Revised, February 2000
What follows is a summary of the pattern of LogFlash 1 study: how the words
move from pile to pile.

Presented in Lesson order for each session (top to bottom of ladder).

First Mode: New Word Review
Recog 1  -> lojban recog -> Under Control
Dropback ->   "      "   -> Recog 1
New Word ->   "      "   -> Recog 1

(This gets people exposed to the entire list quickly, making it likely
that they will get better percentages the second time through, giving
them some ability to read, and knowing the scope of the gismu list.)
Experience with New Word Review Mode now suggests that it is too slow to be
effective for a serious learner.  The latter should move immediately to 
Gaining Control Mode, and use as large a lesson size as they can stand 
to work on daily.

Second Mode: Gaining Control

*Under Control -> lojban recall -> Under Control (English to Lojban)
Recall 1 -> lojban recall -> Under Control
Recog 3  ->   "    recall -> Recall 1
Recog 2  -> lojban recog -> Recog 3
Recog 1  ->   "      "   -> Recog 2
Dropback ->   "      "   -> Recog 1
New Word ->   "      "   -> Recog 1

*Under Control Lesson consists of words inserted into Under Control
EXACTLY 3 sessions earlier (words put into Under Control are kept
together by session number) + an additional number randomly selected
from other Under Control words to bring the total to approximately the
desired New Word Lesson size.

(User is prompt for an optional change to this mode at the start of any
session after all words are out of the New Word pile in New Word Review
mode.  This is the main new-learner mode.)

Third Mode: Maintenance

*Under Control -> lojban recall -> Under Control (English to Lojban)
Recall 1 -> lojban recall -> Under Control
Recog 2  ->   "    recall -> Recall 1
Recog 1  -> lojban recog -> Recog 2
Dropback ->   "      "   -> Recog 1

*As with Gaining Control, the Session Number when a word is put into
Under Control is saved.  For Under Control Lesson in this mode, however.
we search for the oldest session number, and then test all words with
that session number.  Successful words are given the current session
number, thus making them newest session number.  Aging then rotates them
back to oldest etc.  In effect what this is giving is an infinite stack
of Under Control piles, of which we only test one each session.

(User is prompted for an optional change to this mode from Gaining
Control or Brush-Up modes at the start of any session when all New Words
are out of the New Word Mode, and no pile below Under Control has more
than 'Number of New Words' words in it.)

When a person enters Maintenance mode from Gaining Control for the first
time, Recall will typically be at 30-50% on Under Control Words.  Each
succeeding pass through the complete deck will tend to cut the error
rate in half.  Regrouping is suggested after about two complete passes
through the deck in this mode, and about every 3 passes thereafter,
generally doubling your lesson size at each regrouping.  When you reach
97% recall average on lesson sizes over 100 words per lesson for an
entire deck, you can be said to have achieved Mastery.  At that point,
even long term lapse in use of Logflash will be unlikely to drop later
efforts below 80-90%.

Fourth Mode: Brush-Up

This mode is identical to Gaining Control, except that there is no Recog
3 pile:

*Under Control -> lojban recall -> Under Control (English to Lojban)
Recall 1 -> lojban recall -> Under Control
Recog 2  ->   "    recall -> Recall 1
Recog 1  ->   "      "   -> Recog 2
Dropback ->   "      "   -> Recog 1
New Word ->   "      "   -> Recog 1

* Under Control as defined for Gaining Control Mode.

(The purpose of Brush-up Mode is to support such a person who has
stopped use of LogFlash for at least 6 months, vcausing him to want to
work the words up the ladder again.  Since Recog simply is not lost to
any great degree even after a long period of time, the extra Recog
lesson can be skipped, allowing faster progress to Recall, which tends
to be worth more concentration.)

Default number of words in each New Word Lesson is 20.  This is changeable to
 a maximum of 250.

Maximum number of words in each lesson is unlimited.  This may be reduced.  If
 there are more words in a lesson than the maximum, the rest are saved for the
 next session.

There exists a Skip New Words Flag that may be set at any time, with the stated
 effect.  (These last two options are two methods of dealing with error pile
 buildup due to just low percentages (use max size) or a period of inactivity
 (use skip new words, and possibly also max lesson size).  The idea is to
 get people to stick with a consistent and somewhat ambitious number of
 New Words, rather than to keep fiddling with it, or to get depressed
 when progress slows or lessons get too long.)

In the default option, error lessons occur after each main lesson, and
 after all are correct N times (N default 6, but now changeable), then go
 to the next lesson.  At the end of a session, all error words are moved
 to the Dropback pile.

There is a 'Dropback flag' option.  If reset, then error words for any
 pile are moved to the next pile lower, rather than to the Dropback pile
 at the bottom of the ladder.  (Dropback errors are returned to the
 Dropback pile.)  This is an experimental option that people have asked
 for, but is not recommended.

There is an automatic review given at the start of any New Word lesson,
 and any error lesson, which may be exited at will.  The current program
 allows review to display 1 word at a time in a format like the test
 screen, or 4 words at a time in a slightly condensed version of the test
 screen, with paging up and down among the screens.  Optional review or
 printing of all words, or all words for a given lesson, is possible from
 the Main Menu.

Here is how LogFlash 2 (which has not been updated lately) teaches rafsi:

The same input file is used, but different internal tables are generated
from it.  This is why the gismu list has cmavo on it, when those cmavo
have rafsi assigned.

Recog Lessons:

In any lesson, there are three passes made through all the words.  All
words with a CVC rafsi are presented in a random order, the user
identifying the actual word from the rafsi.  Then repeat for CCV, and
then for CVV.  Thus there could be as many frames as 3*#words.  Since
null rafsi frames are skipped, in actuality the typical number of tested
frames is less than the number of words per lesson.

Errors are noted for each frame, and the error lesson tests you only on
the frames in which you erred.  A word which has NO errors in ANY of the
three frames is advanced to the next higher rung.  Thus a word with no
rafsi is not tested in Recog lessons at all.  This has been a mild
problem, making these words both too easy and too hard at the Recall
lesson level, since the user will not have seen them since New Word
pre-lesson Review 3 sessions before.

At the end of a session, all words with at least 1 error frame will be
in error piles, and are moved to the Failure pile (now to be renamed
Dropback, of course).

Recall Lessons:

The user is presented with the word, and three blank fields labelled
CVC, CCV, CVV.  The user must type in the correct rafsi and/or leave
blank if none for ALL 3 fields (TAB used between fields), then ENTER
to check all 3.  An error in any one, including either an omission of
a rafsi or an extraneous rafsi, marks the whole word an error.  The
3-entry field is tested 6 times.

At the end of a session, all error words will be in error piles, and are
moved to the Failure pile.  However, as that is a Recog pile, the test
reverts to 'triple frame' mode.

Since you are more likely to miss words with multiple rafsi, the Failure
pile tends to gain a high pile count, and higher than average frames per

Under Control is a Recall Lesson, with words selected as for LogFlash 1
of the appropriate mode.

There are 3 modes.  Gaining Control which tests Lojban<-> rafsi.  This
graduates to Brush-up which tests English keyword<->rafsi (and thus
should not be used until the user is reasonably proficient with LogFlash
1), and thence to Maintenance, also English keyword<-> rafsi.

Nora finished Gaining Control, reaching a reliable skill of 30%.  I
finished both Gaining Control and Brush-up, and 1 pass of Maintenance
reaching a skill of 70-80% (of which 10% more were gismu to keyword
errors).  This was 2+ years ago.

The lessons are more difficult, and hence higher error rates and slower
progress.  HOWEVER, the density of the rafsi packing is so high, and the
possible choices for each word so limited that by the time you reach the
end of the list, half of the word/rafsi relations are known merely by
elimination.  Thus there is a hump that is really noticeable in the
learning curve.

The new Logflash 2 (which is back-burnered indefinitely, due to lack of demand
will add two things, and change one.

1) There will be a New Word Review mode with two lessons as with LF1 and
LF3.  These Recog lessons will present all 3 rafsi on a single screen,
thus a single frame per word.

2) In Gaining Control and Brushup modes, the Recog 2 lesson will be of
the type just described, thus allowing a kind of review of the rafsi for
a word as a gestalt for the more difficult Recog3->Recall 1 lesson.

3) The lujvo making program will be integrated into LogFlash.  There
will be a Recog lujvo lesson added as part of the gestalt Recog 2 above,
and a Recall lujvo lesson added as part of the Under Control Lesson.
Thus there will be more chances to get words wrong, buyt more variety
and application.  The speed of learning is thus expected to increase

In each of the lujvo-making lessons a lujvo problem will be presented.

In Recog lujvo, a randomly generated lujvo will be presented for each
word, using each rafsi form.  The user must then give the Lojban, then
in Brush-Up/Maintenance, the English tanru that is reflected in the
lujvo.  Only words in the current lesson will be used, and some type of
check-off scheme will ensure that some frame will use each rafsi form
for a word, including 4-letter and 5-letter forms.  The total number of
frames will thus be 1-2 times the number of words.

In Recall lujvo, a tanru in Lojban or English depending on mode will be
presented, and the user must type in a valid lujvo.  Options will permit
the word to be graded correct if 1) any valid lujvo form 2) one of the
top N scoring lujvo forms 3) only the top scoring lujvo form, with the
default as 2) with N = 5.

All of this is subject to change.

Finally, here is how LogFlash 3 teaches cmavo:

The LogFlash 3 program is virtually identical to LogFlash 1, except that
field sizes are changed, necessitating some minor screen rearrangement.
The rafsi field is replaced by the selma'o field, and the Lojban text
field is larger.