×
DATE-INDEPENDENT LOCAL ORDER
UNAUDITED. No record-level date-independent local-order audit has been performed.
Foreign instrument 01 · distributed systems meets archival correspondence
Darwin Had No Global Clock
UNDERDETERMINED
No target clock can be built from the audited darwin-correspondence export.
The pinned Epsilon darwin-correspondence collection has 15,238 letters, but its bulk metadata does not establish date-independent local order or matched receive events. Date provenance is unaudited, so an independent target is also unavailable. The separate 1,536-letter darwin-family-letters collection is outside this audit.
First make the clock work where it belongs. Then carry it into the scoped archive collection and press start. The refusal is the result.
I · Home calibration
Make fifteen events tell time
Lamport's 1978 Figure 1 supplies three processes, their local event sequences, and six sent messages. Mattern's 1989 vector rule labels the same history. Press once to recompute all vectors, all 210 ordered distinct pair tests, and the two relations printed in Lamport's figure.
Lamport Figure 1, relabelled with Mattern vector time
Waiting to run.
p1 ? r4published p1 relation
p3 ? q3published p3 relation
? / 210vector order versus reachability
The local process arrows are part of the input. So are the six message pairings. The clock discovers neither. It turns that supplied causal history into labels that can travel with events.
II · Foreign target
Now take away the clock's inputs
The pinned darwin-correspondence collection contains 15,238 letter XML files. Editorial dates are covered because they are the proposed target, and the separate 1,536-letter darwin-family-letters collection is outside this audit.
Target gate
×
MATCHED RECEIVE EVENTS
MISSING. 15,237 addressee descriptions contain 0 receive dates and yield zero accepted matched receive events.
×
INDEPENDENT TARGET
UNAUDITED. Of 7,219 unbracketed point dates, zero have been reviewed here for independent provenance.
Not tested yet. No chronology percentage will appear because no honest eligible-pair denominator exists.
Date sorting supplies hidden answers, while the other requirements remain missing or unaudited. The result is underdetermined, not zero percent.
III · The sophisticated dismissal
What does the clock add to the same-edge graph?
Mattern's Theorem 10 makes strict vector dominance equivalent to happened-before when both receive the same directed edges. The pair laboratory is therefore a secondary theorem and implementation check on Lamport's 15-event home graph, not a Darwin chronology result. Pick any two distinct home events and ask both representations.
Pair laboratory
Run the home calibration first.
pendingcomputed comparable unordered pairs
pendingcomputed incomparable unordered pairs
pendingcomputed same-edge disagreements
The stronger ordinary target method is documentary-critical interval chronology: datelines, postmarks, replies, textual references, publications, events, endorsements, paper, watermarks, and writing style. The catalogue covers 13,934 records with point dates and 1,304 with interval or open dates, but scoring it against its own catalogue would be circular. Its independent accuracy remains unaudited.
IV · Adapter audit
Every bridge, and what falls through it
Each button removes one mapping from the target input and rebuilds the evidence ledger. Rows with no accepted instances are reported as unexercised.
TEI @xml:id or CSV id → stable document key
Strict use: identifies a letter only.
Ablation:
LOSS: A document identifier is not an event time or a receive event.
sent persName key → candidate author process
Strict use: names the author process.
Ablation:
LOSS: Authorship collapses secretaries, coauthors, copying, dispatch, and memory into one actor.
received persName key → candidate destination process
Strict use: names the intended recipient.
Ablation:
LOSS: An addressee does not prove delivery, reading, or a time of receipt.
date-blind exact acknowledgment in B → constraint A to B
Strict use: allowed after human uniqueness review.
Ablation:
LOSS: A binary edge erases transmission route and receipt moment; ambiguous references are dropped.
provenance-audited reply relation → constraint A to B
Strict use: allowed only when provenance is independent of dates.
Ablation:
LOSS: Editorial judgment becomes one bit and may secretly depend on the target date.
manuscript dateline or observed postmark → held-out target interval
Strict use: never constructs or sorts the graph.
Ablation:
LOSS: A dateline follows a calendar convention and proves neither dispatch nor receipt.
No mapping disabled yet.
Dropped from the causal engine
- CSV sorting_date and sent date
- editorial abstract and footnotes in the strict result
- sender and recipient addresses
- shelfmark, language, extent, and keywords
- transcript text without an audited exact reference
- numeric suffix, CSV row order, and website result order
Conventions supplied by hand
- A letter begins as one composition or send event at its author process.
- A receive event exists only when receipt is recorded or independently proved.
- One exact uniquely resolvable acknowledgment can create one causal constraint.
- No edge comes from people, topic, geography, archive order, identifier order, or dates.
- Every accepted edge stores IDs, evidence, evidence type, reviewer, and provenance status.
- A cycle invalidates the adapter; it is never repaired by consulting dates.
- Incomparable means only that no causal path was observed.
V · Fault bench
Break what it can see, then what it cannot
A fault the engine claims to catch
Add r4 → p1 to a graph that already contains a path from p1 to r4. That creates a causal cycle.
Unmutated.
A fault it should miss
Change the date and place on a complete target-shaped fixture. Both versions enter the real adapter. The safe path must stay identical, while a deliberately leaky date-sorting control must change.
Unmutated.
The blind spot is a negative control with a positive control beside it. The verifier requires metadata isolation and also requires deliberate date leakage to be detectable.
VI · The check
The check that stands behind this page
Home results below are bound to a fresh browser computation. Target counts are named frozen source observations. The offline check independently reads the same public modules, reproduces view models and fault paths, and checks the frozen reduction, HTML bindings, frontmatter, share asset, and transfer manifest.
Home benchmark, live recomputation
pendingcomputed events across 3 processes
pendingcomputed unordered event pairs
pendingcomputed dominance versus path mismatches
Published checks: pending; pending.
| Event | Computed vector | Event | Computed vector | Event | Computed vector |
|---|---|---|---|---|---|
| p1 | pending | q1 | pending | r1 | pending |
| p2 | pending | q2 | pending | r2 | pending |
| p3 | pending | q3 | pending | r3 | pending |
| p4 | pending | q4 | pending | r4 | pending |
| q5 | pending | ||||
| q6 | pending | ||||
| q7 | pending |
Convention: each three-component clock starts at [0,0,0]. Tick the local component before each event; a receive then takes the componentwise maximum with the sent vector.
Target archive, frozen full-corpus audit
COMMIT ZIPHTTP 200 after redirects · 93,705,444 compressed bytes · 204,347,124 expanded bytes · 77,447 entries · SHA-256 1bee8e8e9ef311d84f1dba6f064a6a39a5b0bc665e2de568279e33f071e50813
LICENCECC BY-NC 4.0 · retrieved 10 September 2026
CSV2,348,852 bytes · 15,238 parsed records
| Observation | Count | Meaning for the transfer |
|---|---|---|
| darwin-correspondence letter XML files | 15,238 | the complete scoped collection |
| excluded darwin-family-letters files | 1,536 | a separate collection, named but not silently included |
| sent actions with a date | 15,238 | editorial target exists, but is excluded from clock input |
| received actions | 15,237 | recipient identity is usually present |
| received actions with a date | 0 | no receipt time or reading order |
| correspContext elements | 0 | no structured reply relation |
| structurally empty placeholder bodies | 15,238 | each body is semantically empty of transcript text |
| non-empty bodies | 0 | exact acknowledgments cannot be extracted here |
| editorial abstracts | 15,229 | available only as conjectural candidates in the toggle |
| reply-like abstract screen hits | 730 | lexical candidates, not accepted causal edges |
| CSV IDs matched to XML IDs | 15,238 | independent parser cross-check inside the repository |
The scout recorded d7a5261795882df3dbf24752649673bce04a4a0882258788d7fee8acc4d62485 for the moving main.zip response. This build fetched the commit-addressed archive instead. GitHub generated a different ZIP byte stream, so the source commit is the stable identity and both ZIP hashes are disclosed.
Transfer ledger
| Gate | Value | Assertion |
|---|---|---|
| home benchmark passed | pending | all vectors and 210 ordered pair checks |
| adapter double checked | pending | 15,238 CSV IDs match 15,238 TEI IDs |
| held-out target available | pending | date provenance has not been independently audited |
| baseline beaten or distinct | pending | same-edge identity is a diagnostic, not the native baseline |
| planted fault caught | pending | validator cycle must contain the injected reverse edge |
| known blind spot demonstrated | pending | real adapter ignores metadata, positive leakage control reacts |
What remains uncertain
Of 7,219 unbracketed point dates, this audit cannot distinguish manuscript datelines, postmarks, recipient endorsements, and later editorial normalization. It did not manually classify 15,229 abstracts or 730 reply-like screen hits, test 67,197 linked graphics, prove the three transcript spot-check links date-blind, or estimate independent accuracy for documentary-critical chronology.
In the three named transcript spot checks, two candidate reply constraints agree with displayed dates, but the reply text and editorial notes expose those dates. Supplying either of the two unsupported Darwin-local orders moves the illustrative three-pair agreement from 66.667% to 100%, a 33.333 percentage-point adapter sensitivity. Those figures are excluded from the strict result.
Written permission, a date-redacted relation audit, image-level rights metadata, and a provenance-stratified manuscript sample would settle those questions. A permission-cleared subset might support partial-order inference without a total order for every participant. It would not change the theorem-level identity between vector dominance and reachability on the same accepted graph.
Sources and boundaries
- Leslie Lamport, Time, Clocks, and the Ordering of Events in a Distributed System, 1978, Figure 1. Lamport defines happened-before and scalar logical clocks, not vector time.
- Friedemann Mattern, Virtual Time and Global States of Distributed Systems. The fetched file identifies itself as an edited version of the October 1988 workshop paper and carries 1989 copyright. The vector update rule and equivalence are credited here. Colin Fidge independently published vector timestamps in 1988.
- Cambridge Collections Epsilon data repository, pinned commit shown above, CC BY-NC 4.0. Only the separately published GitHub archive was bulk-read.
- Darwin Correspondence Project, Editorial policy and practice. It documents dating from replies, internal text, publications, external events, endorsements, paper, watermarks, and writing style.