Source: census_work · Run 20261016T060747Z · As of 2026-10-16T06:07:47Z. Aggregate source holdings; yield estimates are planning figures.
Holdings census executive report
Domain bocg, run 20261016T060747Z, 4 roots. Counts and digests only; detail tables stay in the run directory.
1. Headline
The roots hold 17,501 distinct units and 33,916 unique spans. In-domain text outside the already-measured base comes to 2,901,068 words, which the measured anchor turns into 609 raw episodes, 304 ratified after adjudication survival, at about 43.5 SME hours. The structural count from the span ledger is 1,233 candidate spans in 183 episode-shaped components; the two readings should be reconciled by the blind sample, not averaged.
| measure | value |
|---|---|
| distinct units | 17,501 |
| unique spans | 33,916 |
| in-domain words outside the measured base | 2,901,068 |
| raw episodes (word-anchored) | 609 |
| ratified episodes (x survival) | 304 |
| candidate spans (structural) | 1,233 |
| episode-shaped components | 183 |
| SME hours at the measured throughput | 43.5 |
2. What is on the machine
| root | units | distinct | in-domain units | in-domain words | outside base | dominant eras |
|---|---|---|---|---|---|---|
| archives | 583 | 570 | 10 | 1,492,722 | 230,677 | 2010-2014 2,443, 2020-2024 1,418, 2025+ 748 |
| working-folder | 230 | 127 | 42 | 404,572 | 137,627 | 2020-2024 2,345, 2025+ (mtime) 2 |
| forensic-mail | 15,499 | 14,244 | 3,419 | 1,419,434 | 1,419,434 | 2010-2014 8,957, 2005-2009 3,940, 2015-2019 2,681 |
| forensic-docs | 2,898 | 2,560 | 39 | 4,468,510 | 1,113,330 | 2015-2019 (mtime) 1,024, 2010-2014 412, 2010-2014 (mtime) 381 |
Documents counted but not read: 1,407. Bulk text units above the segmentation cap hold 3,355,180 in-domain words that do not count until segmented.
3. What can be assigned
Time: of 33,916 unique spans, 35% are dated in place (a date in the span or its message header), 7% are assignable by inheritance from a dated neighbour, thread or file name, and 58% cannot be dated from content. Precision: {'day': 12991, 'None': 19794, 'year': 936, 'month': 195}.
Institution: 12% assigned by declaration, 0% assignable from mail domains, 88% unassignable. Institutions seen: 4.
4. What links
Link degree over unique spans: 23,607 isolated, 4,092 with one or two links, 3,345 with three to nine, 2,872 with ten or more. 26,279 components, of which 408 are episode-shaped (two or more spans, two or more dated days, a decision). Component size bands: {'0': 23607, '10+': 221, '1-2': 1785, '3-9': 666}.
5. What it would yield, by SKU
Probabilistic, in SKU order. Expected counts over candidate spans with Poisson 95% intervals; base rates anchored on the measured stratum, structural weights declared, survival measured.
| SKU | expected | 95% interval | ratified (x survival) | spans with mass |
|---|---|---|---|---|
| trace | 108.0 | 87.7 to 128.4 | 54.0 | 3,781 |
| episode | 671.6 | 620.8 to 722.3 | 335.8 | 3,487 |
| outcome | 142.6 | 119.2 to 166.0 | 71.3 | 3,487 |
| atom | 3137.9 | 3028.1 to 3247.6 | 1568.9 | 3,781 |
Base: 2,376 spans in the measured base carry the anchor counts {'trace': 62, 'episode': 321, 'outcome': 56, 'atom': 2348}; provenance {'base_rate': 'anchored', 'weights': 'finger_in_air', 'survival': 'measured (click adjudication 10/20)'}.
Word-anchored cross-check and structure:
Word-anchored: 609 raw / 304 ratified episodes, 4,453 atoms, 106 consequential threads, 43.5 SME hours. Structural: 1,233 candidate spans (201,283 words) in 183 episode-shaped components. Complexity of unique spans: 29,267 low, 4,188 medium, 461 high (mean score 0.62).
| division | trace E | episode E | outcome E | atom E | word-anchored raw episodes |
|---|---|---|---|---|---|
| prime_brokerage_financing | 12.2 | 73.2 | 16.0 | 365.8 | 123 |
| em_trading | 2.6 | 15.8 | 3.1 | 79.2 | 73 |
| equities_trading | 11.2 | 68.9 | 15.0 | 311.6 | 70 |
| rates_trading | 10.0 | 62.9 | 13.5 | 293.1 | 68 |
| credit_trading | 13.3 | 83.7 | 17.0 | 416.0 | 64 |
| fx_trading | 7.5 | 48.8 | 10.0 | 220.1 | 47 |
| buyside_trading_execution | 5.3 | 32.5 | 7.0 | 171.5 | 43 |
| commodities_trading | 9.7 | 58.4 | 13.2 | 241.4 | 40 |
| trade_lifecycle_operations | 3.6 | 21.8 | 4.6 | 94.5 | 13 |
| listed_derivatives_client_clearing | 3.0 | 17.7 | 3.9 | 80.4 | 12 |
6. Quality and risk
- Duplication across spans: {'unique': 33916, 'exact-duplicate': 4308, 'near-duplicate': 4227, 'boilerplate': 3199, 'quoted-reply': 2440}. Unique share 71%.
- Chunking: words per span p10/p50/p90 41/178/317; spans per unit p50 2.
- Vocabulary terms excluded as generic in this run: office.
- Unread material: 1,407 documents; 3,355,180 bulk in-domain words awaiting segmentation.
- All magnitudes are planning figures anchored on one measured stratum. Everything censused is calibration base, never product.
7. Change since the previous run
Previous run 20261016T053121Z: 0 units added, 0 changed, 0 removed; in-domain words +0; raw episodes +0.
8. Decisions needed
- Adjudicate the blind sample so the structural and word-anchored readings are replaced by a measurement for the mailbox stratum.
- Decide whether bulk text units are segmented (they hold in-domain words that currently count for nothing) or excluded by rule.
- Decide whether the unread documents are worth text extraction, given their era and count.
- Confirm the institution declarations in the config; unassignable spans cannot be sold as any institution's calibration base.
Download Markdown · Public technical report (JSON) · Publication manifest