Jesse.Markowitz/TheLadder
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

story-to-pack: add the corpora and the catalogue v2/v3

Browse Source 
Fourteen corpora split and classified: aesop, bierce, chekhov, holmes,
keefe, lawson, lorimer, maupassant, nobody, plaintales, poe, torchy,
wallingford and winesburg, each with its splitter and the hand-written
groups and pages; catalogue v2 and v3; and the shared splitters
gutenberg_chunks.py, se_split.py and se_build.py.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019rwKTmug58sEsJ72AuWsEi
This commit is contained in:
JesseMarkowitz
2026-10-07 06:05:51 -04:00
co-authored by Claude Opus 5.5
parent e5617b86ba
commit 3a4758172d
566 changed files with 303901 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/story-to-pack/probe/torchy/README.md
+363
View File
@@ -0,0 +1,363 @@
# Torchy reference set: the Torchy stories (Sewell Ford, 1911-1917)
A hand-made comparison baseline for the story-to-pack pipeline, a sibling of the Jacobs set in
`../jacobs/` and the Keefe set in `../keefe/`. Claude wrote every summary, predicament, group and
option by reading the scenes. **No local or hosted inference was used.** Compare the local-model
runs against it. It is not ground truth, just one careful reading.
The narrator is a slangy office boy at a big New York trust who is promoted, in the third volume, to
private secretary. **It is the closest source in the project to the game's Corporate Ladder pack**,
and the only corpus so far whose protagonist climbs.
The reading was split across **three readers**, one per volume (stories 0-18, 19-35, 36-54), each
writing every scene's record plus a notes file of the recurring situations they saw. The **grouping
was done in one pass** by a fourth reader over all 738 records and the three notes files, going back
to the scene text where a record left it unsure. So no reader's grouping was reconciled with
another's: the groups are one judgement, made across the whole corpus at once.
| File | What it is |
| --- | --- |
| `corpus/*.txt` | the three Project Gutenberg volumes: *Torchy* (#20626), *On With Torchy* (#17301), *Torchy, Private Sec.* (#20627) |
| `split_gutenberg.py` | → `stories.json` (55 stories, one per chapter) and `chunks.json` (738 scenes, same 320-word paragraph rule as `../chunk.py`) |
| `NOT_STORY.json` | the chunks that are publisher's advertising, kept out of every group |
| `claude/batches/sNN.json` | per story: `{chunk, story, summary, predicament}`, the same schema as `../jacobs/claude/batches/` |
| `claude/batches/NOTES-*.md` | the three readers' notes on recurring situations and protagonist standing, the input to grouping |
| `claude/groups_src.py` | the groups, options and per-story protagonist standing |
| `claude/build.py` | checks everything, then writes `claude/records.json` and `claude/groups.json`. It is identical to Jacobs' copy, since its paths are relative to the file. |
| `claude/page.py` | → `claude/REFERENCE.html`, the readable page (Jacobs' copy with the title, heading, subtitle and the unassigned note changed) |
Rebuild: `python3 split_gutenberg.py && cd claude && python3 build.py && python3 page.py`.
## What is in it
- **738 records**: 735 scenes of story and 3 of advertising. The predicaments run a median of 20
words, range 15 to 24. They follow the `redescribe.py` rules, so there are no names, firms, cities
or story-only objects. The `topic_words` leak check flags 464 of 735, and almost all of it is role
words: boss 221, sweetheart 80, suitor 44, rival 32, guest 31, visitor 30. These are kept on
purpose: the design's cast is roles, and this corpus is about exactly those roles. Place words are
rare (office 11, party 10, room 8, house 8, wedding 5).
- **Advertising.** `NOT_STORY.json` lists chunks 245, 736 and 737 as the publisher's advertising.
They are in **no group**. Chunks 244 and 735 are real story endings with advertising after them;
their records follow the story part, and both are used (244 in G03, 735 in G06).
- **39 groups**, each a situation that recurs in **three or more stories**. **All 39 pass the
catalogue's recurrence rule** (at least 5 scenes, at least 3 stories, no story over 50%), with
**stories 30 and 31 counted as one story** for both the story count and the 50% cap. They are one
story in two parts: the lobster boat and the hired boatman in 30 lead straight into the picnic, the
fog and the unmasking in 31. Six groups span 3 stories, 11 span 4, 9 span 5, 5 span 6, 2 span 7,
2 span 8 and 4 span 9. Each scene sits in at most one group: **559 are assigned and 179 are not**
(176 story scenes plus the 3 advertising chunks). All 55 stories feed at least one group.
- **Two different Marjories.** In story 23, Marjorie ("Peggy") is the boss's young ward. In stories
25, 26 and 35, Marjorie is the heir's married sister. Nothing in the grouping depends on the
name: story 23's ward is in G01-G03, G07 and G34 as the boss's young dependent, and the sister is in
G05-G07 and G21 as the heir's family. Both are `boss_family`, which is what they are.
- **Options**: 5 per group, 195 in all. 164 come from what someone in a scene did or refused to do
(`scene:N`), 16 are archetypes and 15 fill gaps. **10 scene options cite a scene outside their own
group**, and `groups.json` marks each one `in_group: false`. Most are the move a character makes a
scene or two after the predicament (51 after 50, 173 after 172), or the same move made in a
neighbouring group's story (317, teasing the raging boss, is also an order he carries out).
- **Moves**: spine paths with a direction only (`+`, `-`, `risk`, `set`), and no magnitudes.
Relationships move 191 times, `standing` 53, `savvy` 49, `cheek` 44, flags 32 and `money` 30.
Money is thin because nobody in these stories bargains over pay (see *Against Keefe*). Relationship
roles are generic. `boss` is whoever is over you at the top, the old head or his heir, and it is
by far the commonest (57). The others are `sweetheart`, `boss_family`, `supervisor` (the fussy
official between you and the top), `visitor`, `guardian`, `rival`, `old_friends`, `charge`,
`outcast`, `culprit`, `hunted`, `benefactor`, `snob` and a few more. Flags record what the player
did: `bluffed`, `covered`, `hushed_up`, `meddled`, `recommended`, `refused_order`, `let_slip`,
`owned_up`, `told_a_tale`, `in_disguise` and others.
- **Skills** proposed for this corpus: **`cheek`**, for saying the bold thing to people above you
(the frank answer, the bluff, the backchat), and **`savvy`**, for street sense: seeing through a
fraud, reading a room, solving by common sense what the experts dress up. These are the two things
that carry Torchy through every story. `savvy` is Keefe's label too, which is deliberate: Keefe
lacks it and Torchy has it, so one label can serve a pack built from both. `tact` was considered
as the second skill, because the volume 3 promotion is won by handling callers smoothly (487-488).
It was dropped because tact here always shows as a relationship move, and `cheek` carries more of
the decisions.
- **Stage** comes from protagonist standing, judged **per story, not per scene**, in `STANDING` in
`groups_src.py`. It is judged **by rank**, the job he holds in the story, so that all three volumes
use one measure. He is the office boy, **low**, in stories 0-36 (the promotion comes only in the
last scenes of 36, 491-493). He is private secretary, **middle**, in 37-49. He is **high** in 50-54,
where he sits in on directors' meetings as acting head of the subsidiary and advises the heir as
a near equal. The readers used different measures. Volume 1's reader used rank (all low) and noted
that his *pull* rises from story 12. Volume 2's reader used trust and reach, and rated 21, 24, 25,
29, 32, 33 and 34 middle, the stories where he acts in the boss's name. Volume 3's reader used rank
and rated 48 and 49 high for his newly found family. Here those seven volume 2 stories stay low,
since he is still the office boy, and 48-49 stay middle, since finding his family changes his
standing with the guardian, not his job. **The groups come out 32 low, 6 middle and 1 high.** That
skew is honest: 37 of the 55 stories are before the promotion, and the situations that recur most
are the office boy's. The rise is real, and it shows in *which* situations appear, not in the stage
counts (see *Before and after the promotion*).
## Groups against the existing catalogue
The 82-entry catalogue is in `../catalogue/v2/catalogue_v2_src.py`. The mapping below follows its
rule: a group joins an entry when the player faces the same decision from the same side, and it is
new when the decision differs. It was judged by reading, not by `classify.py`. **24 groups land on 19
existing entries, and 15 are new to the catalogue.** Two of the 15, G13 and G26, recur situation types
that Keefe's reference also found and listed as new. Those two are **cross-corpus confirmations**, so
13 are new to both. Neither Keefe nor Torchy is in `SOURCES` yet. Adding them is a separate change to
that file.
"Books" says where the group's scenes fall: *office boy* is before the promotion (chunks up to 490),
*secretary* is after it, and *both (thin)* means one side has only one story.
| Group | Scenes / stories | Books | Catalogue |
| --- | --- | --- | --- |
| G01 Deciding who gets past the door | 14 / 9 | both | **new** (`the_gate`) |
| G02 Judging a caller on sight | 11 / 5 | office boy | `mistook_them` (its own example) |
| G03 Called in, and expecting the sack | 15 / 8 | office boy | `called_to_account`, sharpened; Keefe G07 |
| G04 A petty supervisor who resents you | 15 / 9 | office boy | `power_over_you`, from below a patron; Keefe G26 turned round |
| G05 Pulled off your job for a family chore | 17 / 9 | both (thin) | **new** (`family_errand`, nearest `bow_for_your_living`) |
| G06 Sent to stand in for the boss | 22 / 5 | both | **new** (`stand_in`); partly Keefe G29 |
| G07 The boss's family troubles land on you | 24 / 7 | both | **new** (`boss_family_trouble`, nearest `peace_at_home` from outside) |
| G08 Someone in the boss's family is being taken in | 16 / 3 | both (thin) | `go_between`, widened |
| G09 Getting rid of an embarrassing guest | 18 / 5 | both | `unwelcome_in_home`, widened |
| G10 Talking the boss into backing someone overlooked | 12 / 6 | both (thin) | **new** (`champion`, nearest `go_between`) |
| G11 Someone everyone makes fun of | 21 / 5 | both | `despised`, from the onlooker's side; Keefe G25 |
| G12 Minding someone who will not be minded | 19 / 4 | office boy | **new** (`minder`, nearest `benefactor`) |
| G13 Calling on a friend from the bottom rung | 12 / 8 | both | **new** to the catalogue; **confirms Keefe G34** |
| G14 A bluff on the spot | 13 / 6 | both (thin) | `pretence`, widened |
| G15 Passing for grander than you are | 12 / 4 | both | `pretence` |
| G16 Out of place among the rich | 9 / 7 | both (thin) | `outside_looking_in`, widened; mirror of Keefe G19 |
| G17 Making yourself fit for the next rung | 6 / 4 | both (thin) | **new** (`looking_the_part`) |
| G18 Still on the bottom rung | 6 / 5 | office boy | `dead_end`, sharpened |
| G19 A sweetheart far above you | 11 / 5 | both (thin) | `station` |
| G20 Her guardian says you are not good enough | 34 / 9 | both | `family_against_your_path`, from the suitor's side |
| G21 Contriving a way to be near her | 28 / 6 | both | **new** (`forbidden_meeting`, nearest `pretence` and `rival`) |
| G22 A rival for her in the room | 21 / 5 | both (thin) | `rival` |
| G23 The boss notices you are in love | 5 / 4 | office boy | **new**; Keefe G09 inverted |
| G24 Meddling in the boss's love life | 23 / 4 | secretary | **new**; Keefe G09 inverted; partly `go_between` |
| G25 Stiffening a timid man's spine | 15 / 6 | both | **new** (`stiffen_spine`, nearest `courage_doubted` from the coach's side) |
| G26 Newly promoted, and not everyone accepts it | 16 / 5 | both (thin) | **new** to the catalogue; **confirms Keefe G06** |
| G27 Everyone trusts the expert, and you think he is wrong | 15 / 4 | both | `against_the_verdict`, sharpened; contradicts Keefe G08 |
| G28 A wrongdoer at your mercy | 13 / 3 | both (thin) | `mercy_or_law` |
| G29 Standing by someone the law or a mob is after | 10 / 3 | both (thin) | `protecting_hunted` |
| G30 An order from above you think is wrong | 8 / 3 | office boy | `pressed_to_wrong` (314-317 are nearer `orders_without_reasons`) |
| G31 Pumped for what you know | 7 / 4 | office boy | `their_secret` |
| G32 Covering for the boss's absence | 10 / 4 | both | `their_secret` (its own example) |
| G33 Someone wants to trade on your access | 11 / 3 | both (thin) | **new** (`trade_on_access`, nearest `after_your_money`) |
| G34 The boss in a temper | 9 / 5 | both (thin) | `power_over_you`, partly; contrasts Keefe G04 |
| G35 A benefactor who wants to run your life | 9 / 3 | office boy | `kept_dependent` |
| G36 Taken in by someone richer | 10 / 4 | both | `benefactor` |
| G37 Helping a couple against the family | 9 / 4 | both | `family_against_your_path`, from the helper's side |
| G38 A love let slip, and one more chance | 15 / 4 | both | **new** (nearest `changed` and `too_late`) |
| G39 A snob shown up | 18 / 6 | both | `mistook_them`, from the onlooker's side |
Five entries take two groups each: `mistook_them` (G02 at the door, G39 watching someone else),
`power_over_you` (G04, G34), `pretence` (G14 a quick bluff, G15 a sustained front),
`family_against_your_path` (G20 the suitor, G37 the ally) and `their_secret` (G31 an outsider fishing,
G32 covering an absence). In each pair the side or the decision differs enough to keep them apart.
**The readers' flagged situations, and where they went:**
- **Summoned to be fired, then kept for being frank** is G03, 15 scenes over 8 stories. It sharpens
`called_to_account`: the frank answer, not the flattering one, is what saves him (155-156, 20-21,
65-66).
- **Personal errands for the boss's family** is G05, which is new. The family's *troubles* are G07,
its *gullible members* G08 and its *unwanted guests* G09. The readers flagged one situation, and
the scenes split into four decisions: do the chore, fix the scrape, warn the dupe, remove the guest.
- **Deciding who gets through the gate** is G01, which is new. **Misjudging a humble visitor who
turns out to matter** is G02, `mistook_them` exactly. When someone *else* is the snob and the junior
watches the reversal, it is G39.
- **A petty supervisor who resents you** is G04, `power_over_you`, with a twist: the supervisor is
outranked by the junior's patrons, so the junior needles him and gets away with it.
- **Persuading the boss to promote an overlooked colleague** is G10, which is new. The Corporate
Ladder pack already has this event (`flags.recommended_trevor`), so a hand-written pack and a corpus
found it independently.
- **Refusing an order to falsify the books** happens in one story only (14, 183-186). It recurs once
widened to **an order from above you think is wrong** (G30), which adds the boss's rash telegram
that nobody dares stop (24) and the reporter told to praise a fraud or lose his job (34).
- **Being sent to represent the boss far above your rank** is G06, which is new: a host (21), an art
"commissioner" coached in stock phrases (25), a negotiator told to announce he is the office boy
(33), a stand-in at the guardian's charity show (46), and the man in charge of the boss's wedding (54).
- **The boss's family troubles landing on you** is G07, which is new.
- **The sweetheart's guardian who objects to your standing** is G20, the largest group (34 scenes, 9
stories). The volume 3 reader suggested keeping "judged by rank" apart from "judged by birth". It
does not recur on its own: birth is the objection only in stories 35 (476) and 49 (666-673). So it
stays one group, and its scenes show the objection moving from his job (485, 492) to his name (666).
- **After the promotion, being called Sir by the old hierarchy and undercut by a jealous senior** is
G26. It is Keefe's G06 found again (see below).
## Before and after the promotion
The promotion comes at chunk 491, near the end of story 36. Before it are 490 scenes of story; after
it are 245. Both halves are grouped at about the same rate (77% and 75%), so the difference between
them is in *which* situations recur, not in how much.
**Office boy only (nine groups, no scene after the promotion).** Called in and expecting the sack
(G03). The petty supervisor (G04). Judging a caller on sight (G02). Minding someone who will not be
minded (G12). Still on the bottom rung (G18). An order you think is wrong (G30). Pumped for what you
know (G31). The boss noticing you are in love (G23). A benefactor who runs someone's life (G35).
Five more are office-boy situations with a single story after the promotion: the family chore (G05,
one scene in 53), out of place among the rich (G16), a sweetheart far above you (G19), a rival in
the room (G22) and the boss in a temper (G34).
What they share is **being judged or used from above**. The office boy's risk is the sack. He is
summoned, lectured, sent on chores, questioned and given orders. His own power is the gate, and his
typical mistake is misjudging who is at it. Once he is promoted, every one of these stops. The
volume 3 reader noticed the same: the only threat to his job in that volume is his own threat to
quit (490, just before the promotion), and afterwards his fear is of looking like a dummy in the post
(498), not of the sack.
**Secretary only, or nearly (three groups).**
- **Meddling in the boss's love life (G24)**, 23 scenes over stories 50-53, none before. He plants a
reminder, sends roses in the boss's name, reads the boss's private letter files and pleads his case
to the woman. It is the volume's commonest new move.
- **Newly promoted, and not everyone accepts it (G26)**. The new boy calls him Sir (494). The stiff
senior clerk defers, and is won over when Torchy refuses the deference (495-496). A senior manager
who wanted the post for his cousin takes his only assignment (497). The old head, back from a
cure, orders him about as the boy again (524-525). Its one earlier instance is a *friend's*
promotion in story 7 (96-97), seen from outside, where the boss who resents the rise sets an
impossible task.
- **Someone wants to trade on your access (G33)**. A broker fishes for inside information (509-511)
and a long-lost uncle wants introductions to rich men (646-653). Its one earlier instance (383-385)
comes through the heir's name, not Torchy's own position.
**Through both, with the situation changing shape.**
- **The door (G01) turns round.** The office boy blocks callers and fits them into the boss's day.
The secretary receives them with tact (487-488, the scene that earns the promotion), entertains
them (606) and soothes them (633). By 721 the directors besiege *his* desk. The readers put it
plainly: he used to guard a door, and now he is the one callers try to use.
- **Favours reverse direction.** Before, the boss helps with the junior's courtship (G23: the heir
congratulates him, grants time off, wishes he could do more). After, the junior manages the boss's
(G24). 678 names the debt outright.
- **The bottom rung becomes a resource used from above (G13).** Before the promotion he calls on old
messenger-boy and newspaper friends as one of them (26, 88, 100). After it he trades on his old
rank's slang and pays the runner (531-532), and his newspaper friend runs the boss's wedding film
(731).
- **The guardian's objection moves (G20).** Before, she objects to his job: she cross-examines him
about his post and plans to shame him (485-486). The promotion answers that, and she finds him at a
desk (492). After, she objects to his lack of a name (666), and only his newly found family answers
that (673). So **20 of G20's 34 scenes come after the promotion**. Rank settled the office, and did
not settle the drawing room.
- **Standing in for the boss (G06)** grows more personal. Before, he hosts a supplier, buys a
painting and negotiates a railway charter. After, he runs the boss's charity errand and the boss's
wedding.
- **Stiffening a timid man's spine (G25)** runs evenly through both (8 scenes before, 7 after), and
its last instance is the boss himself (696-697).
**What this says about a pack's stages.** A pack's stages should differ by *which events can fire*,
not only by the numbers on the same events. This corpus gives a clean split. A first stage draws on
being judged from above: the summons, the supervisor, the chore, the order, the gate. A second stage
drops all of those. It adds the new rank not being accepted (G26), people trying to use your access
(G33) and responsibility for the boss's private life (G24). A third group of events should carry
**stage-specific descriptions** of one situation: the door, the stand-in, the old friends and the
objecting family all recur in both stages, and each looks different in each. That is the
description-variants engine change DESIGN.md already flags, with a concrete use.
## Against Keefe
Keefe (`../keefe/`) is the other literal career ladder, a ballplayer and then a soldier. His
reference listed 19 new situation types, 13 of them about work. Torchy recurs some of them and
contradicts others.
**Confirmed across corpora:**
- **Keefe G06, a step up goes to your head** (new there), is **Torchy G26**. Left in charge, he
"feels puffed up and important" (523). The move Keefe's options call wise, keeping on as before
with your old equals, is exactly what wins over the resentful senior clerk (496). Torchy adds the
side Keefe lacks: the *old hierarchy* resisting the step, from above (497, 525) and from below (494).
- **Keefe G34, leaning on the friend back home** (new there), is **Torchy G13**, widened. For Keefe
it is one loyal friend asked for loans. For Torchy it is a whole network from the rung he left:
messenger boys, a former messenger now trading on the street, a sports columnist and a reporter.
And he pays them back (101, 532).
- **Keefe G07, called to account** (`called_to_account`), is **Torchy G03**. It is the same
predicament, but the winning move is the opposite (see below).
- **Keefe G25, a cruel joke on someone who cannot take it**, is part of **Torchy G11**. In 608-616 a
hearty practical joker bullies a frail violinist and is shamed for it. With the old clerk the office
laughs at (177-180) and the plain typist (578-581), it makes one group about the person everyone
mocks.
- Partly: Keefe G29, *a new skill that looks easy*, appears in G06 when the office boy is coached in
stock art-critic phrases and blurts them at the wrong moment (328-331). Keefe G11, *the big occasion
and you are not ready*, appears in G27, where he freezes before the directors and then wins the room
(504-506). Neither recurs on its own in Torchy.
**Contradicted, or turned round:**
- **Pay talks (Keefe G01) and threatening to leave for a rival (G02) do not appear.** They were
Keefe's spine: G02 was his largest group. Torchy never names a figure. Pay rises arrive unasked:
the old head doubles his salary for resourcefulness (536), and the boss appoints a friend at five
thousand a year on Torchy's word (201). His one threat to quit
(490) is about his *title* before his sweetheart, not money, and it works. This bears out Keefe's
own finding that pay bargaining belongs to a labour market with rival employers, which a draftee and
an office boy at a trust both lack. `money` moves only 30 times here, against Keefe's 38 in a corpus
a quarter larger.
- **Blaming everyone else (Keefe G05) is reversed.** Keefe's commonest failing, spread over 14
stories, is putting his failures on others. Torchy's commonest turn is owning up. Called in to be
fired, he tells the plain truth and offers to go, and it is the frankness that keeps him (G03). The
predicament is the same, but the move each corpus rewards is opposite. A pack could carry the same
options and let the corpus set which pays.
- **Keefe G08, sure you know better than the boss, is contradicted by G27.** In Keefe the junior who
thinks he knows better is always deluded. In Torchy he is usually right: common sense beats the
amateur investigator (266-270), the senior manager's verdict on the inventor (501-507) and the
detective with the packet of sawdust (599-601). **Whether trusting your own judgement pays is a
property of the narrator, not of the situation**, and a generator should not hard-wire it.
- **Keefe G04, the boss who manages by mockery, is turned round.** Torchy's bosses rage (G34) but do
not mock. The mockery flows *upward*: the junior teases the raging boss that his temper is bluff
(317), and the boss adopts the junior's catchphrase (323). The contempt in Torchy comes from the
petty middle, the supervisor of G04, and the junior answers it in kind. That is Keefe's G26,
*needled by a colleague*, from the needler's side.
- **Keefe G09, the boss takes a hand in your private life, is inverted twice.** In Keefe the boss
meddles to protect his investment, and the player bows or refuses. In Torchy the boss helps the
junior's romance out of kindness (G23), and after the promotion the junior takes a hand in the
boss's (G24). The catalogue's `bow_for_your_living` has no entry for either direction of that
kindness.
The shared finding is the one Keefe's own test pointed to: **situations about standing inside a
hierarchy transfer between corpora, and situations about a labour market do not.** Being called to
account, the step up that goes to your head, the old network and the person everyone mocks all
recur. Pay bargaining and the threat to leave do not. Torchy adds one thing Keefe could not show. His
hierarchy is climbed, not endured, so its situations divide by rung.
## How this corpus differs
- **It rises.** Every earlier corpus had a protagonist at one level, or no fixed protagonist at all.
Keefe had no story ranking high. Here the same narrator is low, then middle, then high, and the
situations change with him (above).
- **It is the richest source yet for the office itself.** Sixteen groups are about work and the
people above and beside you (G01-G06, G10, G13, G14, G17, G18, G26, G27, G30, G32, G33). The
Corporate Ladder pack's own events turn up here: recommending a colleague (G10,
`recommended_trevor`), covering for someone (G32, `covered_for_trevor`), a colleague's promotion
(G26, `trevor_promoted`), and "looking the part" upkeep (G17, `lifestyle`).
- **Half of it is courtship and the boss's family.** The readers found that nearly every story has
two strands, an office errand and the courtship. The four courtship groups, G19-G22, hold 94 of the
559 assigned scenes, and the four about the boss's family, G05 and G07-G09, hold 75. After the promotion the work largely leaves the office: only stories 36, 37, 39, 43,
50 and 54 have much business in them.
- **It is comic, and the narrator is shrewd.** Keefe's comedy was dramatic irony: the narrator never
sees the joke. Torchy's slang hides a narrator who usually does see it. So the `scene:N` options are
often the wise move, and fewer groups need a filled "what a wiser person would do" option: 15 fills
here against Keefe's 41.
- **Story-long arcs are common.** Many situations run over five to ten consecutive scenes of one
story (the guardian in 49, the boss's courtship in 50-51, the snob and the grocer in 46). The 50%
cap is what keeps those from becoming one-story groups.
## Known defects
- **Truncated endings.** `../chunk.py` emits a trailing paragraph run only when it reaches 80 words,
so the short closing passage of six stories was dropped from `chunks.json` (it is still in
`stories.json`). **Story 2 loses the promotion that resolves it**: the heir agrees to give the clerk
"a show to get his name painted on a door", so chunk 40 ends on "what would you suggest?". G10's
first scene (40) therefore has its outcome missing. Story 12 loses the boss's promise of a pipe organ
to the rest home. Story 16 loses a closing jab at the secretary. Story 19 loses the tournament's
result: Vee chose him, and he has theatre seats. Story 24 loses the size of the boss's tip, which
was in hundred-dollar notes, so record 325's "caddie's tip" understates it. Story 44 loses his
keeping the child's secret from Vee. Stories 18 and 54 lose only advertising. This was not worked
around: re-chunking would shift the chunk ids the records are written against.
- **Advertising was not cut by the splitter** (`split_gutenberg.py` says so and why), so chunks 245,
736 and 737 exist and are listed in `NOT_STORY.json`. Their records say they belong in no group, and
none is in one.
- **One-story arcs stay unassigned even when they are strong situations.** Examples: the dying
inventor who refuses a fortune and flies out to sea (story 16, 206-214, the corpus's one tragedy);
the dismasting at sea (52, 702-708); the foundling's origins revealed (48, 650-662). None recurs in
three stories.
- **The cap holds out scenes that fit.** More of the secretary's theft in story 6 (81, 83) would fit
G28 but would push story 6 over half. So would the heir's bet in story 1 (22-23) in G05.
- **One recurrence was seen but not grouped, because the player's side flips.** Getting or losing a
post through connections appears in five stories: Torchy picked over the supervisor's head (15), a
clerk whose family pulled strings (30), a stockholder's son who cannot be fired (288), a waiter's
job given to the chef's cousin (356), and a manager who wanted the secretary's post for his cousin
(497). In some the player is the favoured one and in others the one passed over, so it is two
decisions, and neither recurs alone.
- No batch record was found to be wrong. The readers' notes cite example chunk ids that sometimes
differ from the final groups, because the notes were input, not a grouping.