← Seen or Said
API

Drive Seen or Said from your own code

The one metered task, beat-check, is available over HTTP. It takes 1 documentary (its name in your words and how much of it you saw) and up to 6 beats you remember, each with how the film showed it and the page's class, and returns what each beat actually shows, the leap taken on trust and a plain question to check it, plus the beat to check first.

The division of labour first: you pick how each beat was shown, a fixed rule gives the film its verdict and the page picks the beat to check first; the model explains them. After a run the page checks that the verdict and every class match, that the check-first line starts with the page's beat, that every number comes from the beats and that no ids leak into the prose.

Base URL and the envelope

Every endpoint lives under https://api.skillsafe.ai/v1/app-api, and every response uses one envelope:

{ "ok": true,  "data":  { ... } }
{ "ok": false, "error": { "code": "...", "message": "..." } }

The token is minted for this app (the guest call takes {"slug":"seen-or-said"}), so no slug header is needed afterwards. Send it as Authorization: Bearer …. The run body is the input object, sent bare.

StatusCodeMeaning
400validation_errorThe body is not a JSON object, or a required field (task, film, watched, beats, facts) is missing.
401unauthorizedNo token, or a stale one. Mint a guest token or sign in again.
402insufficient_creditsThe balance is under min_credits. Price with /estimate first.
403forbiddenA guest token tried to run: a reading is metered and needs a personal sign-in.
404not_foundUnknown job id.
429rate_limitedBack off and retry.
5xxserver_errorTransient. Retry with the same Idempotency-Key so a retry never double-bills.

The task field, the input and the classes

Every run sends task: beat-check (the one task is beat-check); a missing or different one is still answered as beat-check. Each beat's class is how it was shown: seen (the camera was there), found (old footage, a photo or a document), said (an interview), told (a narrator or on-screen text), staged (acted out or reconstructed), cut (2 shots placed side by side) or unclear. The verdict, first match wins: seen and found make at least half the beats on-record; staged and cut at least half rebuilt; said and told at least half on-their-word; otherwise mixed. Check first: the beat furthest from what the camera saw (unclear, cut, staged, told, said, found, seen), the earlier row on a tie.

FieldRequiredWhat it holds
taskyesAlways beat-check.
filmyesThe documentary in the person's words, up to 60 characters.
watchedyeswatched it all, watched part of it or saw a clip or trailer only.
beatsyes1 line per beat: beat 1 | the film says: ... | shown as: the camera was there as it happened | page class: seen.
factsyesThe page's lines: film verdict:, check first:, firmest:, classes:.

The reply

job.output.output is a JSON string:

{"task": "beat-check", "verdict": "on-record", "headline": "...", "beats": [{"beat": 1, "known": "seen", "shows": "...", "leap": "...", "ask": "...?"}, ...], "check_first": "...", "closing": "..."}
FieldWhat it holds
verdictThe page's verdict id, copied.
headlineWhat the verdict rests on, at most 16 words.
beats1 entry per beat, in order: beat, known (copied from the page), shows (at most 25 words), leap (at most 25 words) and ask (a question, at most 20 words).
check_firstThe beat to check first and why, ending with the firmest beat, at most 40 words.
closing1 light sentence, at most 20 words.

The page then checks the reply (BeatKit.check) and lists any disagreement under "The page's check".

Worked examples - every verdict

These are the 4 free samples on the page: the exact input the page sends and the reply that replays for free.

City fox family - seen, told, found, said; mostly on the record

Input:

{
 "task": "beat-check",
 "film": "a nature film about a city fox family",
 "watched": "watched it all",
 "beats": "beat 1 | the film says: the vixen moved her cubs to a new den under a shed | shown as: the camera was there as it happened | page class: seen\nbeat 2 | the film says: the cubs were born in early spring | shown as: a narrator or on-screen text said it | page class: told\nbeat 3 | the film says: a neighbour's old photos show foxes in the same garden years ago | shown as: old footage, a photo or a document | page class: found\nbeat 4 | the film says: city foxes are bolder than country foxes | shown as: someone said it in an interview | page class: said",
 "facts": "film verdict: on-record (seen or found 2 of 4 beats)\ncheck first: beat 2 (told, the furthest from what the camera saw)\nfirmest: beat 1 (seen)\nclasses: seen 1, found 1, said 1, told 1, staged 0, cut 0, unclear 0"
}

Reply:

{
 "task": "beat-check",
 "verdict": "on-record",
 "headline": "This film rests on the record for 2 of 4 beats you listed.",
 "beats": [
  {
   "beat": 1,
   "known": "seen",
   "shows": "You get footage of the vixen moving her cubs under the shed into the new den as it happens.",
   "leap": "The camera does not show what led to the move or what happened to the den after the shot ends.",
   "ask": "What happened just before or after the vixen moved her cubs to the den?"
  },
  {
   "beat": 2,
   "known": "told",
   "shows": "You are told in words on screen that the cubs were born in early spring.",
   "leap": "You still rely on the narrator's word for this detail, since no footage confirms the timing shown here.",
   "ask": "How could you check that the cubs were born in early spring?"
  },
  {
   "beat": 3,
   "known": "found",
   "shows": "You see old photos kept by a neighbour, showing foxes in the same garden years ago.",
   "leap": "The photos alone do not say exactly when they were taken or who took them.",
   "ask": "Could you find out when the neighbour took those photos of the garden?"
  },
  {
   "beat": 4,
   "known": "said",
   "shows": "You hear someone say, in an interview, that city foxes are bolder than country foxes.",
   "leap": "This rests on that person's own word, without footage comparing city and country foxes directly.",
   "ask": "What did the person comparing city and country foxes base the claim on?"
  }
 ],
 "check_first": "Beat 2 is worth checking first since it rests on what is told rather than on what a camera recorded, while beat 1 keeps the most direct link to the screen.",
 "closing": "Maybe take another look at beat 2 next time you watch this fox family film."
}

Painting found in an attic - staged, staged, said, cut, seen; mostly rebuilt

Input:

{
 "task": "beat-check",
 "film": "the lost painting film",
 "watched": "watched part of it",
 "beats": "beat 1 | the film says: the painting hung in the hallway for decades without anyone noticing | shown as: acted out or reconstructed | page class: staged\nbeat 2 | the film says: the grandmother bought it at a flea market | shown as: acted out or reconstructed | page class: staged\nbeat 3 | the film says: an expert says the brushwork matches the painter | shown as: someone said it in an interview | page class: said\nbeat 4 | the film says: after the auction the family looks upset, as if they lost money | shown as: suggested by putting 2 shots side by side | page class: cut\nbeat 5 | the film says: the restorer cleans the canvas in her studio | shown as: the camera was there as it happened | page class: seen",
 "facts": "film verdict: rebuilt (staged or cut 3 of 5 beats)\ncheck first: beat 4 (cut, the furthest from what the camera saw)\nfirmest: beat 5 (seen)\nclasses: seen 1, found 0, said 1, told 0, staged 2, cut 1, unclear 0"
}

Reply:

{
 "task": "beat-check",
 "verdict": "rebuilt",
 "headline": "This film rests on rebuilt scenes for 3 of 5 beats you listed.",
 "beats": [
  {
   "beat": 1,
   "known": "staged",
   "shows": "You watch a reenacted scene showing the painting hanging unnoticed in the hallway for decades.",
   "leap": "This scene was staged afterward to stand in for the moment, not filmed as it happened.",
   "ask": "What would show how long the painting hung in the hallway before anyone noticed?"
  },
  {
   "beat": 2,
   "known": "staged",
   "shows": "You see a reenacted scene of the grandmother buying it at a flea market.",
   "leap": "This recreates the purchase after the fact, rather than catching it happening at the market.",
   "ask": "What would confirm that the grandmother bought it at that flea market?"
  },
  {
   "beat": 3,
   "known": "said",
   "shows": "You hear an expert say in an interview that the brushwork matches the painter.",
   "leap": "This rests on the expert's own word, without other proof shown for the brushwork claim.",
   "ask": "What would help confirm that the brushwork matches the painter?"
  },
  {
   "beat": 4,
   "known": "cut",
   "shows": "You see 2 shots placed together - the auction, then the family looking upset.",
   "leap": "The edit links the family's look to the auction, without showing what caused it or any money lost.",
   "ask": "What happened between the auction and the family looking upset?"
  },
  {
   "beat": 5,
   "known": "seen",
   "shows": "You watch the restorer clean the canvas in her studio as it happens on camera.",
   "leap": "The camera does not show what happens to the canvas before or after this moment in the studio.",
   "ask": "What might the camera have missed just before or after cleaning the canvas?"
  }
 ],
 "check_first": "Beat 4 is worth checking first since it rests on an edited link rather than on what a camera recorded, while beat 5 keeps the most direct link to the screen.",
 "closing": "Maybe watch beat 4 again sometime to see how the auction scene is cut together."
}

Island post office - told, said, said, found; mostly on their word

Input:

{
 "task": "beat-check",
 "film": "island post office documentary",
 "watched": "watched it all",
 "beats": "beat 1 | the film says: it is the oldest post office still open in the country | shown as: a narrator or on-screen text said it | page class: told\nbeat 2 | the film says: letters once came by rowing boat | shown as: someone said it in an interview | page class: said\nbeat 3 | the film says: the post office keeps the island together | shown as: someone said it in an interview | page class: said\nbeat 4 | the film says: a 1950s photo shows a queue outside | shown as: old footage, a photo or a document | page class: found",
 "facts": "film verdict: on-their-word (said or told 3 of 4 beats)\ncheck first: beat 1 (told, the furthest from what the camera saw)\nfirmest: beat 4 (found)\nclasses: seen 0, found 1, said 2, told 1, staged 0, cut 0, unclear 0"
}

Reply:

{
 "task": "beat-check",
 "verdict": "on-their-word",
 "headline": "This film rests on their word for 3 of 4 beats you listed.",
 "beats": [
  {
   "beat": 1,
   "known": "told",
   "shows": "You are told on screen that it is the oldest post office still open in the country.",
   "leap": "You still rely on the narrator's word for this claim, without seeing proof it is the oldest.",
   "ask": "How could you check whether it is the oldest post office still open in the country?"
  },
  {
   "beat": 2,
   "known": "said",
   "shows": "You hear someone say in an interview that letters once came by rowing boat.",
   "leap": "This rests on that person's own word, without any record showing letters arriving by rowing boat.",
   "ask": "What would help confirm that letters once came by rowing boat?"
  },
  {
   "beat": 3,
   "known": "said",
   "shows": "You hear someone say in an interview that the post office keeps the island together.",
   "leap": "This rests on that person's own word about what the post office means for the island.",
   "ask": "What else could show that the post office keeps the island together?"
  },
  {
   "beat": 4,
   "known": "found",
   "shows": "You see an old photo from the 1950s showing a queue outside.",
   "leap": "The photo alone does not say why the queue was there or who took the picture.",
   "ask": "What would explain why the queue was outside in that 1950s photo?"
  }
 ],
 "check_first": "Beat 1 is worth checking first since it rests on what is told rather than on what a camera recorded, while beat 4 keeps the most direct link to the screen.",
 "closing": "Maybe take another look at beat 1 next time you watch the island documentary."
}

Lighthouse keeper's diary - told, staged, found, cut, unclear; a mix

Input:

{
 "task": "beat-check",
 "film": "the lighthouse diary film",
 "watched": "watched it all",
 "beats": "beat 1 | the film says: the keeper wrote in the diary every night for 30 years | shown as: a narrator or on-screen text said it | page class: told\nbeat 2 | the film says: a storm smashed the lamp room | shown as: acted out or reconstructed | page class: staged\nbeat 3 | the film says: the last diary page is shown up close | shown as: old footage, a photo or a document | page class: found\nbeat 4 | the film says: the keeper was lonely - the diary cuts to an empty sea | shown as: suggested by putting 2 shots side by side | page class: cut\nbeat 5 | the film says: the diary ended up in a museum | shown as: not sure how it was shown | page class: unclear",
 "facts": "film verdict: mixed (no group reaches half of 5 beats)\ncheck first: beat 5 (unclear, how it was shown is not known yet)\nfirmest: beat 3 (found)\nclasses: seen 0, found 1, said 0, told 1, staged 1, cut 1, unclear 1"
}

Reply:

{
 "task": "beat-check",
 "verdict": "mixed",
 "headline": "No kind of evidence carries this film on its own, so it reads as a mix.",
 "beats": [
  {
   "beat": 1,
   "known": "told",
   "shows": "You are told on screen that the keeper wrote in the diary every night for 30 years.",
   "leap": "You still rely on the narrator's word for this span, since no footage covers 30 years of writing.",
   "ask": "How could you check that the keeper wrote in the diary for 30 years?"
  },
  {
   "beat": 2,
   "known": "staged",
   "shows": "You watch a reenacted scene of a storm smashing the lamp room.",
   "leap": "This recreates the storm after the fact, rather than catching the lamp room damage as it happened.",
   "ask": "What would confirm that a storm smashed the lamp room?"
  },
  {
   "beat": 3,
   "known": "found",
   "shows": "You see the last diary page shown up close, as a record on screen.",
   "leap": "The page alone does not say when it was written or by whom.",
   "ask": "What would show when this last diary page was written?"
  },
  {
   "beat": 4,
   "known": "cut",
   "shows": "You see 2 shots placed together - the diary, then a shot of an empty sea.",
   "leap": "The edit links the empty sea to the keeper being lonely, without showing that feeling directly.",
   "ask": "What would show why the diary cuts to an empty sea here?"
  },
  {
   "beat": 5,
   "known": "unclear",
   "shows": "How the diary ending up in the museum was put on screen is not yet clear.",
   "leap": "Since how it was shown is not known yet, how much this covers cannot be judged.",
   "ask": "What would show how the diary ended up in the museum?"
  }
 ],
 "check_first": "Beat 5 is worth checking first since how it was shown is not known yet, while beat 3 keeps the most direct link to the screen.",
 "closing": "Maybe take another look at beat 5 next time you watch the lighthouse diary film."
}

Build the input yourself

Load the engine in node (global.window = global; require("./beatkit.js")), then:

const K = BeatKit;
const built = K.buildInput(K.form({ film: "a nature film about a city fox family", watched: "all",
  beats: [{ text: "the vixen moved her cubs to a new den", how: "seen" }, { text: "the cubs were born in early spring", how: "told" }] }));
built.input.facts;                      // the page's lines: verdict, check first, firmest, class counts
const wire = K.toWire(built.input);     // POST this to /estimate and /run
// then: const reply = K.normalize(K.parseReply(job.output.output));
// built.facts.verdict                  // the verdict the page's rule gives
// K.check(reply, built).notes          // the page's re-check of the reply
// K.md(built, reply, "my film")        // the check as Markdown

Step 1 — a tiny client helper

One function that sends JSON and unwraps the envelope. Paste a personal token from /tokens.html, or leave it empty and mint a guest token in step 2.

BASE="https://api.skillsafe.ai/v1/app-api"
TOKEN="paste-your-token-here"   # from https://seen-or-said.skillsafe.ai/tokens.html
api() { # api METHOD PATH [JSON-BODY]
  curl -s -X "$1" "$BASE$2" -H "Content-Type: application/json" \
    ${TOKEN:+-H "Authorization: Bearer $TOKEN"} ${3:+--data "$3"}
}

Step 2 — a token

A guest token is enough for /me and /estimate. A run is metered, so it needs a personal token: sign in on /tokens.html and copy it from there.

# a guest token (enough for /me and /estimate); a run needs your personal token from /tokens.html
TOKEN=$(curl -s -X POST "$BASE/guest" -H "Content-Type: application/json" -d '{"slug":"seen-or-said"}' | jq -r '.data.token')

Step 3 — GET /me

Who the token is (subject_type: user or guest) and the credit balance.

api GET /me | jq '.data | {subject_type, credits}' 

Step 4 — POST /estimate

Free. The body is the run input itself, the bare object (never wrapped in {"input": ...}). hold_credits is a reservation, not the price; min_credits is what the balance must cover; model_alias reads gpt-terra.

# input.json: the run input object - {"task": "beat-check", "film": "a nature film about a city fox family", "watched": "watched it all", "beats": "beat 1 | the film says: ... | shown as: ... | page class: seen", "facts": "film verdict: on-record ..."}
# - see the field table
api POST /estimate "$(cat input.json)" | jq '.data | {hold_credits, min_credits, model, model_alias, markup_bps, warnings}' 

Step 5 — POST /run, then poll the job

Send an Idempotency-Key built from the lane and the input: a retry with the same key never bills twice. The page uses seen-or-said:<task>:<inputHash>:a<attempt> (the one task is beat-check), and its one reformat retry (with retry_note added) is attempt 2, a new key and a second run. The job's output.output is the reply text; truncated: true means the balance cut it short.

KEY="seen-or-said:beat-check:$(shasum -a 256 input.json | cut -c1-16):a1"
JOB=$(curl -s -X POST "$BASE/run" -H "Content-Type: application/json" -H "Authorization: Bearer $TOKEN" \
  -H "Idempotency-Key: $KEY" --data @input.json | jq -r '.data.job_id')
while :; do
  J=$(api GET "/jobs/$JOB"); S=$(echo "$J" | jq -r '.data.status')
  case "$S" in succeeded|failed|cancelled) break;; esac; sleep 2
done
echo "$J" | jq -r '.data.output.output' > reply.json   # one JSON object, the reply contract
echo "$J" | jq '.data | {charged_credits, truncated}' 

Step 6 — POST /run-stream

The same run as Server-Sent Events, with the same input.json. Browsers receive progress ticks rather than text deltas; the final done event carries the reply.

# Server-Sent Events: tick/delta events while the job runs, then a final event with the job
curl -N -s -X POST "$BASE/run-stream" -H "Content-Type: application/json" -H "Authorization: Bearer $TOKEN" \
  -H "Idempotency-Key: seen-or-said:beat-check:$(shasum -a 256 input.json | cut -c1-16):a1" --data @input.json