One catalog row, stated exactly
Knifemaking / Bladesmithing: "Forging, grinding, heat treating and finishing blades in a metal workshop." The row is rated Hard, lists a head-mounted phone or a multiview rig and is Available on request.
Now recruiting capture teams worldwide. Get paid to record first-person video of real work.
Apply to captureHard task
Requirements
Knifemaking / Bladesmithing: "Forging, grinding, heat treating and finishing blades in a metal workshop." The row is rated Hard, lists a head-mounted phone or a multiview rig and is Available on request.
Hand pose, contact events and action segments are recorded on one clock, with a visibility flag on each joint for fingers that stock, a handle or a tool hides.
A calibrated stereo pair and an MCAP container are agreed in the skill spec. The row does not state either, and this page does not claim them for the row.
No steel, temperature, hardness, tool, machine or outcome is stated for this row, and this page makes no safety or compliance claim and gives no instructions.
Definition
Stereo knifemaking and bladesmithing data is first-person video of a person doing the work named in one catalog row: forging, grinding, heat treating and finishing blades in a metal workshop. It is recorded from the contributor point of view with two calibrated cameras, and with synchronized sensor streams alongside, so the frames show what the hands are doing with the stock, the tool and the blade rather than a distant view of a shop. Searchers also phrase it as stereo knifemaking video, blade forging and grinding robot data, a bladesmithing dataset or custom metal workshop data collection at scale. The data is the same, collected as a custom collection scoped to your task.
On this site the data is human capture: a paid, consented contributor performs the task from a written skill spec. It is not robot or teleoperation data, and the page makes no claim that any robot has been trained on it or that any robot makes a blade. Stereo describes the capture and MCAP describes the container. They are separate choices, and neither is stated on the row itself.
Catalog row
One row sits in the Metal Fabrication & Welding Workshop environment, in the Manufacturing & Workshops sector. "Knifemaking / Bladesmithing" reads: "Forging, grinding, heat treating and finishing blades in a metal workshop." It is rated Hard, its capture is Phone (head-mounted) / multiview rig and its status is Available on request. The environment shared example line reads "Metal shop with welding bays, grinders, cutting and fitting benches". It is one line for every row in that environment, so it is quoted here rather than assigned to this task.
Available on request means the contributors and site access for the row are not yet set up for immediate capture, so the timeline is quoted against your spec instead of being fixed in advance. It does not mean a stock of blade footage exists, and this page claims no volume, hours or delivered clips for the row. The row names forging, grinding, heat treating and finishing, blades and a metal workshop. It names no steel, no temperature, no tool or machine model and no outcome, so if your program needs any of them, they are written into the skill spec rather than assumed.
The environment lists eleven rows in total. Four are rated Hard, seven are rated Medium and none is rated Easy. The other ten are listed as Ready to deploy. The first table below reproduces the knifemaking row and its nearest neighbours exactly as the catalog states them.
Why it is hard
The site answer is the Skill column. The catalog rates the row Hard, and that is all "Hard" means here. This page imports no outside rubric and states no failure rate, tolerance or claim about how skilled any worker is. What the site does say is in the row text. The task is four named stages on one workpiece: forging, grinding, heat treating and finishing. The stages are listed as one run of work on a single blade, so the later stages depend on what the earlier ones left behind. The site publishes no step list, no limit and no outcome definition for any stage.
The site capture copy adds two things that bear on this work. The first is that the part that matters is often the part a hand is covering, which is why hand pose carries a per-joint visibility flag for hidden fingers. The second is that a stereo pair recovers depth from a calibrated frame pair, so the shape of the part sits in the same frame as the hands. The challenge table below maps each difficulty to the stream or label the site already states, and says Not stated where nothing is.
Capture
The catalog lists one capture mode for the row: Phone (head-mounted) / multiview rig. A head-mounted phone records the first-person view of real work, and a multiview rig is the alternative mode the row lists. A wrist-mounted camera is available alongside the head camera for close-range detail of the hands and the tool. The row does not state stereo capture, and this page does not describe the row as stereo capture. Stereo reaches this row only through the site general stereo offer: a capture choice agreed in the skill spec and confirmed on a Spec Pilot.
The site describes two stereo setups and does not say which one a blade making order would use. The off-the-shelf stereo head rig, dataset 06, is dual fisheye cameras at 1080p per eye with a 62 mm baseline, per-unit calibration, a 9-axis IMU and hand-tracking output, shipped as MCAP. The hard-environments stack is a global-shutter 4K/60 head camera, two 1080p/60 wrist cameras, a depth sensor at 848×480 and 30 fps and 200 Hz 6-DoF IMUs. Which of these applies to a metal workshop, how many cameras a spec calls for and how a rig is mounted in a shop are Not stated. Confirm them in the skill spec.
What stereo adds
The site states that a stereo pair at human-eye spacing makes depth recoverable from every frame pair, and that a calibrated pair measures disparity between two views at a fixed, known baseline rather than guessing depth from a single frame. For blade work, that puts the stock, the tool and the blade in the same capture as the hands, so the distance between a hand and a workpiece is carried by the data instead of inferred from one view. The site gives no depth accuracy figure for these tasks, so what a buyer does with that depth is their pipeline question.
The site also notes that glare-heavy and low-texture scenes make single-frame depth worse, and that depth degrades on wet and reflective surfaces while confidence maps mask the low-confidence regions. How depth behaves on bare or polished metal in a shop is not stated, so this page promises no figure and raises it as a point to settle in the spec before capture. Calibration is re-solved at the start and end of every session under a 0.28 px RMS reprojection gate, and streams are hardware-triggered onto one clock with a 2.0 ms cross-stream ceiling.
Episodes
Each attempt is delivered as one episode. The standard episode carries action segments, 3D hand pose for both hands with a per-joint visibility flag, contact events that mark when a hand meets an item, an objects list, and per-frame instance masks stored at the depth resolution of 848×480. Hand pose is 21 joints per hand, 3D, in metres in the head-camera frame. Each episode also carries one task instruction, which is the only language data the site states. An episode might cover one stage or a run of stages on one blade; where each episode starts and ends is written into the spec, because the site sets no episode length.
Action segments divide an episode into start- and end-timestamped spans, each carrying start_ns, end_ns, verb and noun, so forging, grinding, heat treating and finishing can each be their own labeled span if the spec asks for it. Which verbs and nouns are used is set in the skill spec. The site lists no label for the outcome of a stage, no label for the state of a blade, and no torque, grip-force or tactile stream in standard capture, so ask for any such label in the spec.
Neighbouring pages
The egocentric knifemaking and bladesmithing page covers the same catalog row from the first-person side: the row, the streams, the privacy rules and the delivery formats. This page is the stereo and container counterpart. It takes the same row and sets out what a calibrated pair and an MCAP container would add, and it keeps the capture description exactly as the catalog states it. The two pages describe one row, not two datasets, and the sibling is the better first read if you only need the row itself.
The stereo welding and metal fabrication page covers the other workshop rows on the stereo side, and states that the knife-forging row is covered on the knifemaking page and is not counted there. The stereo grinding and deburring page covers the two grinding rows, which are about welds, seams and cut edges, and the site does not say they are blade tasks. The stereo weld inspection and cleanup page covers one inspection row, and the stereo jewelry and metalsmithing page covers a studio bench with small pieces. The comparison table below sets each page against this one.
Force and touch
Standard Firsthand capture is cameras, depth and IMU. A separate glove tier records a tactile signal, and the tactile glove page describes it. The site does not state that the glove tier is available for the knifemaking row, and it states no pressure, force or grip value for any stage of this work. If grip pressure on a handle or a tool matters to your policy, say so in the spec so the glove tier can be scoped and quoted.
Do not assume standard capture carries a force signal: it carries video, depth and motion from the IMU, plus the hand pose and contact events derived on the same clock.
MCAP
MCAP is an open source container file format for multimodal log data, supporting multiple channels of timestamped pre-serialized data. It is serialization-agnostic, so messages can be Protobuf, DDS (CDR), ROS, JSON and more. Its design is row-oriented and append-only, it stores schemas alongside the data, and it offers an optional index plus optional LZ4 or Zstandard chunk compression and optional CRC checksums. Those facts come from mcap.dev and the foxglove/mcap repository, both linked below. Firsthand is not affiliated with Foxglove or the MCAP maintainers.
Firsthand offers MCAP delivery, and dataset 06 is stated as shipping as MCAP. For this row the site states a stereo pair, depth with confidence maps, hand pose, IMU, contact events, action segments and calibration elsewhere. It does not state which of those are channels in an MCAP file, how they are named, what encoding they use or at what rate. No topic name, message schema or rate is stated on this page, and none should be assumed. Name the streams that matter in the skill spec and confirm them on a Spec Pilot batch. The stream map below sets each stream beside what the site states about it.
Delivery and license
Delivery is in the format your stack already reads: RLDS, LeRobot, WebDataset, HDF5, zarr or Rerun .rrd, or raw streams with metadata, with readable converter source alongside, and MCAP can ship alone or beside any of them. The episodes follow the same episode schema as every other Firsthand collection, so a loader that reads another Firsthand dataset reads this one without changes. Which container suits a lab depends on its reader and its training loop, and the site does not rank them.
The license is perpetual, irrevocable and worldwide, buyer-owned and non-exclusive by default, with exclusivity available as a separate term. The executed agreement controls. Official documentation for MCAP, RLDS, LeRobot and WebDataset is linked in the sources below.
Validation
Validation runs against signed accept criteria written into the spec before capture, and a reject log ships with each batch. Episodes are machine-checked against stated reject criteria, such as a sync breach, the manipulated object out of frame for more than 15 percent of the episode, or more than 35 percent of depth pixels in the manipulation region invalid, and a human reviewer then checks each one. Only footage that clears the criteria is billed as a validated hour.
For this row, the spec has to say which item counts as the manipulated object at each stage and where the manipulation region is, because the stock, the tool and the blade are different items at different moments. What counts as a finished stage is also written down, since the site states no outcome definition. Rejected items are recaptured at our cost, and the site states no accuracy figure for stage labels, so what counts as a pass for a stage boundary is set in the accept criteria.
Capture and privacy
Every contributor signs a written release, and each delivered episode links to it through consent artefact IDs. At ingest, automated detection finds faces, plates and screens and blurs them irreversibly, and speech is scrubbed of names, addresses and card numbers. A human reviewer confirms that no un-consented person is identifiable in any frame, footage with bystanders who cannot be de-identified is discarded, and the anonymization record ships with the data.
The site does not state a separate rule for other text in a shop, such as drawings, labels, maker marks or machine displays, so this page does not claim they are blurred. It also states nothing about the ownership of a design or a finished piece, and Firsthand does not capture at a customer own facility, so the setting a program uses is agreed in the skill spec before capture begins. This page makes no safety, protective equipment, qualification or compliance claim, because the site makes none for this row, and it gives no instructions for any step. None of this is legal advice.
Speed and scale
From signed spec to first validated batch, the published timeline is a median of 48 hours where coverage is deep and three to four weeks where it is thin and contributors need to be recruited first. The row is Available on request, so the range that applies is confirmed when the spec is quoted rather than assumed. After a first batch, on-demand top-ups run against the same accept criteria, so you can add variation without rewriting the spec.
Blade making is custom collection at scale through a network of thousands of capture providers, for a consistent, recurring data flow to labs. The site states no provider count, hours or throughput for this row, and none of its off-the-shelf dataset descriptions names it, so treat it as a custom collection. Only dataset 06, the stereo head rig, is stated as shipping as MCAP, and a lab can use that stock footage to check the rig and the format before commissioning the task.
Getting started
Describe the stages your policy needs, using the four the row names: forging, grinding, heat treating and finishing. Say where each episode starts and ends, what you expect to see as the stock, the tool and the blade, what counts as a finished stage and the variation your pipeline is missing. Then name the streams you want, whether stereo is part of the capture, and whether MCAP is the only delivery or sits beside RLDS, LeRobot or WebDataset. That becomes a written skill spec, quoted for reach, timeline and price before any capture begins.
Most programs begin with a Spec Pilot, a small, fast first batch that proves the spec and the accept criteria on real footage before volume scales. Collections are billed per validated hour. Counts of sites, episodes and hours are set in the skill spec. Our skill spec guide walks through how to write one.
Capture setup
| Part | As stated on this site |
|---|---|
| Capture mode (Knifemaking / Bladesmithing) | Phone (head-mounted) / multiview rig, as the catalog lists it |
| Wrist camera | Available alongside the head camera for close-range detail |
| Stereo head rig (dataset 06) | Dual fisheye, 1080p per eye, 62 mm baseline, per-unit calibration, 9-axis IMU, hand-tracking output; shipped as MCAP |
| Hard-environments stack | 4K/60 global-shutter head camera, 120° lens; two 1080p/60 wrist cameras; depth 848×480 at 30 fps; 200 Hz 6-DoF IMU |
| Sync | Hardware-triggered onto one clock; 2.0 ms cross-stream ceiling |
| Calibration | Re-solved at the start and end of every session; 0.28 px RMS gate |
| Hand pose | 21 joints per hand, 3D, in metres in the head-camera frame; per-joint visibility flag |
| Stereo on this row | Not stated; the site general stereo offer, agreed in the skill spec and confirmed on a Spec Pilot |
| Which streams apply in a metal workshop | Not stated; confirmed in the skill spec |
Every collection runs through the same seven-stage end-to-end custom collection pipeline.
Live catalog rows
| Task | Description | Environment | Skill | Capture | Status |
|---|---|---|---|---|---|
| Knifemaking / Bladesmithing | Forging, grinding, heat treating and finishing blades in a metal workshop. | Metal Fabrication & Welding Workshop | Hard | Phone (head-mounted) / multiview rig | Available on request |
| Grinding & Deburring | Grind welds and cut edges smooth and remove burrs with grinder and hand files. | Metal Fabrication & Welding Workshop | Medium | Phone (head-mounted) / multiview rig | Ready to deploy |
| Welding | Not reproduced here; see the environment page for the full row text | Metal Fabrication & Welding Workshop | Hard | Phone (head-mounted) / multiview rig | Ready to deploy |
Neighbouring pages
| Page | Main topic | Where Knifemaking / Bladesmithing appears |
|---|---|---|
| This page | The Knifemaking / Bladesmithing row, with stereo and MCAP as spec-level choices | The whole page |
| Egocentric knifemaking and bladesmithing data | The same row, from the first-person side | The whole page; the sibling |
| Stereo welding and metal fabrication data | The other workshop rows on the stereo and MCAP side | Stated as covered on the knifemaking page, not counted there |
| Stereo grinding and deburring data | The two grinding rows, about welds, seams and cut edges | Not covered; separate rows |
| Stereo weld inspection and cleanup data | One weld inspection row | Not covered |
| Stereo jewelry and metalsmithing data | Studio bench rows with small metal pieces | Not covered; different environment |
| Egocentric stereo data in MCAP | The MCAP delivery format for stereo capture in general | Not covered; format page |
| Hard-task data collection | The hub of rows the catalog rates Hard | The row is in the Hard-row table and links here |
Capture challenges
| Workshop challenge | What goes wrong | Stated stream or label |
|---|---|---|
| Hands cover the stock, handle or tool | Fingers hide the part that matters | 3D hand pose with per-joint visibility flags; per-frame instance masks; a wrist view alongside the head camera; reject if the object is out of frame for more than 15% of an episode |
| Hand meets tool or workpiece | The hand and the item touch in quick succession | Contact events mark when a hand meets an item; no finer contact label is stated |
| Glare and reflective metal | Depth readings degrade on a reflective surface | The site states that depth degrades on wet and reflective surfaces and that a confidence map is delivered to mask low-confidence regions; reject if more than 35% of depth pixels in the manipulation region are invalid |
| Four stages on one blade | It is unclear where one stage ends and the next begins | Action segments, one per stage as written in the spec; no accuracy figure for stage labels is stated |
| State of the blade between stages | What a stage leaves behind is a visual state | Not stated: no blade-state, outcome or quality label is listed |
| Motion blur and occlusion | Fast reaches between stock, tool and body | Global-shutter head camera avoids rolling-shutter skew; visibility flag for hidden joints |
| Grip and handling force | How hard a handle or tool is held | Not stated: no force, torque or tactile stream in standard capture; the glove tier is a separate tier |
Stream map
| Stream or item | What the site states | In an MCAP delivery, as the site states it |
|---|---|---|
| Stereo pair | Dual fisheye, 1080p per eye, 62 mm baseline (dataset 06); for this row, agreed in the skill spec | Dataset 06 is stated as shipped as MCAP; channel layout not stated |
| Depth with confidence maps | Recovered from the calibrated stereo pair; confidence map masks low-confidence regions | Not stated |
| Hand pose | 21 joints per hand with a per-joint visibility flag | Channel layout and joint order not stated |
| Wrist views | Wrist camera available alongside the head camera | Not stated |
| IMU | 9-axis (dataset 06); 6-DoF at 200 Hz (hard-environments stack) | Channel layout and rate not stated |
| Contact events, action segments | Listed in the site episode schema; action segments carry start_ns, end_ns, verb and noun | Not stated |
| Calibration | Measured every session and shipped with the episode | Placement in the file not stated |
| Topic names, schemas, encoding, rates | Not stated | Agreed in the skill spec |
Delivered vs not stated
| Item | Status | Detail |
|---|---|---|
| Action segments | Delivered | start_ns, end_ns, verb, noun, one per stage |
| 3D hand pose | Delivered | 21 joints per hand; per-joint visibility flag for occluded fingers |
| Contact events | Delivered | The moment a hand meets an item |
| Per-frame instance masks | Delivered | Stored at 848×480 |
| Consent and anonymization records | Delivered | Consent artefact IDs per episode; anonymization record shipped with the data |
| Formats | Delivered | MCAP, RLDS, LeRobot, WebDataset, HDF5, zarr, Rerun .rrd or raw streams with metadata |
| Steel or material types, tool and machine models, process parameters | Not stated | Agreed in the skill spec |
| Stage-outcome, blade-state or quality labels | Not stated | No such label is listed |
| MCAP topics, schemas and rates | Not stated | Agreed in the skill spec |
| Cycle times and throughput | Not stated | No such figure is listed |
| Number of sites, episodes and hours | Not stated | Set per program in the skill spec |
| Safety, qualification or compliance claims | Not stated | This page makes none |
Example footage
One real clip already published on this site. It is a format and rig example: a head-mounted stereo capture with hands in frame. It is not knifemaking, bladesmithing or any task from this page.
No knifemaking or bladesmithing clip is published on the site, so this is the nearest stereo clip, labelled as a rig example. Request a sample to see delivered files. See stereo sample page, request an egocentric data sample.
FAQ
Sources
Tell us the stages you need, where each episode starts and ends, what you expect to see as the stock, the tool and the blade, whether stereo and MCAP are part of the delivery, and the angles you want. We will quote reach, timeline and price against a written spec.