REGRESSION — the rehearsal flies these, not you
Do not fly these to test them. Every row here cites only issues that are CLOSED and attested. They are kept because a tested thing is not deleted, it becomes the regression check — but re-flying settled work is how a card grows to a hundred and sixty rows and a pilot stops reading it. tools/ladder_rehearsal.py scores the structural half of these over real SRS, real Polly and real Whisper in about five minutes, with nobody in a jet. Run that instead. Anything here that a machine genuinely cannot judge — whether a seam is audible, whether it sounds like one person — is on EARS ONLY. If one of these starts failing, it has ROTTED: reopen the issue it cites and put the row back on a flight.
84 rows · in the order you meet them · say the ID and the issue number on the radio
D1[R#42] P1 first flight
Lead: "approach, <handle>, request creation of Apex flight"
"<handle>, you are now the lead of Apex flight. Each member of Apex check in to be joined" — and he uses your HANDLE, not a slot name
D2[R#42] , P1 first flight
Wingman, formed up inside a mile: "approach, <handle>, joining Apex"
"Roger <handle>, joined to Apex." From then on both of you are answered as Apex
D3[R#42] P1 first flight
A third aircraft well outside a mile tries to join
Refused with the number: "negative, radar shows you N miles from Apex — you must be within one mile to join"
D4[R#42] P1 first flight
Ask to join a flight that does not exist
"unable, Bolt flight doesn't exist" — an answer, not silence
D5[R#42] P1 first flight
As a flight, request the approach
Worked as one aeroplane: one clearance, one sequence number
D7[R#42] lead loss P2
Lead leaves the slot or crashes while the flight exists
The flight dissolves, survivors are asked their intentions, nobody inherits the name
D8[R#42] P2
Say something intra-flight — "Apex two, tighten it up"
The controller says nothing. It was not addressed to him
F1[R#41] P1 first flight
Fly the approach to a full stop
Handed to Tower over the runway, not at a range on final, and nothing tells you to climb once you are down
F2[R#41], [R#19] P1 first flight
Go around from short final
No handoff to Tower, and no reversal back towards the field while you are climbing out. A go-around at half a mile is closer than a landing at one
F3[R#41] P2
Touch and go
You are not handed to Tower for the few seconds you are on the runway — runway_touch is deliberately not acted on
F3b[R#164], [R#77] P1 first flight
Touch and go, and then say nothing — or get airborne again at any point after Tower has said goodbye
Nobody hands a flying aeroplane to Ground. The roll-out goodbye moves you to taxi_in while you are still rolling, so an aeroplane that then flies is a ground controller's aeroplane in the air; before 13 August nothing could retrieve it and you sat on 121.900, airborne. If you have already checked in with Ground, Batumi Tower takes you back — "contact Batumi Tower one one eight decimal six". A ghost flew both halves on 13 August and both fired. What is yours: on the touch-and-go the goodbye is followed about fourteen seconds later by "contact Batumi Approach one two four decimal four two five", with nothing from you in between. It is defensible — you are airborne, leaving, and Batumi has no Departure seat — but it is two frequency changes in a quarter of a minute at a hundred feet, and whether that is a controller working or a controller thrashing is an ear's question
F4[R#41] P2
Land, stop, leave the slot, take a new aeroplane and check in
The old callsign is gone from the board. Watch /diag — a leftover here is the ghost that held a real pilot in the stack for a whole approach
F4b[R#13] P1 first flight
Say something Whisper will mangle — a hurried call, a name half-swallowed, someone else's callsign — then check /diag
No aeroplane appeared that does not exist. One P-47 once entered the stack as "Hammer 1-1", "Hammer 1-3", "All 4" and "Maintained 2" — four aircraft, three imaginary, each with a place in the queue. With one ship that is untidy; with two a ghost at the head holds real pilots for an aeroplane that will never arrive. The identity work since means a radio is bound by RADAR and a mangled name should reach nothing, but the closure condition here has always been a live sortie and nobody has flown one
F5[R#88], [R#77] P1 first flight
Land, and say nothing at all for the rest of the sortie
On the roll-out Tower says "welcome, exit the runway and contact Batumi Ground one two one decimal niner", and you reach a stand without ever keying the mic. This was the last dead end in the ladder. Tower should not clear you to taxi to parking — that is Ground's; he names the man who owns it
F5b[R#77] P1 first flight
Check WHICH Ground, and when
It is the ARRIVAL field's — Batumi 121.900, never Kobuleti's 121.800 (this row said 118.600 until 13 August, which is Batumi Tower — the wrong number for the right controller, on the row about naming the right one). And it arrives during the roll-out, not after you are clear: there is no runway polygon in the aerodrome table, so "he has vacated" is not a fact this system can observe, and issuing it early is what a real tower does anyway
G10[R#1] P3
Airborne later — "what am I doing" or "where am I going next"
He answers from your plan without you repeating it, and does not read you ranges you do not need
What the aeroplane can do decides how much help you get: a moving map
wants the fix named and nothing else, a Mustang wants position reports and
vectors. It is keyed on the type radar reports, so nobody has to declare it — if
you are in the Mustang and he is terse, or in something modern and he will not
stop reading you ranges, that is the bug.
G3[R#1] P1 first flight
Ask again for BatumiTest, and read it back with one number wrong
He corrects that number only and asks for it again — not the whole clearance, and never a shrug
A controller who accepts a wrong read-back has recorded an agreement
that was never made. Get the squawk wrong, or the altitude; the failure to watch
for is him reading the entire clearance again, which on a busy frequency is how a
read-back correction becomes worse than the error.
G5[R#1] P1 first flight
"request clearance for the CAS over Tsutsnvati"
A question naming both — the plain one, and the one with the beacon letdown on return. He must not pick for you
Two plans on file are the same sortie flown two ways, differing only in
the recovery at the end. Asking is the correct answer and the hard one; the easy
failure sounds decisive and clears you on somebody else's plan. Listen for
whether he describes them by what they ARE — a name you were given yesterday is
not something to recognise under a running engine.
G6[R#1] P1 first flight
Answer it — "the beacon letdown one"
Kettle: eleven thousand, and the route out through Ingress to Tsutsnvati
And having asked, he has to use your answer. Check the level: eleven
thousand is Kettle, three thousand is Marlin, and getting Samovar instead is
invisible until you are on the wrong approach at the end of the sortie.
G7[R#1] P2
"request clearance to Vaziani"
"nothing on file" and a question. Not the nearest plan read out as though you had asked for it
The opposite failure to G5. A resolver that always picks its best match
never says "nothing on file", which looks perfect right up until the night it
routes you somewhere you never asked to go.
G8[R#1] amends P2
After a clearance, change your mind — "request a change, make it NellisTest"
A complete new clearance (four thousand), read back again. He must not say only what changed
An amendment REPLACES; it does not accumulate. Two live plans for one
aeroplane is the ambiguity the whole design removes, and a controller who reads
you only the changed part leaves you holding half of one clearance and half of
another. Your read-back also stops counting — you agreed to the old one.
G9[R#1] one plan, two flights P2
With Shooter: both of you ask for Marlin, one after the other
Both get it, both are cleared to three thousand, and neither clearance changes the other's
The reason plans became templates. Assignment COPIES, so one plan can be
flown by two aeroplanes at once and by the squadron again next week. Script-
checked already; what a script cannot tell you is whether it sounds right on the
radio when two men ask for the same sortie a minute apart.
H13[R#39] P2
Fly the approach fast — 400 kt or more — and watch the turn from base onto final
He may overshoot: known open, criterion 2. The sweep at 450 kt says 181 dithering events and a 5.1 nm turn radius against a 3 nm base leg — so expect left/right reversals, not a clean overshoot. Report how far through you go
The half of it that is not fixed, written down so you do not report it
as new. At 450 knots the turn from base onto final still overshoots, and the
number you give me — how far through the centreline you went — is what sizes the
fix, because the intercept has to scale with your groundspeed.
H14[R#39] P1 first flight
The ladder, step 1. F-16, whole approach at 300 kt or less
Clean. The sweep flies this exact grid at 300 kt: 1296/1296 arrive, zero dithering. Anything else here is new and matters more than H16
H15[R#39] P1 first flight
The ladder, step 2. Same approach in the P-51
Clean, and for a different reason than H14 — this is the speed the engine was built at, so a failure here is a regression, not a limit
H16[R#39] P1 first flight
The ladder, step 3. F-16, 450 kt or more inbound
Predicted to fail, and how: reversals. At 450 kt a 30° bank gives a 5.1 nm turn radius and the base leg is built around 3, so the engine orders a turn he cannot make and then orders the opposite. If it fails any OTHER way, that is new information and worth more than the expected one
H17[R#39] P1 first flight
Note every speed instruction you are given, and the number
You are asked to slow down at all — this has never been heard on the radio. In the F-16 never below 210 kt; the published profile wanted 174, which is the P-51's number
H18[R#50] P1 first flight
Cleared for the approach, ask for it again. Once you have the approach clearance, request it a second time — "Sockeye, request the approach"
"You are cleared", in some form. Never "number two" — behind whom? There is nobody else up. On Fred's 31 July sortie this told the only aeroplane in the sky he was second behind himself, four transmissions running at 44 nm, and he declared an emergency to break the loop. From the cockpit it is indistinguishable from being forgotten
H19[R#50] P1 first flight
Then change frequency and check in again while still cleared
Still cleared. The root cause of H18 was a check-in resetting a CLEARED aircraft to en-route on every frequency change — so the ladder, which changes frequency seven times, was the thing most likely to trigger it
H22[R#39] P1 first flight
Once established on final, listen for what happens to the speed restriction
"Resume normal speed", once, and no further speed assignments. On final the pilot owns his own speed — the controller does not know your fuel or stores
H23P1 Take off and climb out normally. Listen all the way to the hand-off to Center
No approach vectors on the departure. He must not turn you, must not descend you, and must not tell you you have gone around. On 10 August a departure off Kobuleti was flown on Batumi's approach geometry — six headings and a descent to two thousand while climbing out to five, thirty miles from either field
[R#86]
H25P1 Immediately after take-off, listen for two calls
You are not told you are established, cleared, or on final. On 11 August the approach geometry was asked about an aeroplane at 0.6 nm and 472 feet climbing off Kobuleti, said yes, and the engine cleared him for an approach he had not started — which wedged the phase for the rest of the flight
[R#93]
H27P1 En route, level at your cleared altitude, report it
He agrees with it. On 11 August a pilot level at five thousand — the altitude on his own clearance — was told "assigned altitude is five thousand five hundred, not five thousand" and made to climb. Two ideas of one number, and the strip is the one that counts
[R#98]
H28P1 With company: two other aircraft holding. Note every altitude the controller gives out
No two aeroplanes at one level. On 11 August, the first time three arrivals were ever sequenced at once, the aircraft cleared for the approach kept the bottom of the stack and the next arrival was assigned the same altitude — both over the beacon, both at five thousand. This is the accident the deterministic engine exists to prevent and it is the DEFAULT outcome of three aeroplanes arriving together. Reproducible in three lines with no radio; see #108, which is a design call and may be answered by fixing the board instead
[R#108]
H7[R#19] P2
Inbound, 2–3 nm off course, between 11 and 16 nm — get yourself there on purpose
He gives you an intercept heading. If he sends you OUTBOUND to reposition, that is E1 and it is still open
E1, mapped: inbound, 2 nm or more off, between 11 and 16 nm, and he
sends you outbound to reposition rather than giving you a heading. Under 1.5 nm
off, or beyond 18 nm, it behaves. You will have to put yourself there on purpose
— it is a narrow band and you did not meet it last night.
H8[R#20] P2
Arrive from behind the field, 8–12 nm on the departure side, and ask for the approach
He takes you out and brings you round. Circling near the field is E2
E2. Two starts in 1,296 still orbit on the sweep, both 12 nm behind the
field. The touchdown fix took it from three to two, so it may already be gone
from the air; this is the cheapest way to find out.
J1[R#40] P1 first flight
Fly with your usual SRS name and check in normally. Read the authority column
radar — the chain with no microphone in it: SRS name → the name radar prints → your track. If it reads plan or roster on every call, the roster fix did not take and I want to know before Andre, not after
J2[R#40] P1 first flight
Andre checks in cold — a radio this controller has never heard, nothing filed for him, no setup of any kind
He is identified and vectored on his first call. Ask engineering what authority said. radar means his SRS name matched his DCS name, which is the expected and best case; plan or elimination means the chain did not close and I want the reason
J3[R#40] P1 first flight
Garble or omit your callsign on one call — mumble it, or say only "request the approach"
He answers, and your identity does not move. The aeroplane you are in decides who you are; the words are only a claim
J4[R#40] P1 first flight
With a second aircraft up, say his callsign by mistake on one of your calls
Your report must not be filed against his aeroplane. That is a separation error, and it is the one nothing in the system used to report
J5[R#40] guards P2
Read back an instruction sloppily — "maintained two thousand", "left three sixty"
No new aeroplane appears. Ask engineering for the board if unsure. These exact phrases each invented an aircraft that took a holding level
J6[R#40] P2
Get radar to lose you — sit on the ramp, or go well out — then call in
"Not radar identified, say your position and altitude", and nobody is held behind you. Being unseen must cost you a place in the queue, not give you one
K1[R#43] the window P1 first flight
Say "approach, <you>, I have a question". Let him say standby. Then let the other pilot take a vector and read it back. Then wait
He comes back to you unprompted: "<you>, go ahead with your questions". He must not need reminding, and he must not have forgotten what it was about
K2[R#43] P1 first flight
Same as K1, but make the interleaved call one that needs a tool — a vector, a clearance or a "call you in five miles"
Identical result. Tool calls used to cost two messages each and quietly shortened his memory exactly when it was busiest
K3[R#43] P2
Ask him something that refers back four or five calls — "that heading you gave me earlier, say again"
He knows. If he asks which heading, the window is still too short and I want the number it was set to
K4[R#43] no stale situation P2
Somewhere mid-sortie, ask engineering to dump the last message sent to the model
No radar picture in any remembered turn — only your words and his replies. A stale scope in there is the bug, whatever else looks fine
K5[R#43] P3
Fly a long sortie, then ask engineering for the transcript
Postgres still has all of it. Trimming what he is SENT must not trim what can be replayed afterwards
L1[R#40] P1 first flight
With a flight formed, say something intra-flight on the ATC frequency — "Apex two, tighten it up"
He answers. Previously this was dropped and you heard nothing. What he says is the model's judgement; that he says something is the fix
L2[R#40] P1 first flight
Same, then check how he addresses you on the next call
Still your handle or your flight — never "Apex 1-2". A member number is not a name anybody is addressed by, and it must not become your label
L3[R#40] P1 first flight
Say "engineering, L3 check, how do you read" while on the approach frequency
Engineering answers and logs it, and you are not put in a mode — your next call goes to ATC with no "engineering, clear" needed. Say his name again next time you want him
L4[R#40] P1 first flight
Read back an instruction correctly
Silence is correct here. He does not acknowledge your acknowledgement. If he answers every read-back the frequency will fill up with two of you, and it gets worse with four
L5[R#40] P2
Read back an instruction wrongly — say the wrong altitude
He corrects you. This is the case the silence in L4 is buying
L6[R#40] P2
Transmit with no callsign at all, out of the blue
"Station calling, say your callsign" — a canned local answer that costs no model call. He is not ignoring you
P10[R#155] P1 first flight
Read the two clocks in the header, at any moment
bridge is the age of everything on the board; recorder is the age of the last thing said on the radio. On a quiet frequency the second goes minutes old while the first stays at 1s — that is correct and is not a warning about the bridge. One number used to stand for both, and accused a running bridge of being dead
P11[R#155] P1 first flight
On the knee, at kneeboard width, mid-sortie. Not on a monitor
Nothing scrolls sideways, and everything about you — what he thinks you asked for, what you are cleared for, where the engine has you, which rung of the ladder, and why nothing is happening — is on one card without hunting. The board used to be thirteen columns in a horizontal scroller and intent was off the right-hand edge
P9[R#155] P1 first flight
The moment you notice nothing is happening — holding, or waiting for a handoff that has not come — read your own card, bottom line
NOT DONE names who is keeping you and why: "Batumi Approach keeps him — approach, 20 nm, inbound". That sentence has been written down since 1 August and appeared nowhere. If the line is blank while nothing is happening, the monitor is not reaching the decision at all, which is a different fault and worth saying on the radio
Q0P1 Before saying anything else, ask Kobuleti Clearance for a clearance on a sortie you have not flown yet
He issues one. He must not say you are already cleared, and must not say your read-back was correct — you have not read anything back. A plan on file is not a clearance issued, and not a clearance acknowledged; "read-back correct" is the phrase that ends Delivery's business and hands you to Ground, so asserting it unprompted skips the rung it exists to close. Found by the ladder rehearsal on a deliberately clean board
[R#105]
Q1P1 Preset 1. "Kobuleti Clearance, Sockeye, request clearance." — your own callsign, the one radar prints, not the jet's slot name
He answers as Kobuleti Clearance — not Batumi anything — and gives you BatumiTest without asking which. NellisTest is the only other plan on file and it ends three thousand miles away, so nothing here is ambiguous and he must not ask. Note the plan carries NO origin: it is a route anybody may request, and Kobuleti becomes the origin at the moment the clearance is issued. Clearance to Batumi, an altitude, and a departure frequency of 123.300
[R#56]
Q12P3 At any point, ask a Kobuleti controller for something only Batumi can give — "request the ASR" on preset 2
He should send you to the right man rather than inventing an answer. A controller who works one field must not clear you into another's approach
[R#21]
Q13P1 On any Kobuleti frequency ask "say again the ground frequency", then "and tower?"
121.800 and 133.000. Until 7 August he was handed no frequency but Departure's and invented the rest — he answered "Ground is one three three decimal zero" (that is Tower) and "Tower is one one eight decimal zero" (that is Batumi Tower). Both in faultless phraseology. He now carries his own field's list
[R#58]
Q1bP1 Instead say "...request IFR clearance to Batumi."
He should ask which and Domino must be among the ones he offers. Every plan on the board ends at Batumi, so naming the destination narrows nothing — a controller who picks one confidently here has guessed. Before 7 August this resolved to Anvil: a real plan, someone else's sortie, because "Kobuleti" in your callsign line scored against Anvil's task
[R#57]
Q3bP1 On check-in with any controller, listen for the ATIS
He asks: "advise you have information Alpha", or tells you which letter is current if you named one. On 10 August the engine asked on three consecutive transmissions and not one reached the air — it carried no decision, so nothing checked it had been said
[R#90]
Q4bP1 Read the taxi clearance back — "taxi to runway zero seven, holding short of runway zero seven"
Nothing moves. He acknowledges and you stay with Ground. On 11 August that read-back was heard as a REPORT of holding short, so the phase moved and the ladder sent you to Tower before you had taxied an inch. Then, at three miles, say a debug note — it must reach the recorder and not the controller; one at 1,900 ft was classified as "I have landed" and suppressed the whole approach
[R#121]
R1P1 Tune Batumi ATIS on 127.1 and listen for the letter, then check in with Batumi Approach saying "with information ..." using the one you heard
"Information [that letter] is current. Say your request." — the same letter you just heard. This row said "use whatever /diag shows as current" and /diag does not show it: the letter is in the atis table and reaches a pilot only by being broadcast. Listening is the point anyway, and it exercises the transmitter in the same breath
[R#17]
R2P1 Check in claiming the wrong letter — "with Alpha"
"Information Bravo is current now, not Alpha." Then he asks your request. He must not refuse you anything
[R#17]
R3P2 Check in mentioning no letter
"Advise you have information Bravo." A prompt, not a telling-off
[R#17]
R4P1 Any of the above
Every one ends by asking what you want. He must never assume the ASR — that was a real complaint from the air, and there are two approaches published plus a visual
[R#17]
R5P3 Check in with Tower on short final
He says nothing about the ATIS. Quizzing a man at two miles is noise at the worst moment
[R#17]
R6P1 Ask Ground for taxi, note the runway; then ask Tower for take-off
The same runway, both times. It comes from the broadcast, not from each controller's own reading of the wind — that is the whole reason it lives in the database
[R#17]
S1P1 Any altitude, heading or frequency
"niner", "fife", "tree" — never nine, five, three. This is applied to every string on its way to the radio, including the agent's own prose
[R#17]
S2P1 Your callsign
"sock-eye", the fish. It was being read as one Japanese-looking word and came out like the rice wine
[R#17]
S3P1 "Batumi", "Kobuleti"
Three and four syllables run together, not spelled out and not paused between. The old spellings used CAPITALS for stress, which Polly reads as an initialism, and SPACES between syllables, which it reads as word breaks
[R#17]
S4P2 The nine controllers across one sortie
Nine distinct voices and nine distinct manners. Batumi Ground is gruff, Kobuleti Clearance is a pedant, Batumi Approach is the calm one
[R#21]
S5P3 Anything said twice — "roger", "readback correct"
Instant the second time. It is cached after the first render; a pause there means the cache is not hitting
[R#17]
S6P1 Any vector while being repositioned
A heading in fives — "two five zero", "two six five". Never "two six seven"
[R#19]
S7P1 The final approach course
NOT rounded — it is 125, the number on the plate. Rounding a published course is the one place five degrees is wrong
[R#19]
U0P1 Before anything else, on the ramp at Nellis, just ask for a clearance
He gives you one. He does not say you are past his boundary and send you to Los Angeles Center — which is what happened on the first Nevada run, to a stationary aeroplane. "On the ground" was measured from SEA level against a 200 ft threshold, and the Nellis ramp is at 1,849 ft, so every parked jet on the map read as airborne. Tonopah is worse at 5,550
[R#114]
U1P1 Preset 1, "Nellis Clearance, request clearance"
He answers as Nellis Clearance and issues Redflag — Nellis to Nellis via Tonopah, cruise 24,000. Not Silverstate, which is the one-way transit, and not anything Georgian
[R#110]
U2P1 Work the ladder to take-off: Ground 121.800, Tower 132.550
The same rungs as the Caucasus, at Nevada frequencies. Nellis Ground is on 121.800, which is also Kobuleti Ground's number — a coincidence, written down in config/theatres/nevada.toml precisely because it is the sort that makes a wrong answer look right
[R#110]
U3P1 Airborne, climbing out, say nothing
At about 5 nm Nellis Departure, 135.100, unprompted. Then at 25 nm Los Angeles Center, 133.400 — and this is the rung that did not exist. Nevada had no Center at all, so a departure worked cleanly through four controllers and then stayed with Departure for the whole flight, silently. If Center never comes, that is the finding
[R#110]
U4P1 Ask Center for a range or bearing to Tonopah
A real number. The fix catalogue used to be read off core.route, so a Nevada bridge published Kobuleti and Batumi and never Nellis — and a plan naming a fix the table does not hold is refused at delivery. Ask for Nellis too
[R#110]
U5bP1 Anywhere a handoff is due, listen to the whole transmission
It is a handoff or an instruction, never both. On the first Nevada stack run: "Bandit, contact Los Angeles Center one three three decimal four. Good day. Hold at present position, maintain one zero thousand..." — sent away and given an order by the man who sent him, and he cannot obey both. Report the exact words
[R#115]
U7P1 Being vectored, check every altitude against the chart minima
Never below them. Nellis and Tonopah carry their own 48-cell surveys, and the terrain here is a mile and a half up — a Caucasus-shaped assumption is not a small error over the Spring Mountains
[R#110]
U8P2 After landing, the taxi in
Ground has you and nothing hands you back to Tower. Same rung as F6 on the Caucasus, on a map where none of it has ever been flown
[R#100]