Findings
Counts, medians, durations and gaps computed from the observation series. Every finding states its method inline, so it can be redone against the same public data.
No finding computes a score, index, ranking or weighting. Weighting two measures against each other declares how the two should be ranked, which is a methodology rather than arithmetic. An unweighted count survives disagreement about which measure to prefer.
Limitation is written by the computation that produced the figure, at the time it ran — the population a filter dropped, a probe that was withheld, a list a cap truncated. It is not reconstructed afterwards from the published number: a bound written later by a reader of the output would be this site authoring a claim about its own arithmetic. Where a computation stated none, the field renders — rather than a sentence composed to fill it. 12 of 26 findings below carry one.
archive
11 of 11 declared API sources observed
Method. Count of declared API sources with at least one observation recorded. A source requiring a credential that is not configured is recorded as unobserved on every run rather than omitted, so the gap is visible in the series instead of resembling an empty result.
Limitation. —
Figures and evidence
Named figures: —
56 of 69 recorded candidate(s) resolved by rule, 13 with no verdict recorded, 0 confirmed by analysis, 0 confirmed by a person
Method. Every change candidate staged or ruled on here, by stage. A candidate's LATEST verdict decides its stage: one carries a person's confirmation, a chair-ordered withdrawal, a discard and a machine decline, and counting receipts would publish a person-confirmed item this record no longer holds. Rule classes come from the receipts' own words, so the vocabulary is the store's, not a list held here.
Limitation. Stages sum to the detected population and classes to the resolved count; where either stops, a state or class exists that no figure names. One name the order gave — WRONG REFERENT — has no member; one the store holds — DECLARED LISTING — the order did not name. Neither is mapped onto the other.
Figures and evidence
- awaiting
- 13
- class_attribute_residue
- 15
- class_declared_listing
- 1
- class_excluded_chrome
- 21
- class_oscillation
- 18
- class_recorded_as_a_change_of_the_witness
- 1
- confirmed_by_analysis
- 0
- confirmed_by_person
- 0
- detected
- 69
- resolved_by_rule
- 56
4 tracked document(s) changed, 6 of 152 observed record(s) not retrieved, 13 of 69 recorded candidate(s) with no verdict recorded
Method. The figures in the nightly sentence, re-derived here from the stores rather than read from the page. `changed` counts tracked rows whose DOCUMENT-class observation changed on the run's day carrying no decline receipt for it. `not_retrieved` counts records dated that night whose own error field is non-empty, the population the watcher prints. `pending_review` counts queue candidates with no…
Limitation. —
Figures and evidence
- changed
- 4
- not_retrieved
- 6
- observed
- 152
- pending_review
- 13
- source_changes_week
- 12
- staged
- 69
- upcoming_90
- 0
- upcoming_beyond
- 3
53 days of continuous observation across 150 observed sources (132 instruments tracked)
Method. Count of calendar days from the first recorded observation to today, inclusive. Every tracked instrument is fetched and hashed on every run, including runs where nothing changed. The observed-source count exceeds the tracked count because observation records for instruments since removed from scope are retained rather than deleted — the series is not broken by a scope decision.
Limitation. —
Figures and evidence
- days
- 53
- instruments_tracked
- 132
- observed_sources
- 150
16 of 107 recorded status value(s) awaiting a person; 25 row(s) sit off the verification axis in the register tier
Method. The verification axis is the tracked corpus minus the register tier, which promises no person-verification and is stated separately rather than folded into a total. A row is awaiting when its status is unknown, it is not frozen, and it records no status basis — three conditions, because a frozen row and a row with a recorded basis are settled rather than pending.
Limitation. The axis terms do not sum to the tracked total and are not meant to: the register tier is off the axis entirely. Awaiting counts rows no person has ruled on, which is not the same as rows a person declined — nothing here writes a negative verdict.
Figures and evidence
- awaiting
- 16
- on_axis
- 107
- person_verified
- 57
- register
- 25
- total
- 132
6898 instrument-days observed
Method. Count of distinct instrument-days observed: one per source per calendar day on which a record was written, whether or not the content changed. The store holds 16948 raw records across those days, because the loop can run more than once in a day.
Limitation. —
Figures and evidence
- instrument_days
- 6898
- raw_records
- 16948
committee
3 of 6 recorded committees have a scope quoted from their charter
Method. Count of body records carrying a non-empty scope. Scope must be the charter's own words; a charter that cannot be retrieved yields no scope rather than a paraphrase.
Limitation. —
Figures and evidence
Named figures: —
13 of 132 tracked instruments name a producing committee
Method. Count of distinct instrument ids appearing in the produces list of any body record, intersected with the tracked corpus. Counting only: a committee producing more tracked instruments is the count says how many instruments a committee produced and nothing further about the committee.
Limitation. Derived from 6 recorded body records. instruments_involved lists 12 of the 13 counted.
Figures and evidence
Named figures: —
119 tracked instruments name no producing committee
Method. Count of tracked instruments not appearing in any body produces list. States a gap in this archive's committee coverage; it is not evidence that no committee produced the document.
Limitation. Derived from 6 recorded body records. instruments_involved lists 12 of the 119 counted.
Figures and evidence
Named figures: —
corpus
Longest interval between a recorded status date and today: us-21-cfr-part-11, 10,621 days
Method. For each verified instrument, the number of days from the date the SOURCE states for its status to today. This measures the age of the regulatory fact, NOT how long this archive has observed it — the observation series is far younger and its length is reported separately. No weighting is applied.
Limitation. Computed over the 11 instruments carrying both a verified status and a recorded status date, of 132 tracked. An instrument without one is outside this population, not a zero within it.
Figures and evidence
Named figures: —
Median days in current status: 1054
Method. Median of (today − status_as_of) in days, over instruments with a verified status and a recorded status date.
Limitation. Computed over the 11 instruments carrying both a verified status and a recorded status date, of 132 tracked. An instrument without one is outside this population, not a zero within it.
Figures and evidence
Named figures: —
26 provision(s) tracked at article or section level; 0 with an applicability date still in the future
Method. Count of records in the provisions collection, and of those whose recorded applicability_date is later than today. Applicability is recorded per provision because instruments phase in across years and a single instrument-level date would be wrong for most of their articles.
Limitation. —
Figures and evidence
Named figures: —
Status across the tracked corpus: 74 unknown, 39 final, 13 in-force, 3 draft, 2 withdrawn, 1 superseded
Method. Count of instruments by the `status` field, as recorded after human verification against the primary source. Instruments never verified are counted as `unknown`.
Limitation. —
Figures and evidence
- status_draft
- 3
- status_final
- 39
- status_in-force
- 13
- status_superseded
- 1
- status_unknown
- 74
- status_withdrawn
- 2
35 of 37 declared tier-1 documents are tracked; 38 carry a verified status
Method. Count of documents named in sources/tier1-declared.yaml that appear in the tracked corpus at tier 1, and how many of those carry a status confirmed against a primary source. The declared list is a judgement about what belongs and is published so it can be disputed. Both figures are counts.
Limitation. —
Figures and evidence
Named figures: —
20 instrument(s) tracked at tier 1 but not named in the declared list
Method. Count of instruments carrying tier 1 whose id does not appear in sources/tier1-declared.yaml. Either the list is out of date or the tier assignment is. Both are gaps in this archive's own record-keeping, reported rather than reconciled silently.
Limitation. Measured against sources/tier1-declared.yaml, a judgement about what belongs rather than a measurement of the world.
Figures and evidence
Named figures: —
coverage
0 of 69 recorded candidate(s) are offered as a change
Method. Counted from state/triple-projection.json, which derives the three limbs for every file in state/revision-staging/. `offered` is every derivation agreeing; `waiting` is a limb that has not been measured yet; `withheld` is the remainder, each carrying its own reason.
Limitation. These count CANDIDATES, not changes. A candidate is a digest that moved; whether a document changed is what the three derivations decide, and for most of these they decide it did not.
Figures and evidence
- confirmed
- 0
- offered
- 0
- offered_witness
- 1
- staged
- 69
- waiting
- 9
- withheld
- 59
- withheld_by_terms
- 18
18 status pack(s) wait and 0 can be pressed
Method. Counted from state/packs/_index.json: the queued and withheld entries the confirm queue reads. `keypress` counts those its accept gate would admit; `blocked_on_collector` counts those whose pack states no corroborating passage or no capture, which a keystroke cannot supply.
Limitation. Depth is what is waiting, not what is wrong. A pack blocked on a collector is not an item a person has declined to read; it is an item no person can act on, and the two are different facts wearing the same place in a queue.
Figures and evidence
- blocked_on_collector
- 16
- depth
- 18
- keypress
- 0
crawlability
0 source(s) return HTTP 200 to a browser and a non-200 to an AI-agent user-agent
Method. Count of sources where the browser user-agent received HTTP 200 and the AI-agent user-agent did not, on the same URL, seconds apart. This archive sent an agent request to 0 of 132 probed sources: it no longer requests a host as an agent that host's robots.txt refuses, so an observed refusal can only come from a host that permits the agent and then declines it.
Limitation. Declared exclusions are counted separately and are not included here. An agent request was sent to 0 of 132 probed sources.
Figures and evidence
- agent_requests_sent
- 0
- observed_refusals
- 0
- probed_sources
- 132
9 source(s) name an AI agent in robots.txt and disallow it from the whole site
Method. Count of sources whose robots.txt carries a group naming an AI agent with `Disallow: /`. Read from the publisher's own file, identifying as this archive. No request is sent to a host that refuses the agent by name.
Limitation. Read only from a robots.txt this archive could retrieve. 49 of 132 probed sources publish no robots policy it can read, and a declaration cannot be observed on those.
Figures and evidence
Named figures: —
0 source(s) return HTTP 200 to an AI-agent user-agent and a non-200 to a browser
Method. The inverse of the preceding finding, recorded from the same two probes, and bounded the same way: an agent request was sent to 0 of 132 probed sources.
Limitation. —
Figures and evidence
- agent_requests_sent
- 0
- count
- 0
95 of the 132 sources probed for crawlability return the document to a standard HTTP client (132 instruments are tracked in total)
Method. Each source URL is requested twice: once with a desktop browser user-agent and once with an AI-agent user-agent. `yes` means at least one request returned HTTP 200 with 600 or more characters of extractable text. No judgment about the publisher is expressed or implied.
Limitation. An agent request was sent to 0 of 132 probed sources; where a host's robots.txt refuses the agent or cannot be read, only the browser request is made.
Figures and evidence
Named figures: —
49 of 132 probed sources publish no robots policy this archive can read
Method. Count of sources whose /robots.txt could not be retrieved, or returned a body that is not a robots file, when requested with this archive's own user-agent. A policy that cannot be confirmed is not a permission: no agent probe is sent to these hosts.
Limitation. —
Figures and evidence
Named figures: —
observation
0 source URL(s) retrievable at first observation and not at the last
Method. Count of instruments whose most recent observation was not a direct read and which had at least one earlier observation with fetch_method=direct. Three classes are counted and reported separately: refused (fetch_method=blocked), gone (HTTP 404 or 410) and other non-direct. Sources not retrievable from the first observation are not counted as rot.
Limitation. —
Figures and evidence
- rot_blocked
- 0
- rot_gone
- 0
- rot_other_non_direct
- 0
- total
- 0
0 revisions recorded: the collection holds none
Method. Count of records in the revisions collection by `change_class`. A revision is recorded when the SHA-256 of an instrument's normalised text differs from the previous observation. Detection currently holds 69 candidate(s), 69 unconfirmed; candidates are proposals and are not counted here.
Limitation. —
Figures and evidence
- total
- 0
0 silent revision(s): content changed with no declared version or date change
Method. Count of revisions where the normalised text hash changed and no version identifier or declared date differed between the two observations. Version identifiers are matched by pattern in the normalised text. Detection currently holds 69 candidate(s), 69 unconfirmed; candidates are proposals and are not counted here.
Limitation. —
Figures and evidence
- count
- 0
standards
14 standard(s) tracked by lifecycle stage; 8 with a confirmed stage
Method. Count of instruments with at least one standards-progression record. Stage codes are published openly by ISO, CEN/CENELEC and ICH even where the standard's text is paywalled, so progression is observable without reproducing any restricted text.
Limitation. —
Figures and evidence
Named figures: —
Last updated: