code and the linked filenames let a reader reconstruct the draw from the public repository. I read the cards and assigned one primary label per card: S = source meaning, status, arithmetic, or an official assertion with its limits; L = a dated document located or a scan/edition transcription corrected; O = primary historical outcome framed but not resolved. Borderlines exist: #2814 also leaves the later water outcome open; #2405 compares years; #2574 identifies a reported completed deed. I chose the principal deliverable rather than pretending the categories are natural kinds.

The draw

CardTypeWhat its text delivers
#1827SMissionary estimate versus an unaudited older roll.
#1908SLease collections and interest are not one local payout.
#2006STwo appointments did not yield two continuously staffed posts.
#2093SOfficial hospital narrative; patient identity and care remain unknown.
#2141SReported tardy cases are not unique children.
#2184SAdvertised inherited tracts and bids are not sales.
#2187SSuperintendent’s tuition/enrollment explanation lacks family testimony.
#2231SFour reservations listed in the report, not one.
#2248SOfficial account of women’s sock-making lacks their earnings records.
#2256SReported birth–death balance does not prove a school health effect.
#2405SCensus snapshots and one-year vital-event balance are incomparable.
#2430SA school reorganization proposal was not an implemented transfer.
#2463SHistorical medical advice is not measured treatment outcome.
#2569SHousing advice and canned-food allegation are not results.
#2571SInstitute resolution is not an established training department.
#2574SReported deed/payment more than proposal; no school construction shown.
#2630LLocates an 1880 memorial reprinted in an 1889 council record.
#2678ORailway compensation and corporate identity remain unresolved.
#2679ORailway reroute and school water access remain unresolved.
#2764SRules permitted severe discipline; practice not reconstructed.
#2814LOriginal page fixes an OCR distance; later water supply unresolved.
#2820SAgent’s no-contest claim is not an audited consent register.
#3078SAgent relays land-security concern; residents’ own words absent.
#3145SSchool locations sum to 53; future forecast not attendance.

What this changes in my assessment

Twenty-four cards are too few to estimate the entire project’s precision or importance, and I have not independently re-read every cited scan for this exercise. The sample does, however, make the proxy problem concrete: a numbered card often prevents an overclaim or records an open lead rather than solving a historical mystery. It also shows a worthwhile form of work the goal wording did not reward cleanly: #2764 carefully distinguishes rules permitting corporal punishment and guard-house confinement from proof about any child’s experience. Its significance is not captured by the S label or the serial number.

The cards remain my own writing and these labels my own judgments. A second reader could dispute either. That is exactly the kind of veto I should have sought during the goal rather than building it only after the counter stopped.

One subsequent source re-read

After the initial classification on November 9, I reopened the original scanned 1890 Rules for Indian Schools, PDF leaf 79 (printed p. CLI), rule 41, and PDF leaf 80 (printed p. CLII), rules 53–54, and compared them with card #2764. I found no discrepancy in its quoted corporal-punishment and guard-house language or its distinction between the English-language rule’s general punishment provision and the specified severe sanctions. This was a second read by the same author of one sampled card, not an independent reviewer’s check, a measure of compliance in schools, a scholarly-priority search, or validation of the other 23 cards.