S7 Reliability of facts and verification against the originals

At each fact the app shows how reliable it is, and the card in the tree has one mark for the whole person. The research can verify a fact against the original record: in the background, for one fact, for a person or for a large part of the tree. Certainty, though, comes only when a person compares the fact with the scan of the record and confirms it (Confirm against the scan, S7.8). Verification by the agent gives at most Proven; a person's confirmation is stronger and the agent never changes it. The chapter describes the Strom app 3.26.0 and the research 1.30.0. What works only with the research's bridge and what works without it is summed up in S7.9.

S7.1 The mark at a fact in the person card

In the person card each fact has a mark: who confirmed or proved it, how sure the research is, or that it came from another program's family tree and nobody has verified it. The steps differ by a tick, dots and a frame, never by colour alone.

• The tip at the mark (on a click): the fact's sources; with two readers two readers agree; for a fact that stayed probable or possible after verification, the line “Stays {degree} · verified by the research on {date}” and the reason why it could not be proven; for a confirmed one “✓ Confirmed against the scan on {date}”; for a fact from the family tree that the family tree cites, cited in the family tree, not verified in Strom. The tip also holds the actions Verify (S7.3), Confirm against the scan and Undo (S7.8), where they apply. From app 3.27.0 each source in the bubble has the reliability of the citation at this fact (Original, Copy / index, Uncertain), not the source's default reliability (A2.6.1).
• Enforced by the code:
  • two dots go only to a fact for which two readers of the record agreed when the research verified it (R8.8.1); a fact proven otherwise has one;
  • only a person gives ✓: by confirming against the scan, or by writing, changing or putting back the value themselves, or by accepting a proposal of the research. A fact the research took in from a family tree from the Strom app and nobody has changed since has the mark of its degree of certainty, without a degree From family tree (○○); never ✓ (S6.3);
  • the marks of a research tree are the same with the bridge and without it, also right after the page loads: the app takes them from the stored tree, not from the running bridge;
  • a person's confirmation holds only while the fact has the confirmed value (S7.8).
• Limits: the app does not yet show the mark No source at a fact: a fact without a source has no mark. A person without a source is shown by the mark on the card (S7.2).
• Controls: the person card → the mark at the fact; the help What the marks mean. Settings → Person card → Reliability at facts in the person card (on by default).
• Since: app 3.26.0, research 1.30.0

S7.2 One mark on the card in the tree

A person's card in the tree has one mark in the bottom right corner: how well the person is backed and who stands behind the facts. It takes the weakest of the card's three facts: name, birth and death.

• Enforced by the code:
  • only facts that have a degree count; when neither the name, nor the birth, nor the death has one, the person's other facts and those of their families decide (residence, occupation, a marriage): one confirmed or proven fact, or a citation of quality 2 or 3, is enough and the mark is ●○ (the person is known from a record, for example as the father in a child's baptism);
  • from app 3.27.0 the quality of a citation is its own reliability, otherwise the source's default reliability (A2.6.1): the same source can be a citation of quality 3 at the occupation and of quality 2 at the birth. A record of a change by the research that is not cited at the fact counts by its source;
  • a research tree whose stored data carry the research's origin of facts has the mark by degrees right after the page loads, the same with the bridge and without it; this holds for trees stored earlier too;
  • a tree that does not know where its facts come from (without research and without an import from another program) shows the mark by the person's citations: ●● documented, ●○ partly documented, the dotted frame no source. A tree with neither sources nor origin has no mark;
  • a placeholder person (“?”) has no mark.
• The tip at the mark: “Weakest fact on the card: …” and in words what the mark means.
• Controls: Settings → Person card → Show on card → Sources and reliability (on by default).
• Since: app 3.26.0

S7.3 Verify at a fact

At a birth, baptism, marriage, death and burial that is neither proven nor confirmed, the tip at the mark offers Verify. The research then verifies the fact against the original record: in the background, apart from the work on the lines.

• Preview before starting (Verify {fact}):
  • What is verified;
  • How: “AI reads the record {record}” (the record is known, low · one known record), or “AI looks for the record by the family tree's citation ({citation}) and reads it”, or “AI looks for the record in the registers and reads it” (higher · the record is still to be found);
  • Estimate: about how many agent sessions it takes;
  • “Runs in the research's background. The result shows at the fact.” When searching also “If the record is not found, the fact stays from the family tree and the tip says why.”
• After starting (Start verifying) the fact has a mark with a clock “Waiting to be verified · {n}th in the research's queue” and Verify is not offered at it again.
• Verify first: a fact the research queued for verification by itself (S7.6) and that is not next in line has Verify first in its tip. The same preview with the estimate as at Verify follows; after Start verifying the research moves the task with this fact to the head of its queue and the tip shows the place the research answers (“1st in the research's queue”).
  • From research 1.31.0: a task the person moved up this way is kept as asked for by a person (asked) in the file for the app, in the bridge's state (/status) and in the counts of strom verify status, until a session on it ends. The app then does not offer Verify first at the fact again after a reload; before, it offered it again. This holds for the whole person too (S7.4).
• Enforced by the code:
  • a verification asked for by a person goes first in the research's queue (R2);
  • Verify is offered only for a tree linked to the research, when the bridge runs and the research can verify (S7.9);
  • name, sex, divorce and other events are not verified;
  • Verify first is not offered at a fact whose verification a person asked for already (it is first in the queue), at a fact waiting for scans or for an archive's permission, at the one next in line, and when the research does not give the fact's place. A task the agent is working on right now the research does not move.
• Limits: a tree without research has, instead of the action, the line “With the ancestor research an AI agent verifies the fact in the registers itself.” in the tip. When the research does not respond: “Verifying could not start; the research does not respond.”; when there is nothing to verify: “The research has nothing to verify here now.” When the research refuses the verification, the tip says why, for example that the research is an archive, that it no longer has the fact, or that it is busy right now.
• Controls: app: the person card → the fact's mark → Verify (or Verify first) → Start verifying. Terminal: strom verify facts P…|F…|E… [--fact …] [--dry-run].
• Since: research 1.30.0, app 3.26.0

S7.4 Verify facts at a person

All unverified facts of one person can be sent to be verified at once: with the button Verify facts (N) in the person card and in the tip at the mark on the card, or with the item Verify facts · N in the person's menu.

• Where:
  • the person card: the button Verify facts (N) above the section Events. When the person has nothing to offer for verification but their facts are waiting to be verified, it shows “N facts waiting to be verified” instead of the button; otherwise nothing is there;
  • the tip at the mark on the card in the tree (S7.2): Verify facts (N), the same as in the person card;
  • the person's menu: Verify facts · N, while verifying Verifying · N.
• The window Verify the person's facts: “{name} · N unverified facts”, a list of facts to tick, each low or higher, “Estimate: {n} low, {n} higher” (“Low = read a known record. Higher = the record is still to be found.”) and the button Verify (N) by the ticked ones. After starting “N facts queued for verification.” and the menu item shows Verifying · N.
• Verify first for the whole person: when the person has nothing to offer for verification and some of their facts waits further down the queue as the research queued it by itself, the person's menu has the item Verify first under Verifying · N, and the tip at the mark on the card has the button Verify first under “N facts waiting to be verified”. It opens the person's preview with this title; after starting, the research moves the tasks with the person's waiting facts to the head of its queue (S7.3). The person card keeps only the line “N facts waiting to be verified”.
• Enforced by the code: N is exactly what the research takes: the facts of the person and their families that the research knows (residences too, and facts already proven by another reader), except those confirmed by a person, those waiting to be verified, the name and the sex. The count in the menu, in the window and in the message after starting is the same.
• Limits: the button and the item show only with the research's bridge running (S7.9) and only when at least one fact of the person offers Verify.
• Controls: app: the person card → Verify facts (N) above Events; the tip at the mark on the card → Verify facts (N) or Verify first; the person's menu → Verify facts · N, or Verify first under Verifying · N; then in the window Verify (N). Terminal: strom verify facts P….
• Since: research 1.30.0, app 3.26.0

S7.5 Verify facts in the tree

Verification against the originals can be started for a whole line, a family, or the whole tree. The window Verify facts in the tree shows the scope, the order and the cost estimate first and verifies only unverified facts (probable, possible, lead, from family tree).

• Where: Verify facts in the tree… in the section Verifying in What the research knows (S3), in the research menu and on the card What next after an import as Check facts against sources (S6.4).
• Choices:
  • Scope: Line: {name}, The {surname} family, Whole tree;
  • Order: Nearest ancestors first, Weakest evidence first, Longest unverified first;
  • Try on 5 people first: “After 5 people it stops and shows the result.” The button is then Try on 5, otherwise Verify (N).
• Enforced by the code:
  • it takes births, baptisms, marriages, deaths and burials with a citation of the original that no person has confirmed; proven ones too (another reader reads them), except those the current model has already verified; and leads from another program's family tree and from a family tree from the Strom app that nobody has changed (the record is searched for);
  • never living people (born less than 100 years ago with no death recorded);
  • one task per person (a marriage per family), with low priority, after the work on the lines;
  • only a person may start it: in the app, or at their own terminal. The agent may only run --dry-run (how many and at what cost).
• Stop: the section Verifying shows “Waiting to be verified: N facts · verified one by one, nearest ancestors first” and Stop (“Verification stopped.”). Stopping drops the tasks nobody has started yet; started ones run to the end. Dropped facts are not queued again by themselves, only with a new citation of the original or when asked for again.
• Controls: app: see above. Terminal: strom verify bulk --scope line|lineage|tree [--of G…|P…] [--order ancestors|weakest|oldest] [--try N] [--budget …] [--dry-run], strom verify status, strom verify stop [--bulk].
• Since: research 1.30.0, app 3.26.0

S7.6 Verifying by itself in the research's queue

A fact that cites the original record and stayed probable or possible is verified against it by itself: the research queues it without anybody asking.

• Enforced by the code:
  • the queue is filled after the planning of each session and after strom sync;
  • it takes births, baptisms, marriages, deaths and burials of people linked to the tree, off side branches, not living; never in the Strom archive;
  • one task per person (a marriage per family), after the work on the lines; after every 5 sessions one verification task moves forward (the nearest ancestor first);
  • a fact verified, kept with a reason, confirmed or lowered is not queued again, only with a new citation of the original or when a person asks;
  • a task that waits for scans or for an archive's permission waits like any other (R6.2.1, R6.3.1).
• Where the count shows: strom check (“note: N facts with the original cited waiting to be verified against it → strom verify status”), the strom doctor row verification against the originals (“N facts with the original cited waiting to be verified, bit by bit”), strom stats, strom verify status (Verification against the originals: waiting, of them asked for, waiting for the scans to be downloaded, waiting for an archive's permission; verified and kept with the reason; confirmed against the scan; and why a fact was left out of the selection) and in the app the section Verifying (S7.5).
• Since: research 1.30.0

S7.6.1 A one-time fix of older researches

A research started before version 1.30.0 has facts citing the original that nobody has verified against the original. The fix Facts to verify against their originals queues them for verification once.

• Enforced by the code:
  • the fix runs by itself once, at the first start of the new version after the research's backup; then similar facts are queued as they come (S7.6);
  • only tasks: nothing on the facts changes. “Nothing is raised without a look at the original; no value is overwritten.”;
  • strom fixes undo verify-originals drops the tasks nobody has started; started or finished ones stay;
  • in the Strom archive it is only noted that it ran.
• In the app: after the research is updated, once the message “N facts with an original are waiting to be verified, one by one.”
• Controls: terminal: strom fixes (the list of fixes and what they did), strom fixes undo verify-originals.
• Since: research 1.30.0, app 3.26.0

S7.7 The result of verifying

The session that verifies a fact opens the scan of the record and reads it. The result is one of two: the fact is proven by the scan, or it stays as it was, with a reason.

• Enforced by the code:
  • proving is possible only with a scan read; without a look at the original nothing is proven;
  • a kept fact has a reason that the app shows in the tip (“Stays {degree} · verified by the research on {date}”);
  • verification overwrites no value. When the record says something else, a proposal or a conflict arises for a person to decide (S6.3, R5); a fact entered or confirmed by a person the agent never changes;
  • the result is written only by the session doing the verification task (strom verify done).
• Why: the agent proves, a person confirms. Even two readers agreeing is a reading by AI; certainty comes only from a person with the scan (S7.8).
• Since: research 1.30.0, app 3.26.0

S7.8 Confirm against the scan: confirmation by a person

The strongest mark is given by a person: they look at the scan of the record, compare it with the fact and in the tip at the mark click Confirm against the scan. The fact is then Confirmed (✓), and in place of the action the tip shows the state “✓ Confirmed against the scan on {date}” and Undo.

• Enforced by the code:
  • Confirm against the scan is offered only for a fact with a source: a citation, or the record the research read the value from. A fact without a source does not have it: there is nothing to confirm against;
  • what is confirmed is the value the person sees. When the research has changed it meanwhile, nothing is confirmed: the tip closes, the app says “The fact has changed meanwhile; nothing was confirmed. Only the research's current value can be confirmed.” and loads the research's current version (a tree with changes the research does not have yet asks first, Load the research version?). A value changed in the app and not sent yet is not sent to be confirmed; the tip says the same;
  • Undo takes the confirmation back (“Confirmation undone”); the fact then has the degree it had before;
  • a confirmation holds only for the confirmed value: a change of the fact ends it, and when the value comes back it holds again;
  • a confirmed fact the agent never changes, it only proposes (S6.3);
  • a tree without research: the app keeps the confirmation itself, at a fact that cites a source, when the tree can be edited. The confirmation and its undoing go with the tree into the research, also through a GEDCOM file;
  • a confirmation the app keeps is a change of the tree: it has a step back (“confirming a fact of {name} against the scan”, “undoing the confirmation of {name}”) and an entry in the change history (“Fact of {name} confirmed against the scan”, “Confirmation against the scan undone: {name}”);
  • a tree linked to the research: the research writes the confirmation through the bridge, only when the bridge runs and the research can do it; the name and the sex are not confirmed here;
  • in the terminal only a person at their own terminal confirms: strom verify confirm E… [--undo]. To the agent strom refuses the confirmation.
• Limits: a tree linked to the research whose bridge does not run offers neither Confirm against the scan nor Undo; confirming works only with the bridge running, or in the terminal. When the research refuses the confirmation, the tip says why, for example that the research is an archive, or that another session holds it; “Not changed: the research does not respond.” only when it does not respond.
• Why: certainty about a fact comes from a person seeing the record themselves. Verification by the agent is a reading by AI; ✓ says that a person saw the record. A confirmed transcript (A3.1.1, R3.4.1) confirms the text of the record, Confirm against the scan a fact of the person.
• Controls: app: the person card → the fact's mark → Confirm against the scan / Undo. Terminal: strom verify confirm E…, strom verify confirm E… --undo.
• Since: research 1.30.0, app 3.26.0

S7.9 What works with the bridge and what without it

• Enforced by the code: the research verifies even without the app: in the terminal strom verify facts, strom verify bulk, strom verify status, strom verify stop and (only a person) strom verify confirm work. The research's bridge: S1.1.
• Since: research 1.30.0, app 3.26.0